Document [original]

Type 2 diabetes in urban Ghana - the role of

anthropometry and nutrition

vorgelegt von

Diplom-Ernährungswissenschaftlerin,

Master of Science in Epidemiology

Laura Frank

geb. in Berlin

von der Fakultät VII – Wirtschaft und Management der

Technischen Universität Berlin

zur Erlangung des akademischen Grades

Doktorin der Gesundheitswissenschaften / Public Health

- Dr. P. H. -

genehmigte Dissertation

Promotionsausschuss:

Vorsitzende: Prof. Dr. Jacqueline Müller-Nordhorn

Gutachter: Prof. Dr. Matthias Schulze

Gutachter: Prof. Dr. Reinhard Busse

Tag der wissenschaftlichen Aussprache: 25.04.2017

Berlin 2017

TABLE OF CONTENTS I

TABLE OF CONTENTS

TABLE OF CONTENTS ..................................................................................................... I

LIST OF TABLES .............................................................................................................III

LIST OF FIGURES ........................................................................................................... V

LIST OF ABBREVIATIONS .............................................................................................. VI

SUMMARY ..................................................................................................................... VII

ZUSAMMENFASSUNG .................................................................................................... X

1 INTRODUCTION ........................................................................................................... 1

1.1 Type 2 diabetes mellitus ..................................................................................... 1

1.1.1 Definition ........................................................................................................... 1

1.1.2 Epidemiology .................................................................................................... 1

1.2 Overweight and obesity ...................................................................................... 2

1.2.1 Definition ........................................................................................................... 2

1.2.2 Epidemiology .................................................................................................... 4

1.2.3 Overweight and obesity and risk of type 2 diabetes .......................................... 4

1.2.4 Cut-offs for obesity measures ........................................................................... 5

1.3 Nutrition ................................................................................................................... 9

1.3.1 Dietary behavior in SSA .................................................................................... 9

1.3.2 Methods to derive dietary patterns ...................................................................10

1.3.3 Dietary patterns in SSA ....................................................................................12

1.3.4 Dietary patterns and type 2 diabetes ................................................................15

1.4 Public health relevance ..........................................................................................19

1.5 Objectives and research questions .........................................................................20

2 STUDY POPULATION AND METHODS .......................................................................22

2.1 Kumasi Diabetes and Hypertension study ..............................................................22

2.1.1 Study setting ....................................................................................................22

2.1.2 Study design and study population ..................................................................22

2.1.3 Data collection .................................................................................................23

2.1.4 Analytical study population ...............................................................................28

2.2 Statistical analysis ..................................................................................................29

2.2.1 Anthropometry .................................................................................................30

2.2.2 Factor analysis .................................................................................................32

2.2.5 Reduced rank regression .................................................................................34

TABLE OF CONTENTS II

3 RESULTS .....................................................................................................................37

3.1 Characterization of study population .......................................................................37

3.2 Anthropometry ........................................................................................................39

3.2.1 Anthropometric characteristics .........................................................................39

3.2.2 Associations between anthropometric measures and type 2 diabetes ..............40

3.2.3 Discrimination of type 2 diabetes cases and controls .......................................45

3.2.4 Examination of cut-offs for obesity measures ...................................................47

3.3 Nutrition ..................................................................................................................49

3.3.1 Intake of energy, macronutrients and food groups ...........................................49

3.3.2 Dietary patterns derived by factor analysis .......................................................51

3.3.3 Associations between dietary patterns and type 2 diabetes .............................57

3.3.4 Dietary pattern derived by reduced rank regression .........................................60

3.3.5 Association between RRR-derived dietary pattern and type 2 diabetes ...........64

4 DISCUSSION ...............................................................................................................72

4.1 Discussion of results ..............................................................................................72

4.1.1 Anthropometric characteristics .........................................................................72

4.1.2 Associations between anthropometric measures and type 2 diabetes ..............72

4.1.3 Discrimination of type 2 diabetes cases and controls .......................................74

4.1.4 Examination of cut-offs for obesity measures ...................................................75

4.1.5 Intake of energy, macronutrients and food groups ...........................................76

4.1.6 Dietary patterns derived by factor analysis .......................................................76

4.1.7 Associations between dietary patterns and type 2 diabetes .............................78

4.1.8 Dietary pattern derived by reduced rank regression .........................................81

4.1.9 Association between RRR-derived dietary pattern and type 2 diabetes ...........81

4.2 Discussion of methods ...........................................................................................84

4.2.1 Study design and study population ..................................................................84

4.2.2 Data quality ......................................................................................................85

4.2.3 Statistical methods ...........................................................................................88

4.3 Public health relevance ..........................................................................................91

5 CONCLUSION AND FURTHER PERSPECTIVES ........................................................93

6 REFERENCES .............................................................................................................95

APPENDIX .................................................................................................................... 106

DANKSAGUNG ............................................................................................................. 122

EIDESSTATTLICHE ERKLÄRUNG ............................................................................... 123

LIST OF TABLES III

LIST OF TABLES

Table 1: Criteria for the diagnosis of type 2 diabetes by the American Diabetes

Association ....................................................................................................................... 1

Table 2: International classification of adult weight according to BMI-categories (WHO,

2006) ................................................................................................................................ 3

Table 3: Existing thresholds for abdominal obesity of various organizations ..................... 3

Table 4: Studies that investigated the association between anthropometric measures and

diabetes in SSA ................................................................................................................ 8

Table 5: Dietary patterns derived by exploratory methods among SSA populations .........13

Table 6: Dietary patterns derived by reduced rank regression and type 2 diabetes risk

among Caucasian populations .........................................................................................16

Table 7: Descriptive characteristics of 1221 urban Ghanaian participants of the KDH study

........................................................................................................................................38

Table 8: Anthropometric characteristics among women and men of the KDH study .........39

Table 9: Age-adjusted Spearman correlation coefficients for anthropometric measures

among women and men ..................................................................................................39

Table 10: Multivariate-adjusted ORs (95% CI) for type 2 diabetes by different

anthropometric measures among 922 women .................................................................41

Table 11: Multivariate-adjusted ORs (95% CI) for type 2 diabetes by different

anthropometric measures among 299 men ......................................................................42

Table 12: Multivariate-adjusted ORs (95% CI) for type 2 diabetes per 1SD by different

anthropometric measures among women and men .........................................................44

Table 13: Effect modification of the association between BMI and type 2 diabetes by

socioeconomic status .......................................................................................................45

Table 14: Sensitivity and specificity of diabetes cases identified using sex-specific cut-off-

points and Youden index for BMI, WC and WHR .............................................................47

Table 15: Calculation of sensitivity and specificity of diabetes cases identified using sex-

specific cut-off-points and Youden index for BMI, WC and WHR in participants with a good

glycaemic control (FPG < 7mmol/L) .................................................................................48

Table 16: Median energy and macronutrient intake and number of meals per day in the

study population ...............................................................................................................49

Table 17: Characteristics by quintiles of the “purchase” dietary pattern among 679 controls

of the KDH study ..............................................................................................................54

Table 18: Characteristics by quintiles of the “traditional” dietary pattern among 679

controls of the KDH study ................................................................................................55

LIST OF TABLES IV

Table 19: Multivariate-adjusted ORs (95% CI) for type 2 diabetes per quintiles and per 1

SD of dietary pattern scores .............................................................................................58

Table 20: Effect modification of the associations between dietary patterns and type 2

diabetes by age, sex, BMI, central obesity and SES sum score .......................................59

Table 21: Explained biomarker variation of the three response scores ............................61

Table 22: Weight of biomarkers in response scores derived by RRR ...............................61

Table 23 : Explained biomarker variation of the three dietary pattern scores ...................62

Table 24: Factor loadings of all 35 food items derived by reduced rank regression ..........63

Table 25: Characteristics, biomarker concentrations and food intake by quintiles of dietary

pattern score among 668 controls ....................................................................................65

Table 26: Multivariate-adjusted ORs (95% CI) for type 2 diabetes by quintiles and per 1SD

of the dietary pattern score ..............................................................................................67

Table 27: Selected food items identified by linear stepwise regression ............................69

Table 28: Characteristics, biomarker concentrations and food intake by quintiles of the

simplified dietary pattern score among 668 controls ........................................................70

Table 29: Multivariate-adjusted ORs (95% CI) for type 2 diabetes by quintiles and per 1SD

of the simplified dietary pattern score ...............................................................................71

Table 30: Importance of individual food components of the dietary pattern score ............71

Table S1: Input variables of the FFQ for factor analysis and reduced rank regression ... 111

Table S2: Rotated factor-loadings for the two identified dietary patterns among women

and men......................................................................................................................... 114

Table S3: Rotated factor-loadings for the two identified dietary patterns among the total

study population and controls ........................................................................................ 115

Table S4: Rotated factor-loadings for the two identified dietary patterns among the total

study population ............................................................................................................. 116

Table S5: Rotated Factor loadings of a three-factor solution among the total study

population ...................................................................................................................... 117

Table S6: Rotated Factor loadings of a four-factor solution among the total study

population ...................................................................................................................... 118

Table S7: Rotated Factor loadings of a five-factor solution among the total study

population ...................................................................................................................... 119

Table S8: Rotated Factor loadings of a two-factor solution with single fruit and vegetable

items among the total study population .......................................................................... 120

Table S9: Rotated Factor loadings of a two-factor solution with various fruit and vegetable

groups among the total study population........................................................................ 121

LIST OF FIGURES V

LIST OF FIGURES

Figure 1: Construction of the socioeconomic status (SES) sum score .............................26

Figure 2: Flow diagram of analytical study population ......................................................29

Figure 3: Example of a receiver operating characteristic (ROC) curve .............................31

Figure 4: Receiver operating characteristic (ROC) curves of various anthropometric

measures for discriminating type 2 diabetes cases and controls among women and men

........................................................................................................................................46

Figure 5: Dietary energy supply by macronutrients of the total study population ..............49

Figure 6: Food intake (servings/week) among controls and type 2 diabetes cases (FFQ) 50

Figure 7: Scree plot of eigenvalues ≥ 1.0 among the total study population .....................51

Figure 8: Spider graph of the two dietary patterns identified by factor analysis among the

total study population .......................................................................................................52

Figure S1: Food frequency questionnaire used in the KDH study .................................. 106

Figure S2: Example of the assessment of one 24 hour dietary recall used in the KDH

study .............................................................................................................................. 110

Figure S3: Scree plot of eigenvalues ≥ 1.0 among women and men of the KDH study .. 113

Figure S4: Scree plot of eigenvalues ≥ 1.0 among controls and total study population .. 113

LIST OF ABBREVIATIONS VI

LIST OF ABBREVIATIONS

ADA American Diabetes Association

AUC Area under the curve

24HDR 24 hour dietary recall

BMI Body mass index

CI Confidence interval

CIE Change in estimate

CRP C-reactive protein

CV Coefficient of variation

EPIC European Prospective Investigation into Cancer and Nutrition

FFQ Food frequency questionnaire

FPG Fasting plasma glucose

HDL High-density lipoprotein

HOMA-IR Homeostasis model assessment for insulin resistance

HR Hazard ratio

KATH Komfo Anokye Teaching Hospital

KDH Kumasi Diabetes and Hypertension

LDL Low-density lipoprotein

NCD Non-communicable disease

NHANES National Health and Nutrition Examination Survey

NHS Nurses’ Health Study

OR Odds ratio

ROC Receiver operating characteristic

RR Relative risk

RRR Reduced rank regression

SES Socioeconomic status

SSA Sub-Saharan Africa

WHO World Health Organization

WHR Waist-to-hip ratio

WHS Whitehall II Study

WHtR Waist-to-height ratio

SUMMARY VII

SUMMARY

Introduction and objectives: The prevalence of type 2 diabetes is rising worldwide with

a rapid increase in sub-Saharan Africa (SSA) [1, 2]. At the same time, prevalences of

overweight and obesity are increasing dramatically in this region, particularly in urban

areas [3, 4]. However, SSA is still dealing with infectious diseases such as malaria, HIV-

infections and tuberculosis [5]. This double burden poses a major public health challenge

in this region, where financial and health resources are limited. Although obesity and the

nutritional behavior are the main modifiable risk factors for type 2 diabetes [6], their

relationship is only insufficiently investigated in SSA. Therefore, the first objective of this

thesis was to evaluate the associations between various anthropometric measures and

type 2 diabetes and to assess the appropriateness of specific cut-off points for the body

mass index (BMI), waist circumference (WC) and waist-to-hip ratio (WHR) in an urban

Ghanaian study population. The second object of this thesis was to describe the dietary

behavior and to examine the associations between dietary patterns derived by an

exploratory factor analysis and type 2 diabetes. The third aim was to identify a dietary

pattern by using the reduced rank regression (RRR) approach and to evaluate the

association between this pattern and type 2 diabetes in this study population.

Data and Methods: Data from 1221 study participants (542 type 2 diabetes cases and

679 controls) of the Kumasi Diabetes and Hypertension (KDH) study was used. The KDH

study is an unmatched case-control study, which was conducted at the Komfo Anokye

Teaching Hospital (KATH) in Kumasi, Ghana between August 2007 and June 2008. All

participants underwent an anthropometrical examination and the habitual dietary intake

was assessed by one 24 hour dietary recall and a locally specific food frequency

questionnaire (FFQ). Each participant provided a blood sample and type 2 diabetes was

defined as having a fasting plasma glucose ≥ 7mmol/L and/or documented anti-diabetic

medication.

First, associations between various anthropometric measures and type 2 diabetes were

evaluated by multivariate-adjusted logistic regression analysis. Additionally, the

discriminative power and population-specific cut-off points for type 2 diabetes were

identified by receiver operating characteristic curves. Finally, the optimal cut-off points for

BMI, WC and WHR were assessed by the Youden-Index.

SUMMARY VIII

Second, the dietary behavior of the study population was assessed by an FFQ. Dietary

patterns were identified by using an exploratory factor analysis (including 33 food items)

and their associations with type 2 diabetes were evaluated by multivariate-adjusted

logistic regression analysis.

Third, a dietary pattern was identified by using RRR with adiponectin, HDL-cholesterol

and triglycerides as response variables and 35 food items as predictor variables and the

association between the dietary pattern score and type 2 diabetes was assessed applying

multivariate-adjusted logistic regression analysis.

Results: First, measures of central obesity, but not of general obesity, were positively

associated with type 2 diabetes in women and men. Specifically, BMI was not associated

with type 2 diabetes, while WHR showed the strongest association in both sexes,

independent of BMI. Furthermore, WHR showed the best discriminative ability for type 2

diabetes and a cut-off point of ≥ 0.88 in women and ≥ 0.90 in men were the optimal WHR

cut-off points in this SSA population. The recommended cut-off points for BMI and WC

had a poor predictive ability with a low sensitivity and specificity.

Second, the dietary behavior was characterized by a high intake of carbohydrate- and fat-

dense foods, such as plantain, banku, bread, rice, fish and palm oil. Two dietary patterns

were identified by factor analysis: The “purchase” dietary pattern was characterized by a

high consumption of sweets, rice, protein-rich foods (red meat, poultry, eggs and milk),

fruits and vegetables and low consumption of plantain. This pattern was inversely

associated with type 2 diabetes. The “traditional” dietary pattern was characterized by a

high intake of plantain, green leafy vegetables, fish, fermented maize products, and palm

oil and was associated with an increased odds of type 2 diabetes.

Third, a dietary pattern was derived by RRR, which was characterized by a high

consumption of plantain, garden egg and cassava and a low intake of juice, sweets,

vegetable oil, rice, hot chocolate, soft drinks, eggs and red meat. This pattern was

positively associated with serum triglyceride concentrations and negatively with HDL-

cholesterol, but not with adiponectin. The odds for type 2 diabetes increased significantly

with increasing pattern score.

SUMMARY IX

Conclusion: This study highlights the important role of central obesity for the risk of type

2 diabetes in an SSA population. Furthermore, the current recommended cut-off points for

obesity measures are inappropriate to assess diabetes risk in this urban Ghanaian

population. Further investigations are needed to evaluate the rationale of country- or

region-specific cut-off points for anthropometric indices to identify individuals with type 2

diabetes in SSA.

Findings of the second part show that two diverse dietary patterns are identified and

strongly associated with type 2 diabetes in urban Ghana. Therefore, further investigations

are warranted to clarify the determinants of adherences to dietary patterns and to verify

these patterns in other West-African populations.

Findings of the third part suggest that adherence to traditional food items and low

preference for purchased foods relate to increased serum triglycerides and decreased

HDL-cholesterol, both risk factors for type 2 diabetes, and as a consequence may

increase the risk for type 2 diabetes. Finally, the reproducibility of the association between

the RRR-derived dietary pattern and type 2 diabetes should be evaluated in independent

populations.

ZUSAMMENFASSUNG X

ZUSAMMENFASSUNG

Hintergrund und Zielstellung: Die Typ-2-Diabetesprävalenz nimmt weltweit stetig zu,

mit einem rapiden Anstieg besonders in sub-Sahara Afrika (SSA) [1, 2]. Gleichzeitig steigt

die Prävalenz an Übergewicht und Adipositas drastisch an, insbesondere in den

städtischen Gebieten dieser Region [3, 4]. Zudem wird das Gesundheitssystem in SSA

immer noch durch Infektionskrankheiten wie Malaria, HIV-Infektionen und Tuberkulose

belastet [5]. Diese Doppelbelastung stellt eine große gesundheitspolitische

Herausforderung für diese Region dar, in der finanzielle und gesundheitliche Ressourcen

begrenzt sind. Adipositas und das Ernährungsverhalten sind die wichtigsten

modifizierbaren Risikofaktoren für Typ-2-Diabetes [6], jedoch ist ihre Beziehung bislang

nur unzureichend in SSA untersucht. Daher war das erste Ziel dieser Arbeit, die

Zusammenhänge zwischen verschiedenen anthropometrischen Maßen und dem Typ-2-

Diabetes Risiko zu untersuchen, sowie die Übertragbarkeit der spezifischen Grenzwerte

für den Body Mass Index (BMI), den Taillenumfang und dem Taille-Hüft-Verhältnis (WHR)

in einer urbanen ghanaischen Studienpopulation zu beurteilen. Das zweite Ziel war das

Ernährungsverhalten zu charakterisieren und die Zusammenhänge zwischen

Ernährungsmustern, die mittels einer explorativen Faktorenanalyse identifiziert wurden,

und dem Typ-2-Diabetes Risiko zu untersuchen. Das dritte Ziel war es ein

Ernährungsmuster mittels reduzierter Rangregression (RRR) zu identifizieren und dessen

Zusammenhang mit dem Typ-2-Diabetes Risiko in dieser Studienpopulation zu

untersuchen.

Datengrundlage und Methoden: Daten von 1221 Studienteilnehmern (542 Typ-2-

Diabetes Fälle und 679 Kontrollen) der Kumasi Diabetes und Hypertonie (KDH) Studie

wurden analysiert. Die KDH-Studie ist eine ungepaarte Fall-Kontroll-Studie, die am Komfo

Anokye Teaching Hospital (KATH) in Kumasi, Ghana von August 2007 bis Juni 2008

durchgeführt wurde. Die anthropometrische Untersuchung wurde von geschultem

Krankenhauspersonal am leicht bekleideten Studienteilnehmer durchgeführt. Das

Ernährungsverhalten wurde mithilfe eines lokal spezifischen Verzehrshäufigkeits-

fragebogens (FFQ) und eines 24 Stunden Ernährungsprotokolls ermittelt. Jedem

Teilnehmer wurde eine Blutprobe entnommen. Typ-2-Diabetes wurde definiert als

Nüchternblutzucker ≥ 7mmol/L und/oder bekannte Antidiabetika-Behandlung.

Im ersten Teil dieser Dissertation wurden die Zusammenhänge zwischen verschiedenen

anthropometrischen Maßen und dem Typ-2-Diabetes Risiko mittels multivariat-adjustierter

ZUSAMMENFASSUNG XI

logistischer Regressionsanalyse untersucht. Die Fläche unter der Receiver Operating

Characteristic Kurve (ROC-AUC) wurde zum Vergleich der diskriminativen Fähigkeit der

anthropometrischen Maße zur Identifizierung von Diabetikern genutzt, sowie zur

Überprüfung von populationsspezifischen Grenzwerten. Schließlich wurden die optimalen

Grenzwerte mithilfe des Youden-Index ermittelt.

Im zweiten Teil dieser Dissertation wurde das Ernährungsverhalten der Studienpopulation

mithilfe eines FFQs charakterisiert. Des Weiteren wurden Ernährungsmuster mittels einer

explorativen Faktorenanalyse, basierend auf 33 Lebensmitteln des FFQs, identifiziert und

die Assoziationen zwischen den Ernährungsmustern und dem Typ-2-Diabetes Risiko

mittels multivariat-adjustierter logistischer Regressionsanalyse untersucht.

Im dritten Teil dieser Dissertation wurde mittels RRR mit den Aufnahmemengen von 35

Lebensmitteln als Prädiktoren und den Serumkonzentrationen von Adiponektin, HDL-

Cholesterin und Triglyzeriden als Response-Variablen ein Ernährungsmuster hergeleitet

und die Assoziation mit dem Typ-2-Diabetes Risiko mithilfe der multivariat-adjustierten

logistischen Regressionsanalyse untersucht.

Ergebnisse: Die ersten Ergebnisse zeigten, dass Maße der zentralen Adipositas, aber

nicht der generellen Adipositas, positiv mit dem Typ-2-Diabetes Risiko sowohl bei Frauen

als auch bei Männern assoziiert waren. WHR war der stärkste Risikofaktor in beiden

Geschlechtern, unabhängig vom BMI. Der Vergleich der ROC-AUCs zeigte, dass die

WHR das beste anthropometrische Maß zur Identifizierung von Diabetikern sowohl in

Männern als auch in Frauen war. Der optimale WHR-Grenzwert für Frauen lag bei ≥ 0.88

und bei Männern ≥ 0.90 in dieser afrikanischen Bevölkerung. Die empfohlenen

Grenzwerte für BMI und Taillenumfang hingegen hatten eine schlechte Vorhersagekraft

mit einer niedrigen Sensitivität und Spezifität.

Das Ernährungsverhalten war durch eine hohe Aufnahme von kohlenhydrat- und

fettreichen Lebensmitteln (Kochbanane, fermentierte Maisprodukte, Brot, Reis, Fisch und

Palmöl) charakterisiert. Zwei Ernährungsmuster wurden mittels Faktorenanalyse

identifiziert: Das "industriell geprägte" Ernährungsmuster war charakterisiert durch einen

hohen Verzehr von Süßigkeiten, Reis, eiweißreichen Lebensmitteln (rotes Fleisch,

Geflügel, Eier und Milch), Obst und Gemüse und niedrigen Verzehr von Kochbananen.

Dieses Muster war invers mit dem Typ-2-Diabetes Risiko assoziiert. Das "traditionelle"

Ernährungsmuster hingegen war gekennzeichnet durch einen hohen Verzehr von

Kochbananen, grünem Blattgemüse, Fisch, fermentierten Maisprodukten und Palmöl. Es

war mit einem erhöhten Risiko für Typ-2-Diabetes assoziiert.

ZUSAMMENFASSUNG XII

Mithilfe der RRR wurde ein Ernährungsmuster hergeleitet, das durch einen hohen Verzehr

von Kochbananen, Aubergine und Maniok sowie einen geringen Verzehr an Saft,

Süßigkeiten, Pflanzenöl, Reis, Softgetränken, Eier und rotem Fleisch charakterisiert war.

Dieses Muster war mit erhöhten Serumkonzentrationen an Triglyzeriden und erniedrigten

HDL-Cholesterin assoziiert, aber zeigte keinen Zusammenhang mit Adiponektin. Ein

hoher Musterscore war mit einem erhöhten Risiko für Typ-2-Diabetes assoziiert.

Schlussfolgerung:

Diese Studie unterstreicht die wichtige Rolle der zentralen Adipositas für das Risiko von

Typ-2-Diabetes in einer afrikanischen Bevölkerung. Darüber hinaus sind die derzeit

empfohlenen Grenzwerte für Übergewicht und Adipositas ungeeignet, um das Diabetes-

Risiko in dieser ghanaischen Population zu beurteilen. Weitere Untersuchungen sind

notwendig, um Länder- oder Regions-spezifische Grenzwerte für anthropometrische

Maße zu untersuchen, um Personen mit Typ-2-Diabetes in SSA zu identifizieren.

Die Ergebnisse des zweiten Teils zeigen, dass zwei unterschiedliche Ernährungsmuster

identifiziert wurden, die stark mit dem Typ-2-Diabetes Risiko assoziiert sind. Die

Determinanten für die Einhaltung, sowie die Verifizierung dieser Muster sollten in anderen

Westafrikanischen Populationen weiter untersucht werden.

Die Ergebnisse des dritten Teils weisen darauf hin, dass der hohe Verzehr an

„traditionellen“ Lebensmitteln sowie der geringe Verzehr an „industriell geprägten“

Lebensmitteln Serumkonzentration an Triglyzeriden erhöhen und HDL-Cholesterin senken

könnten, beides Risikofaktoren für Typ-2-Diabetes, und dadurch zu einem erhöhten Risiko

für Typ-2-Diabetes führen können. Die Reproduzierbarkeit des Zusammenhanges

zwischen diesem Ernährungsmuster und dem Typ-2-Diabetes Risiko sollte in

unabhängigen Studienpopulationen überprüft werden.

INTRODUCTION 1

1 INTRODUCTION

1.1 Type 2 diabetes mellitus

1.1.1 Definition

Diabetes mellitus is a group of metabolic disorders characterized by the presence of

hyperglycemia resulting from defects in insulin secretion, insulin action, or both [7]. Type 2

diabetes is the most common form of diabetes, accounting for 90-95% of all cases, the

rest are type 1 or gestational diabetes cases. Type 2 diabetes is a multifactorial disease

that is caused by an interaction of genetic and environmental factors. The underlying

pathophysiological mechanism of type 2 diabetes is the combination of insulin resistance

and a progressive pancreatic ß-cell failure [8].

The criteria for the diagnosis of type 2 diabetes are shown in Table 1. The use of glycated

hemoglobin (HbA1c) with a cut point of ≥ 6.5% as an additional diagnostic test to

diagnose diabetes has been recommended by the International Expert Committee in 2009

[9] and endorsed by the American Diabetes Association [10] and the World Health

Organization (WHO) [11]. However, the practicability of using HbA1c for the diagnosis of

type 2 diabetes in SSA is questionable, because of high costs and high prevalences of

hemoglobinopathies such as sickle cell anemia [12].

Table 1: Criteria for the diagnosis of type 2 diabetes by the American Diabetes Association[7]

HbA1c ≥ 6.5%

Fasting plasma glucose (FPG) ≥ 126 mg/dl (7 mmol/L). Fasting is defined as no caloric intake for at

least 8 hours

Two-hour plasma glucose ≥ 200 mg/dl (11.1 mmol/L) during an oral glucose tolerance test (OGTT)

In a patient with classic symptoms of hyperglycemia or hyperglycemic crisis, a random plasma

glucose ≥ 200 mg/dl (11.1 mmol/L)

* in the absence of unequivocal hyperglycemia, criteria 1-3 should be confirmed by repeat testing

1.1.2 Epidemiology

The prevalence of type 2 diabetes is increasing dramatically in SSA, which is mainly

attributed to growing rates of obesity, urbanization, and physical inactivity [1, 3]. The sixth

edition of the International Diabetes Federation diabetes atlas estimated that the number

of adults with diabetes in SSA will double from 21.5 (2014) to 41.5 million people (2035)

within the next 20 years [2]. However, these projections did not consider the increasing

INTRODUCTION 2

role of potential risk factors such as urbanization, changes in lifestyle, and increased

physical inactivity. SSA has the highest percentage (62%) of undiagnosed type 2 diabetes

patients worldwide. In this region, 90% are affected by type 2 diabetes and the prevalence

of type 2 diabetes vary between countries and rural-urban gradients, from 1.0% in rural

Uganda and 12.0% in urban Kenya [13]. In Accra, the capital and largest city of Ghana,

the prevalence of diabetes was 0.4% in 1958 [14]. However, over 40 years later, Amoah

et al. reported a diabetes prevalence of 6.3% in the Greater Accra area of Ghana [15]. In

Africa, 8.6% of all death in adults aged between 20-79 years are due to diabetes and of

this proportion, 76% occurred in people younger than 60 years of age [2]. In 2014, overall

612 billion US dollars were spent for diabetes, with the lowest diabetes-related

expenditure (4.5 billion US dollars) for Africa [2].

1.2 Overweight and obesity

1.2.1 Definition

Obesity results from an imbalance in energy intake and energy expenditure leading to an

abnormal or excessive fat accumulation that may impair health [16]. Different

anthropometric measures are commonly used to classify overweight and obesity in adults.

The body mass index (BMI), a measure of general obesity, is defined as a person’s weight

in kilograms divided by the square of his height in meters (kg/m²). In contrast to BMI, waist

circumference (WC), waist-to-hip ratio (WHR), and waist-to-height ratio (WHtR) are

anthropometric measures that are usually used to assess abdominal obesity. Abdominal

obesity reflects an increased amount of intra-abdominal fat including visceral adipose

tissue that is associated with numerous cardiovascular disease risk factors (decreased

glucose tolerance, reduced insulin sensitivity, and adverse lipid profiles, all risk factors for

type 2 diabetes [17, 18]).

WHR and WHtR are defined as WC/hip circumference and WC/height, respectively. The

established cut-offs for BMI according to the WHO [19] are shown in Table 2. These cut-

offs are derived from mortality and morbidity data from predominately US and European

populations and are associated with an increased risk for type 2 diabetes [20] and

cardiovascular disease [21]. Table 3 shows the current international recommendations for

thresholds of abdominal obesity (WC and WHR) in diverse ethnic populations by different

organizations. A WC cut-off of 80 and 88 cm in women and 94 and 102 cm in men

indicates high and very high risk for metabolic complications, respectively. These cut-offs

correspond to BMI levels at 25.0-29.9 kg/m² and ≥ 30.0 kg/m², respectively, and were

INTRODUCTION 3

developed based on the association between WC and BMI [22], rather than on the

association between WC and disease risk.

Table 2: International classification of adult weight according to BMI-categories (WHO,

2006)[19]

Classification

BMI (kg/m²)

Underweight

< 18.5

Normal weight

≥ 18.5-24.9

Overweight

≥ 25.0-29.9

Obesity

≥ 30.0

BMI: body mass index

Table 3: Existing thresholds for abdominal obesity of various organizations

WC (cm)

WHR

Population

Organization

Women

Men

Women

Men

Europid

IDF [23]

≥ 80

≥ 94

Caucasian

WHO * [24]

≥ 80

≥ 94

≥ 0.85

≥ 0.90

≥ 88

≥ 102

Unites States

AHA/NHLBI (ATP) [25]

≥ 88

≥ 102

US Dep. of Agriculture and US Dep. of

Health and Human Services [26]

≥ 0.88

≥ 0.95

Canada

Health Canada [27]

≥ 88

≥ 102

European

European Cardiovascular Societies [28]

≥ 88

≥ 102

South Asian**,

Japanese and

Chinese

IDF [23]

≥ 80

≥ 90

Asian

WHO [24]

≥ 80

≥ 90

Japanese

Japanese Obesity Society [29]

≥ 85

≥ 90

China

Cooperative Task Force [30]

≥ 85

≥ 80

Middle-East,

Mediterranean

IDF [23]

≥ 80

≥ 94

SSA

IDF [23]

≥ 80

≥ 94

Ethnic South and

Central American

IDF [23]

≥ 80

≥ 90

WC: waist circumference; WHR: waist-to-hip-ratio; * at thresholds of ≥ 80 cm in women and ≥ 94 cm in men:

increased risk, thresholds of ≥ 88 cm in women and ≥ 102 cm in men: substantially increased risk for

metabolic complications; ** based on a Chinese, Malay and Indian-Asian population

INTRODUCTION 4

1.2.2 Epidemiology

Overweight and obesity, which were once problems of developed countries, have

nowadays become one of the main public health topics, particularly in urban areas of

many SSA countries due to dramatically increasing prevalences [4]. Obesity is the major

risk factor for non-communicable diseases (NCD) such as type 2 diabetes. Especially in

urban African populations rates of overweight and obesity are rising [4]. This is mainly

attributed to the ongoing nutrition transition by higher consumption of refined

carbohydrates and fatty and energy-dense foods and physical inactivity. Through a rapid

urbanization during the past decades with increasing motorized transportation

possibilities, the traditionally higher work- (e.g. farming) and transportation-related

physical activity has decreased in SSA [1]. Furthermore, leisure time physical activity is

not very popular, especially among women [31, 32].

In 2013, the age-standardized prevalences of overweight and obesity were 27.9% and

8.1% among adult (≥ 20 years) men and 38.4% and 14% among adult women in Ghana,

respectively [33]. Thus, obesity is more prevalent among Ghanaian women than among

men and especially central obesity is common among women [34]. While in developed

countries obesity is associated with a low socioeconomic status (SES) [35], African

studies showed an opposing positive association between SES and obesity [36-38].

Ziraba et al. reported that the prevalence of overweight and obesity increased by 35%

between 1992 and 2005 in seven African urban cities (including Ghana), with higher

prevalence among women of higher SES compared to their poorer counterparts [4].

However, the highest increase in obesity was observed among the poorest (+50%) and

not among the richest (+7%). Furthermore, obesity prevalences increased by 45-50%

among the non-educated and primary-educated women, compared to a drop of 10%

among women with secondary or higher education [4]. These findings suggest that

obesity has also become a problem of the poor and non-educated urban residents as

seen in their wealthier counterparts before.

1.2.3 Overweight and obesity and risk of type 2 diabetes

There is evidence that general and abdominal obesity is associated with an increased risk

of type 2 diabetes across Asian, European and US populations [20]. In this meta-analysis

of 32 cohort studies BMI, WC, and WHR were similarly associated with incident type 2

diabetes (pooled relative risk per 1 standard deviation for incident diabetes with 95%

confidence interval (CI) were 1.87 [1.67-2.10], 1.87 [1.58-2.20] and 1.88 [1.61-2.19],

INTRODUCTION 5

respectively). However, there is uncertainty whether BMI or central obesity measures are

better discriminators of diabetes risk across diverse ethnic groups. In fact, Vazques et al.

found modest regional differences for WHR (but not for WC or BMI): The association was

stronger in Caucasian compared to Asian populations (Europe: 1.9 [1.7-2.2]), United

States: 1.7 [1.4-2.2] vs. Asia: 1.4 [1.1-1.7] [20]. This is in line with numerous studies

among Asians and African Americans that observed central obesity measures including

WC, WHR and WHtR to be better discriminators of type 2 diabetes than BMI [39-43]. The

Obesity and Asia Collaboration, comprising 21 cross-sectional studies in the Asia-Pacific

region with >263,000 individuals, observed that measures of central obesity, in particular,

WC, were more strongly associated with diabetes and were better discriminators of

diabetes compared to BMI in Asians and Caucasians (with the exception of Caucasian

men) [39].

In contrast, the evidence for these associations among SSA populations is limited. Only a

few cross-sectional studies from SSA investigated the associations between obesity

measures and type 2 diabetes and are summarized in Table 4. These cross-sectional

studies found positive associations for general obesity (BMI) and central obesity

measures (WC, WHR and WHtR) with type 2 diabetes. However, most of the studies

showed that central obesity measures were stronger associated with diabetes compared

to BMI [44-46]. A recent study from Cameroon assessed and compared the strength and

discriminatory power of various anthropometric measures with diabetes [46]. In this study,

central obesity measures were better predictors for diabetes than BMI. WC showed the

best discriminatory ability to identify screen-detected diabetes in this population.

1.2.4 Cut-offs for obesity measures

Although there is evidence of a continuous association between BMI and WC with type 2

diabetes [47-49], cut-off points has been determined and generally accepted for defining

general or central obesity for population screening [50]. Cut-off points for obesity

measures such as BMI and WC have been derived mainly from studies among

Caucasians [22, 51] and their appropriateness for other populations is therefore

questionable [52-54]. A review of 28 studies (four prospective studies and 24 cross-

sectional studies) aimed to identify the optimal cut-off points for WC and WHR for

assessing risk of type 2 diabetes: The optimal cut-off points, derived at the point that

maximizes the sum of sensitivity and specificity, varied across ethnicities and thus Qiao

and Nyamdorj concluded that there is no universal cut-off point that can be applied

worldwide [50]. They also suggest that country- or region-specific cut-off values should be

INTRODUCTION 6

used taking into consideration the purpose for which the cut-off is required and the

availability of resources. In fact, Asian populations [55-57] and black populations [58, 59]

might bear higher risk of diabetes at lower BMI or WC levels than Caucasians. For Asian

populations thresholds of a BMI of 23 to < 27.5 kg/m² were identified, representing an

increased risk and ≥ 27.5 kg/m² representing a high risk for type 2 diabetes or

cardiovascular diseases [57]. Additionally, lower WC and WHR cut-off points for Asians

were observed: for WC 85 cm in men and 80 cm in women and for WHR 0.90 and 0.80,

respectively [39, 60]. It has been suggested that this may be due to ethnic differences in

the association of BMI, body fat distribution and cardiometabolic risk factors. For a given

BMI, Asians have higher levels of visceral adipose tissue and lower skeletal muscle mass

compared to Caucasians [61-63].

Also several aspects of body composition are known to differ between Caucasians and

Africans [64, 65]. Abdominal visceral adipose tissue, measured by computer tomography,

was significantly higher among 1396 white men and women compared to 571 African

American men and women. In contrast, subcutaneous adipose tissue was lower among

whites compared to African Americans [64]. Wagner et al. reviewed the similarities and

differences in the body composition between Blacks and Whites: In general, Blacks had

increased skeletal muscle mass and bone mineral content compared to Whites [65].

These differences in body composition in diverse ethnic groups may have an impact on

the determination of cut-offs for obesity measures. Indeed, a study from the Third National

Health and Nutrition Examination Survey (NHANES) investigated whether black, hispanic

and white Americans have the same values of abdominal obesity (estimated by WC) at

the established BMI cut-off points for overweight (BMI 25-29.9 kg/m²) and obesity (BMI

≥ 30 kg/m²) [66]. WC cut-off points were lower among Blacks and Hispanics compared to

Whites at the corresponding cut-offs for overweight and obesity. For example, the WC cut-

off points that correspond to overweight were in young black men 85 cm, in middle-aged

88 cm and in the elderly 92 cm, while the respective values for white men were 89 cm,

92 cm and 106 cm. Okosun et al. further investigated the predictive ability of the

recommended WC cut-offs for abdominal obesity (≥ 88 cm in women and ≥ 102 cm in

men) to identify type 2 diabetes among overweight Americans in NHANES: Among black

men and women, aged 40-59 years, these cut-offs showed a low sensitivity (23.1%, 95

CI%: [11.4-46.0] and 11.1% [3.4–18.9] with high specificity (84.7% [77.6–91.8] and 96.4%

[89.3–99.4] to detect type 2 diabetes [67].

With respect to SSA, no previous study investigated BMI, WHR and WC cut-off points to

identify type 2 diabetes. A cross-sectional study investigated WC cut-off points to identify

hypertension and recommended 75.6 and 80.5 cm for men and 71.5 and 81.5 cm for

INTRODUCTION 7

women of Nigerian and Cameroon origin to predict hypertension, respectively [68]. The

optimal waist circumference cut-off point to identify the presence of at least two

components of the metabolic syndrome was 86 cm for men and 92 cm for women in rural

South Africans [69]. Another study from urban South Africa found that the waist

circumference cut-off point for the identification of the metabolic syndrome among women

was higher (91cm) compared to the cut-off recommended by the IDF (80cm) [70]. Thus,

the evidence for specific cut-off points among SSA populations is insufficient and further

research is required.

INTRODUCTION 8

Table 4: Studies that investigated the association between anthropometric measures and diabetes in SSA

First author

(year)

Study population

Study

design

Anthropometric measure

Main results

Fisch et al.

(1987) [71]

7,472 (aged ≥ 15

years), rural Mali

cross-

sectional

study

BMI

BMI was positively associated with diabetes: OR with 95% CI for BMI ≥ 29:

3.89 [2.09-7.25]

Aspray et al.

(2000) [72]

770 adults (aged ≥ 15

years), rural and urban

Tanzania

cross-

sectional

study

BMI

WHR

BMI and WHR were positively associated with diabetes (OR with 95% CI)

rural men: 1.83 [1.18-2.83], urban men: 1.65 [1.06-2.55],

rural women: 12.16 [2.12-69.72], urban women: 1.22 [0.98-1.51]

rural men:1.18 [0.86-1.62], urban men: 1.31[1.04-1.65]

rural women: 1.42 [0.91-2.23], urban women: 1.18 [0.97-1.43]

Balde et al.

(2007) [44]

1,537 participants (aged

≥ 35 years), rural and

urban Guinea

cross-

sectional

study

BMI ≥ 30kg/m²

WC ≥ 80cm (women) or ≥94cm (men)

WHR ≥ 0.85 (women) or ≥0.95 (men)

BMI, WC and WHR were positively associated with diabetes in univariate

analysis, but only WC remained independently associated in multivariate

analysis

OR with 95% CI: 1.82 [0.85-3.91]

OR with 95% CI: 1.96 [1.26-3.07]

OR with 95% CI: 1.63 [1.01-2.63]

Giday et al.

(2010) [73]

395 participants, urban

and rural Southern

Ethiopia

cross-

sectional

study

BMI

WHR

BMI and WHR were positively associated with diabetes

OR with 95% CI: 3.96 [1.76-8.92]

OR with 95% CI: 2.83 [1.11-7.26]

Motala et

al.(2008) [45]

1,025 participants, rural

South Africa

cross-

sectional

study

WC was positively and HC inversely associated with diabetes; BMI was not

associated in multivariate analysis

OR with 95% CI: 1.10 [1.04-1.16]

OR with 95% CI: 0.92 [0.87-0.97]

Mbanya et al.

(2015) [46]

8,663 participants,

Cameroon

cross-

sectional

study

BMI

WC, WHtR and HC were positively, BMI and WHR were not associated with

diabetes; WC was best predictor for diabetes in ROC-curve analysis

OR with 95% CI for 1SD: 1.05 [0.98-1.13]

OR with 95% CI for 1SD: 1.30 [1.16-1.46]

WHR

OR with 95% CI for 1SD: 1.05 [1.00-1.16]

WHtR

OR with 95% CI for 1SD: 1.26 [1.11-1.39]

OR with 95% CI for 1SD: 1.18 [1.05-1.34]

BMI: body mass index, CI: confidence interval, HC: hip circumference, OR: odds ratio, ROC: receiver operating characteristic analysis, SSA: sub-Saharan Africa, WC: waist

circumference, WHR: waist-to-hip ratio, WHtR: waist-to-height ratio

INTRODUCTION 9

1.3 Nutrition

1.3.1 Dietary behavior in SSA

Nutrition is one of the main public health concerns in SSA. The prevalence of

undernutrition is one of the highest in the world, particularly in children under the age of

five years [74]. At the same time, this region is also experiencing a rapid increase in

obesity and diet-related chronic diseases [3]. This double burden of malnutrition causes a

major public health challenge in SSA.

Due to increasing rates of urbanization, SSA countries are experiencing a nutrition

transition with a shift from a traditional diet to a western diet. Most of the African countries

are at an early stage of this nutritional change. However, Ghana is one of the countries

that has reached the latter stages [75, 76]. This transition seems to increase nutrition-

related NCD’s by changes in behaviors, diets (high in fat, refined carbohydrates, sugar,

cholesterol and low in fiber), physical inactivity, smoking, alcohol consumption and

increasing obesity prevalences [75, 77]. In fact, studies from urban SSA showed that

buying street and processed foods and eating outside the home is frequent among urban

citizens [78].

Only a few small cross-sectional studies and surveys give information about the dietary

behavior in Ghana. A cross-sectional study among 400 women from rural Ghana reported

that typically three meals per day (79%) were consumed [79]. Two-thirds of the women

cooked two meals per day at home; at least the dinner is prepared at home. Breakfast and

lunch are usually bought at street vendors. Main dishes are made from cereals (maize,

rice and millet), starchy roots (cassava, yam and cocoyam) and plantain. These foods

provide the highest amount of the daily energy intake and diversity of the diet remains low

[80]. They are usually served with soups or stews including fried or grilled fish, meat and

vegetables. Large amounts of spices and palm oil are added. The main source of animal

protein is fish [79], however the consumption of meat [79] and milk [79, 81] is very low.

Vegetables (e.g. pepper, onion, garden eggs and tomato) are consumed on a daily basis

mainly in soups or stews. The consumption of fruits is rather low and is influenced by a

number of factors. Ghanaians with higher income or better education consume more fruits

compared to those with a lower income or education [76]. Furthermore, the intake of fruits

is higher among women than men. Fruits were also more consumed in rural areas,

because of better accessibility and variety depending on the seasonality. In contrast, in

urban areas the consumption of processed foods and snacks are more common [78]. With

regard to the food balance sheets of the Food and Agricultural Organization, the per

INTRODUCTION 10

capita daily energy supply in Ghana has increased from 1,729 kcal in 1990 to 3,003 kcal

in 2011 [82]. About 70% of the daily energy supply is derived by carbohydrates, 20% by

fat and the remaining 10% by protein [76]. Salted food such as fish and meat are often

consumed and the addition of salt at the table is common in Ghana [83, 84]. With respect

to alcohol consumption, 58-63% remain abstinent lifelong [85, 86], of the current drinkers

(20.2%) 7.3% are heavy drinkers with having ≥ 5 units on one or more occasions during

the week [85]. The Study on global AGEing and adult health (SAGE) reported, that the

alcohol consumption in the elderly Ghanaian population (≥ 50 years) of frequent heavy

drinkers is very low (1.1%) [87].

1.3.2 Methods to derive dietary patterns

In the past, nutritional epidemiology has traditionally focused on the association between

single nutrients or foods and a disease outcome. Nevertheless, this approach does not

take into account the complexity of the human diet with their synergistic and interactive

effects. Furthermore, the high degree of inter-correlation among nutrients and among

foods makes it difficult to examine their separate effects. Therefore, dietary pattern

analysis has led to a growing interest in nutritional research, because dietary patterns

reflect different combinations of food intake and allow the assessment of the overall diet

[88].

Generally, dietary patterns are constructed by either an exploratory or a hypothesis-

oriented approach [88-90]. The hypothesis-oriented approach, also called “a priori”

approach takes into account nutrition recommendations that are evidence-based for

nutrition-disease relationships to construct dietary scores or indices. Examples of popular

and well-established dietary pattern scores from Caucasian populations are the Healthy

Eating Index, based on the US Department of Agriculture Food Guide Pyramid [91] and

the Mediterranean Diet score [92]. Both scores rank participants to the degree they

conform to specific dietary recommendations. For example, the Healthy Eating Index is a

single, summary measure of the degree to which an individual’s diet conforms to the

recommendations of the US Department of Agriculture Food Guide Pyramid and to

specific recommendations in the US Dietary Guidelines for Americans [91]. This index

measured the adherence to serving recommendations of ten equally weighted

components (five food groups [grains, vegetables, fruits, milk and meat], four nutrients

[intake of total fat, saturated fat, cholesterol and Sodium] and diet diversity). Each of the

10 components has a score, ranging from 0 to 10, with a possible total index score of 100

[91].

INTRODUCTION 11

The exploratory, also called “a posteriori” techniques are entirely data-driven methods,

such as principal component analysis (PCA), factor analysis or cluster analysis. These

exploratory analyses are all dimension reduction techniques, hypothesis-free, and are

constructed independent of their relevance to the outcome of interest. PCA and factor

analysis identifies food items that are frequently consumed together among a study

population. Both techniques reduce the original set of correlated variables into a smaller

set of uncorrelated variables called principal components or factors [89]. This aggregation

of food items or food groups is based on the degree to which they are correlated with

each other (based on the covariance structure of food variables). In PCA the principle

components are linear combinations of optimally weighted observed variables (food items

or food groups) that account for the largest amount of variation in diet between individuals.

In contrast, factor analysis assumes that the observed variables are linear combinations of

unobservable (latent) factors [93]. Besides the theoretical differences between both

techniques, the results of the factor analysis based on the principal factor method (usually

used in dietary pattern analyses) are generally similar to PCA [90].

In contrast, cluster analysis is used to group individuals into relatively homogenous

subgroups (clusters) based on their similarities in food consumption data. Two methods

are usually used to derive clusters: The hierarchical Ward’s method that minimize the

variance within clusters or the non-hierarchical K-Means method that maximize the

distance between clusters [94]. Further analyses are necessary to identify particular

dietary characteristics and to interpret the identified clusters [88, 93].

The reduced rank regression (RRR) has been proposed as a new dimension-reduction

technique which combines the advantages of the hypothesis-oriented and the exploratory

approach [95]. The RRR considered the etiological scientific knowledge (e.g. about

biomarkers or nutrients that are linked to a disease) and the nutritional information of a

specific study population and is especially useful to identify health-related dietary patterns.

This method determines linear combinations of predictor variables (e.g. food groups) by

maximizing the explained variation in a set of response variables (e.g. nutrients or

biomarkers) that are presumed to be related to the disease of interest [95].

INTRODUCTION 12

1.3.3 Dietary patterns in SSA

With respect to SSA, only five cross-sectional studies in comparatively small and specific

populations have attempted to identify dietary patterns by exploratory methods and are

summarized in Table 5. The five studies from West- [78, 96], East- [97], Central- [98], and

Southern-Africa [99] identified between two and five dietary patterns, which vary due to

differences in the specific foods. Two of the studies obtained dietary patterns that were

characterized by westernized habits. In a study from urban Burkina Faso, two dietary

patterns were identified that were positively associated with the economic levels of

households and with food expenditure [78]. One pattern was labeled as “snacking”, that

was characterized by high consumption of fried foods, vegetable source fats,

sugar/sweetened products, cereals, vegetables, milk and yoghurt, non-fatty meats and

poultry, fresh fish, and roots and tubers. The second dietary pattern was labeled “modern

food” and was characterized by high consumption of non-fatty meats and poultry, fatty

processed meats, eggs, and low intake of cereals, nuts and seeds, beans and pulses, and

vitamin A-rich fruits and vegetables. The “modern food” pattern was positively associated

with overweight, whereas the “snacking” pattern was not associated with overweight. Also

in a study from urban Benin [96], a dietary pattern was identified, that included western

foods: This “transitional” labeled dietary pattern was characterized by a high intake of

white bread and pasta, local roots and tubers, potatoes, meat, eggs, milk and milk

products, sweets and soft drinks. Participants following the “transitional” pattern were of

higher SES and were more often born in the city. In the same study, a second dietary

pattern was identified that was labeled “traditional” and was characterized by high intake

of fruits and grains. Participants adhering to the “traditional” pattern were of lower SES

and were more often born in rural areas [96]. Keding et al. identified five dietary patterns

among 252 women from rural Tanzania; two of them included traditional foods [97]. The

“traditional coast” dietary pattern was characterized by high intake of fruits, nuts, starchy

plants, and fish. The “traditional inland” dietary pattern was characterized by high intake of

cereals, oil and fats, and vegetables. The ”traditional coast” pattern was negatively

associated with hemoglobin levels.

However, these five cross-sectional studies from SSA investigated associations between

dietary patterns and socioeconomic status, overweight or hypertension (Table 5).

Therefore, no previous study examined the association between dietary patterns and type

2 diabetes in a SSA population.

INTRODUCTION 13

Table 5: Dietary patterns derived by exploratory methods among SSA populations

First

author

(year)

Study

population

Study

design

Method

Dietary

assessment

method

Pattern

exposures

Outcome(s)

Main results

Maruapula

et al.

(2007) [99]

1086 elderly,

urban and

rural

Botswana

cross-

sectional

study

factor

analysis

FFQ

"beer" pattern

socioeconomic

conditions

" beer" pattern was negatively associated with

female gender: ß-coefficient (standard error) =

-0.645 (0.108), p<0.001

"meat/fruit" pattern

"meat/fruit" pattern was negatively associated

with religious affiliation to protestant churches:

ß-coefficient (standard error) = 0.481 (0.125),

p<0.001

"vegetable/bread"

pattern

"vegetable/bread" pattern was negatively

associated with rural living: ß-coefficient (standard

error) = -0.646 (0.107), p<0.001

"seasonal

produce" pattern

"seasonal produce" pattern was positively

associated with snacking: ß-coefficient (standard

error) = 0.515 (0.202), p = 0.011

"milk/tea/candy"

pattern

"milk/tea/candy" pattern was positively associated

with religious affiliation to protestant churches: ß-

coefficient (standard error) = 0.281 (0.133),

p = 0.035

Sodjinou

et al.

(2009)

[96]

200 men and

women, urban

Benin

cross-

sectional

study

cluster

analysis

three 24HDR

"traditional"

pattern

overall diet quality

and

sociodemographics

participants of "traditional" pattern had lower

socioeconomic status and were more often born

in rural area

"transitional"

pattern

participants of "transitional" pattern were better

educated and were more often born in urban area

Nkondjock

et al.

(2010) [98]

571 members

of defence

forces, urban

Cameroon

cross-

sectional

study

factor

analysis

FFQ

"fruit and

vegetable" pattern

hypertension

"fruit and vegetable" pattern was inversely

associated with hypertension: OR [95% CI] for

highest quartile = 0.41 [0.20-0.83]

"meat" pattern

"meat" pattern was not associated with

hypertension: OR [95% CI] for highest quartile =

1.12 [0.57-2.18]

FFQ: food frequency questionnaire; 24HDR: 24 hour dietary recall; ß: beta-coefficient; CI: confidence interval; OR: odds ratio, SSA: sub-Saharan Africa

INTRODUCTION 14

Table 5 continued

First

author

(year)

Study

population

Study

design

Method

Dietary

assessment

method

Pattern

exposures

Outcome(s)

Main results

Becquey

et al.

(2010)

[78]

1072 adults,

urban Burkina

Faso

cross-

sectional

study

principal

component

analysis

FFQ

"snacking" pattern

overweight

both pattern were positively associated with

economical level of households and with food

expenditures

"modern food"

pattern

“modern food” pattern was positively associated

with overweight: OR [95% CI] = 1.19 [1.03-1.36]

snacking' pattern was not associated with

overweight: OR [95% CI] = 1.04 [0.95-1.13]

Keding et

al. (2011)

[97]

252 middle-

aged women,

rural

Tanzania

cross-

sectional

study

principal

component

analysis

one 24HDR

"traditional-coast"

pattern

BMI, hemoglobin

level,

socioeconomic

status

"purchase" pattern was positively correlated with

BMI

"traditional-inland"

pattern

"traditional coast" and "purchase" pattern were

negatively correlated and "animal products"

were positively correlated with hemoglobin level

"purchase"

pattern

"animal products" pattern was positively

associated with wealth

"pulses" pattern

"animal products"

pattern

FFQ: food frequency questionnaire; 24HDR: 24 hour dietary recall; ß: beta-coefficient; CI: confidence interval; OR: odds ratio

INTRODUCTION 15

1.3.4 Dietary patterns and type 2 diabetes

Numerous studies have identified prevailing dietary patterns by exploratory methods in

US, European and Asian populations, and investigated their association with the risk of

type 2 diabetes [100-107]. Most of them applied exploratory factor analysis [95-100].

Although the identified dietary patterns were somewhat population specific, there were

notable similarities between the studies from Europe and USA. Most of these studies have

found a healthy pattern (labeled as “prudent” dietary pattern), characterized by a high

consumption of healthy foods such as fruits, vegetables, fish, poultry and whole grains,

associated with a reduced risk of type 2 diabetes [103, 104, 108, 109], and a less healthy

pattern (labeled as “Western” dietary pattern), characterized by a high consumption of

foods such as processed and red meats, fried foods, sweets and desserts, and refined

grains, related to a higher risk of type 2 diabetes [101, 104, 109].

With regard to RRR, only few epidemiological studies investigated the association

between dietary patterns and type 2 diabetes among Caucasian populations [95,110-115].

These prospective studies among Caucasian populations are summarized in Table 6. The

studies identified dietary patterns by the use of different biomarkers or nutrients as

response variables including inflammatory biomarkers [111, 113], HOMA-IR (homeostasis

model assessment for insulin resistance) [112], HbA1c, high-density lipoprotein (HDL)-

cholesterol, inflammatory marker (C-reactive protein (CRP)) and adiponectin [110] or ratio

of polyunsaturated fat intake to saturated fat intake, fiber, dietary magnesium and alcohol

intake [95] that were all strongly associated with type 2 diabetes. Despite the different

biomarkers selected as response variables, there are some similarities between the

dietary patterns: processed meat, sugar-sweetened beverages and refined grains were

main contributors of these patterns and positively associated with type 2 diabetes [111-

113]. Two studies investigated the generalizability of the associations between the RRR-

derived dietary patterns and type 2 diabetes among independent European [114] and US

populations [115]. The European Prospective Investigation into Cancer and Nutrition

(EPIC)-InterAct study reported a good generalizability for three RRR dietary pattern

scores based on the American Nurses’ Health Study (NHS), the German EPIC-Potsdam

Study and the British Whitehall II Study (WHS) [114]. In contrast, the American

Framingham Offspring Study found a good generalizability for the American NHS derived

dietary pattern, but the dietary patterns based on the European studies (EPIC-Potsdam

Study and WHS) were significantly less predictive for type 2 diabetes risk [115].

INTRODUCTION 16

Table 6: Dietary patterns derived by reduced rank regression and type 2 diabetes risk among Caucasian populations

First author

(year)

Study population

Study

design

Dietary

assessment

method

Predictor/Response variables

Pattern exposures

Main results

Hoffmann et

al. (2003) [95]

European Prospective

Investigation into

Cancer and Nutrition

(EPIC)-

Potsdam Study:

192 incident cases and

385 controls

nested

case-

control

study

semi-

quantitative

FFQ with

148 food

items

49 food groups as predictors

and ratio of polyunsaturated fat

intake to saturated fat intake,

fiber intake, dietary magnesium

intake, and alcohol as

responses

high intake of alcohol

and fiber and low in

magnesium

4 factors were identified but only the

4th were associated wi

th diabetes

(RR [95% CI] for 1SD: 0.68 [0.54-

0.85])

Heidemann

et al. (2005)

[110]

EPIC-Potsdam Study:

192 incident cases and

382 controls

nested

case-

control

study

semi-

quantitative

FFQ with

148 food

items

48 food groups as predictors

and HbA1c, HDL-cholesterol, C-

reactive protein (CRP) and

adiponectin as responses

high intake of fresh

fruit and a low intake

of high-

caloric soft

drinks, beer, red meat,

poultry, processed

meat, legumes and

bread (excluding

wholegrain bread)

4 pattern scores were obtained. First

score were used for further analyses.

This pattern was positively associated

with HDL-cholesterol and adiponectin

and inversely with CRP and HbA1c

pattern was inversely associated with

diabetes (OR

[95% CI] comparing

extreme quintiles: 0.27 [0.13–0.64])

Schulze et

al. (2005)

[113]

Nurses’ Health Study

(NHS): 656 type 2

diabetes cases, 694

controls

nested

case-

control

study

semi-

quantitative

FFQ

39 food groups as predictors

and six inflammatory biomarker

as responses: interleukin 6 (IL-

6), soluble tumor necrosis factor

receptor 2 (sTNFR2), C-reactive

protein (CRP), E-selectin,

soluble intracellular cell

adhesion molecule 1 (sICAM-1),

and soluble vascular cell

adhesion molecule 1 (sVCAM-1)

High consumption in

sugar-

sweetened soft

drinks, refined grains,

diet soft drinks, and

processed meat but

low in wine, coffee,

cruciferous

vegetables, and

yellow vegetables,

dietary pattern was positively

correlated with all inflammatory

biomarkers)

pattern was associated with an

increased risk of diabetes (OR [95%

CI] comparing extreme quintiles: 3.09

[1.99-4.79])

NHS:1517 incident

cases among 35,340

women

prospective

cohort study

RR [95% CI] comparing extreme

quintiles: 2.56 [ 2.10-3.12], p for trend

<0.001)

NHS II:724 incident

cases among 89,311

women

prospective cohort

study

relative risks [95% CI] comparing

extreme quintiles: 2.93 [2.18- 3.92], p

for trend <0.001)

FFQ: food frequency questionnaire, EPIC: European Prospective Investigation into Cancer and Nutrition, NHS: Nurses’ Health Study; CI: confidence interval; OR: odds ratio; RR:

relative risk

INTRODUCTION 17

Table 6 continued

First author

(year)

Study

population

Study

design

Dietary

assessment

method

Predictor/Response

variables

Pattern exposures

Main results

McNaugthon

et al. (2008)

[112]

Whitehall II

Study (WHS);

427 incident

cases among

7,339

participants

prospective

cohort study

FFQ with 127

food items

71 food groups (excluding

alcohol) as predictors and

HOMA-IR as response

high consumption of low-

calorie/diet soft drinks, onions,

sugar-sweetened beverages,

burgers and sausages, crisps and

other snacks, and white bread

and low consumption of medium-

/high-fiber breakfast cereals, jam,

French dressing/vinaigrette, and

whole meal bread

pattern was positively

correlated with HOMA-IR

(r= 0.24, P < 0.0001)

HR [95% CI] comparing

extreme quartiles: 2.95

[2.19-3.97]

Liese et al.

(2009) [111]

Insulin

Resistance

Atherosclerosis

Study; 144

incident diabetes

cases among

880 participants

prospective

cohort study

semi-

quantitative

FFQ with 114

food items

33 food groups as

predictors and plasminogen

activator inhibitor-1 (PAI-1)

and fibrinogen as

responses

High intake in red meat, low-fiber

bread and cereal, dried beans,

fried potatoes, tomato vegetables,

eggs, cheese, and cottage

cheese and low intake of wine

pattern was positively

associated with PAI-1 and

fibrinogen

OR [95% CI] comparing

extreme quartiles was 4.3

[1.7-10.8]

FFQ: food frequency questionnaire, EPIC: European Prospective Investigation into Cancer and Nutrition, NHS: Nurses’ Health Study; WHS: Whitehall II Study; HOMA-IR:

homeostasis model assessment for insulin resistance; CI: confidence interval; HR: hazard ratio; OR: odds ratio

INTRODUCTION 18

Table 6 continued

First

author

(year)

Study population

Study

design

Dietary

assessment

method

Predictor/Response

variables

Pattern exposures

Main results

Imamura et

al. (2009)

[115]

Framingham

Offspring Study;

158 incident

diabetes cases

among 2879

participants

prospective

cohort

study

semi-

quantitative

FFQ with 126

food items

3 RRR analyses: 39 food

groups of the NHS, 48 food

groups of the EPIC-Study,

and 71 food groups of the

WHS as predictors and BMI,

fasting glucose, triglyceride,

HDL-cholesterol, and

hypertension (based on

elevated systolic and/or

diastolic blood pressure or

hypertension treatment) as

responses.

NHS dietary pattern: high intake in

red meat, processed meat,

margarine, refined grains, low-

calorie soft drinks and caloric soft

drinks, french fries, fried foods and

pizza and low in dark-yellow

vegetables, green leafy vegetables,

whole grains, wine and other

alcoholic beverages

EPIC dietary pattern: high intake in

red meat, processed meat,

margarine, refined grains, low-

calorie soft drinks, french fries and

pizza and low in wine, beer and

other alcoholic beverages

WHS dietary pattern: high

consumption of beef burgers and

sausages, refined grains, low-

calorie soft drinks, fried foods and

pizza and low in dried fruit (raisins)

Dietary patterns were

positively associated

with diabetes (RR

[95% CI] comparing

extreme quintiles:

NHS: 3.22 [1.93-5.38]

EPIC: 5.46 [3.02-9.87]

WHS: 4.02 [2.39-6.75]

Kröger et

al. (2014)

[114]

EPIC-Study:

France, Spain, UK,

Netherlands,

Germany, Sweden,

Denmark; 9682

incident cases;

(661 incident cases

in subcohort)

12595 participants

in the subcohort

case-cohort

study

country-

specific

validated

dietary

questionnaires

(quantitative

and semi-

quantitative)

RRR1 derived in the NHS

using six inflammatory

markers as responses

RRR2 derived in the EPIC-

Potsdam study using HbA1c,

HDL-cholesterol, CRP and

adiponectin as responses

RRR3 derived in the WHS

using HOMA-IR as response

Dietary patterns were

inversely associated

with diabetes (RR

[95% CI] comparing

extreme quintiles:

RRR1: 0.76 [0.67-0.86]

RRR2: 0.85 [0.75-0.97]

RRR3: 0.65 [0.58-0.73]

FFQ: food frequency questionnaire, EPIC: European Prospective Investigation into Cancer and Nutrition, NHS: Nurses’ Health Study; WHS: Whitehall II Study; CRP: C-reactive

protein; HOMA-IR: homeostasis model assessment for insulin resistance; CI: confidence interval; RR: relative risk; RRR: reduced rank regression

INTRODUCTION 19

1.4 Public health relevance

Type 2 diabetes, once a problem in the developed countries, has now become a global

public health challenge, particularly in the developing countries. While in 2014 an overall

diabetes prevalence of 5.2% for the African region was estimated, SSA is expected to

witness the highest increase in the number of people with diabetes within the next 20

years [2]. This trend has been rising simultaneously with overweight and obesity in this

region [3]. However, SSA is still dealing with infectious diseases such as malaria, HIV-

infections and tuberculosis [5]. This double burden of communicable and non-

communicable diseases poses a major public health challenge in this region, where

financial and health resources are limited. In contrast to the developed countries, where

the majority of people with diabetes are older than 60 years, diabetes occurred in people

in the economically productive age of 30-45 years in SSA [116]. The late diagnosis of

diabetes in SSA, coupled with a poor glycemic control and inequalities in access to anti-

diabetic medications, leads to an early onset of diabetes-related complications and

premature deaths [116, 117]. Furthermore, Africa had the highest prevalence of

undiagnosed diabetes (62%) and the lowest diabetes healthcare expenditure worldwide

[2]. This demonstrates the inadequate response to this growing health burden.

Researchers and policy makers are still focused primarily on the prevention of infectious

diseases. However, an early identification and treatment is essential to avoid severe

complications of diabetes. In SSA, microvascular complications such as retinopathy,

nephropathy and neuropathy are more common than macrovascular complications such

as coronary heart disease, peripheral arterial disease and stroke [118]. A systematic

review found that the prevalence for diabetic retinopathy ranged from 30.2% to 31.6% in

population-based studies and from 7.0% to 62.4% in diabetes clinic based studies [119].

At the time of initial diagnosis 21-25% of the type 2 diabetes patients have a retinopathy

[118]. Although, macrovascular complications are low compared to other regions,

prevalences within type 2 diabetes patients are rising in SSA [120]. Awareness of risk

factors and diabetes in the general population seems to be low [121, 122]. Furthermore,

several studies have consistently shown that diabetes patients have a poor knowledge of

their conditions and how to manage them [123-127]. Contrasting the increasing

prevalences of type 2 diabetes and obesity and the growing public health relevance,

studies on type 2 diabetes are remarkable scarce in SSA.

INTRODUCTION 20

1.5 Objectives and research questions

Although obesity and the nutritional behavior are the main risk factors for type 2 diabetes

[6], essential knowledge on these two risk factors and their relationships with type 2

diabetes barely exist in SSA. Particularly in urban areas of SSA, the prevalence of

overweight and obesity is increasing dramatically [3, 4]. However, the associations

between obesity measures and the risk of type 2 diabetes are only insufficiently examined

in this region [44-46, 71-73]. It is still controversial which measures of overweight and

obesity best reflect an increased risk for type 2 diabetes in non-Caucasian populations

[128]. Furthermore, it is unclear, whether the usual cut-offs for BMI, WC and WHR, which

have been mainly derived among Caucasian populations, are appropriate for other

populations [52-54]. With respect to the nutritional behavior, studies on dietary patterns

from SSA are limited. Only a few cross-sectional studies derived dietary patterns by

exploratory methods [78, 96-99]. However, no previous study investigated their

relationship with type 2 diabetes among an SSA population. Furthermore, no previous

study from SSA applied the RRR method to derive health-related dietary patterns.

To address these research gaps mentioned before, it is of great interest to assess the

impact of these two major risk factors for type 2 diabetes among an SSA population.

Thus, the first aim of this thesis is to investigate the associations between various

anthropometric measures and type 2 diabetes and to assess the appropriateness of

specific cut-off points for BMI, WC and WHR in an urban Ghanaian study population in the

Kumasi Diabetes and Hypertension study. The second aim of this thesis is to describe the

dietary behavior and to examine the associations between dietary patterns derived by an

exploratory factor analysis and type 2 diabetes. Finally, the third aim is to identify a dietary

pattern by using the reduced rank regression (RRR) approach and to evaluate the

association between this dietary pattern and type 2 diabetes.

INTRODUCTION 21

In particular, the following research questions were addressed:

• Is there an association between various anthropometric measures and type

2 diabetes risk?

• Which anthropometric measure has the best discriminative power for the

identification of type 2 diabetes cases?

• Are the current cut-off points for BMI, WC and WHR, which have been

mainly derived among Caucasian populations, transferable to this Ghanaian

study population?

• What are the optimal BMI, WC and WHR cut-off points for the identification

of type 2 diabetes in this study population?

• How is the macronutrient and energy intake and dietary behavior

characterized by one 24 hour dietary recall and a local-specific food

frequency questionnaire in this study population?

• What dietary patterns can be identified by an exploratory factor analysis and

are those patterns associated with type 2 diabetes?

• What dietary pattern can be derived by using food items as predictor

variables and biomarkers that are related to diet and the pathophysiology of

type 2 diabetes as response variables in reduced rank regression, and is this

pattern associated with the risk of type 2 diabetes?

STUDY POPULATION AND METHODS 22

2 STUDY POPULATION AND METHODS

2.1 Kumasi Diabetes and Hypertension study

2.1.1 Study setting

The Kumasi Diabetes and Hypertension (KDH) study was conducted at the Komfo Anokye

Teaching Hospital (KATH) in Kumasi, the Ashanti Region of Ghana, between August 2007

and June 2008 [129, 130]. According to the World Bank, Ghana has a population of

approximately 26.79 million inhabitants and belongs to the “low-middle income countries”

with a gross domestic product (GDP) per year of 38.65 billion US dollars and a gross

national income (GNI) per capita of 1600 US dollars [131]. The official language is English

and Akan is the most widely spoken indigenous language in this country [132]. Since 1957

Ghana had a compulsory school attendance and since 1996 all school-age children

received a free and compulsory quality primary education [133]. These reforms helped to

improve the education system and to reduce the literacy rates. In 2010, the total adult

literacy rate (percentage of persons aged 15 and above who can read and write) was

71.5%, with a notable gap between men (78.3%) and women (65.3%) [131]. In 2003, the

government of Ghana introduced a national health insurance scheme to replace the

existing cash and carry system (pay as you access) and to provide an equitable access

and financial coverage for health care services to all Ghanaian citizens [134]. The

prevalence of diabetes is about 6% in Ghana [15]. Diabetes patients receive care at the

KATH, which is the second largest tertiary hospital in the Kumasi Metropolitan area. At

KATH, the diabetes and hypertension clinics are frequented each by > 100 patients/week

and this hospital has a capacity of 1000 beds.

2.1.2 Study design and study population

The primary aim of this unmatched case-control study was to identify risk factors for type

2 diabetes (and hypertension). Cases were recruited from a pool of patients attending the

diabetes center (n=495) and hypertension clinic (n=451) at the KATH. Friends, neighbors

and community members (n=222) were encouraged to participate as potential controls.

Further preliminary controls came from the outpatient department (n=150) and hospital

staff (n=148).

Type 2 diabetes cases were defined as having fasting plasma glucose (FPG) ≥ 7 mmol/L

and/or documented anti-diabetic medication [135]. Hypertension was defined as having a

mean blood pressure ≥ 140/90 mmHg and/or documented antihypertensive treatment.

Controls were negative for both conditions.

STUDY POPULATION AND METHODS 23

Inclusion criteria were an informed written consent, age ≥ 18 years and residence in the

Kumasi Metropolitan area or adjacent districts. Exclusion criteria were ambiguous results

of glucose and/or blood pressure measurements, known liver cirrhosis or a pregnancy. All

participants were informed on purpose and conduct of the study and provided consent by

signing or thumb printing on a consent form. The study protocol was reviewed and

approved by the Ethics Committee, School of Medical Sciences, University of Science and

Technology, Kumasi.

2.1.3 Data collection

The participants were instructed to remain fasting from 10:00 p.m. on the evening before,

to abstain from alcoholic drinks, smoking and excessive physical activity. On the

examination day, participants underwent a routine physical and clinical examination

including anthropometric measurements and a personal interview on socio-demographic

background, medical history, economic status and physical activity. In addition, fasting

venous blood and urine samples were collected.

Anthropometric and body composition measurements

During the physical examination anthropometric measurements were taken in the

standing position by a trained nurse following standardized procedures. The participants

wore light clothes without shoes. Weight was measured on an electronic personal scale to

the nearest 0.1 kg and height with a stadiometer to the nearest 0.1 cm. Waist

circumference (WC) was determined two fingers’ breadth below the lowest rib using a

measuring tape and hip circumference (HC) at the level of the widest diameter around the

gluteal protuberance (all devices, Seca, Germany). Body composition was assessed by

bioelectric impedance (BIA) (50 kHz; Nutrigard-S, NutriPlus 1.0; Data Input Germany).

BMI was calculated as weight/(height)2 [kg/m²]. Cut-offs for overweight and obesity were

defined according to the WHO classifications (Table 2): Overweight was defined as BMI

25.0–29.9 kg/m² and obesity as BMI ≥ 30.0 kg/m². The WC cut-offs corresponding to BMI-

defined overweight were defined as WC ≥ 80 cm in women and ≥ 94 cm in men, central

obesity was defined as WC ≥ 88 cm in women and ≥ 102 cm in men according to the

WHO thresholds (Table 3). WHR was calculated as WC/HC and WHtR as WC/height. To

identify the optimal cut-off point, central obesity was defined as WHR ≥ 0.85 for women

and ≥ 0.90 for men as recommended by the WHO (Table 3) [24].

STUDY POPULATION AND METHODS 24

Dietary assessment

Food frequency questionnaire (FFQ)

For the nutritional assessment, a locally specific FFQ was designed. This FFQ is depicted

in the appendix (Figure S1). In face-to-face interviews, trained nurses speaking the local

language applied the FFQ to all participants in a separate air-conditioned room after

breakfast. The FFQ queried for the usual weekly consumption of fifty-one food items in ten

food categories over the past 12 months: ‘During the past 12 months, how often did you

usually consume the following foods per week?’ Food categories of the FFQ were based

on the latest Ghana Demographic and Health Survey [136]. These categories were

starchy roots and plantain; cereals and cereal products; animal products; legumes, nuts

and oilseeds; fruits; vegetables; fats and oils; salt and spices; sweets; and liquids

(appendix: Table S1). No portion sizes were available. Thus, the FFQ covered

frequencies, but not quantities, of food consumption. There were six response categories:

never; seldom (1 x per week); 1–2 x per week; 3–4 x per week; 5–6 x per week; daily.

This FFQ has not been validated yet.

24 hour dietary recall (24HDR)

To describe energy and macronutrient intakes at the study population level, a single

24HDR was administered to each participant. An exemplary page of one 24HDR is shown

in the appendix (Figure S2). Trained study personnel speaking the local language applied

the 5-steps multiple-pass method in face-to-face interviews. Food and beverage

consumption between midnight of the pre-last day and midnight of the last day was

recorded in detail. Time and occasion of the meals were included. All consumed foods,

their mode of preparation and their portion sizes, estimated by Ghanaian household

utensils, were documented. The daily intakes of energy (kcal/d), protein, carbohydrates,

total fat and dietary fiber (g/d) were derived from the 24HDR. The estimated portion sizes

were converted into grams, and Ghana specific nutrient composition tables [137-139]

were used to translate the consumed foods into macronutrient and energy intake. For

comparisons of macronutrient intakes between diabetes cases and controls, the

respective values were standardized per 1000 kcal/d.

STUDY POPULATION AND METHODS 25

Covariate assessment

Socio-demographic variables

Socio-demographic data were documented in face-to-face interviews by trained study

personnel speaking the local language. These comprised age, sex, residence, ethnic

group, education (none, primary, secondary, tertiary or other), literacy (able or unable to

read and write), occupation (subsistence farmer, commercial farmer, casual labourer,

artisan, trader, businessman/woman, public servant, unemployed or other), household

assets (electricity, pipe water, fan, fridge, cupboard, radio, tv, bicycle, motor bike, car,

truck and tractor and cattle (yes/no)) and the number of people living in the household.

For the construction of a SES sum score, which was developed by Franziska Jannasch in

2013 [140], the common proxy markers education, occupation and income were used.

The exact procedure of the construction is depicted in Figure 1. First a new variable was

constructed by combining the information regarding education and literacy. This new

variable with four characteristics covered information about having a graduation and being

able to write and read; points from 0 to 3 were given. Occupation, originally a variable with

nine characteristics, was condensed to a new variable with five characteristics, given the

points 0 to 4. Due to the problems of a valid ascertainment of income in Ghana, a list of 11

household assets was recorded. An income score ranging from 0 to 12 points was

constructed based on these assets and the number of people living in the household. The

income score was divided into quartiles, given the points 0 to 3. To create the overall SES

sum score the points of education, occupation and the income score were summed up to

a score ranging from 0 to 10 points. The score were divided in three groups: 0 to 4 points

were defined as very low SES, 5-8 points as low SES and 9 to 10 points were defined as

moderate SES.

STUDY POPULATION AND METHODS 26

Figure 1: Construction of the socioeconomic status (SES) sum score [141]

The common proxy markers education, occupation and income were used to construct the SES

sum score ranging from 0 to 10 points. 0 to 4 points were defined as very low SES, 5-8 as low SES

and 9 to 10 points were defined as moderate SES.

Medical history

A questionnaire for medical history was applied in these interviews including questions on

own and family history of diabetes (yes or no), medications, previous and current diseases

and smoking behavior (never, previous or current).

Physical activity

Physical activity was recorded as the duration (min/week) and type (i.e. intensity) of work-

related, transportation-related and leisure-time physical activity. These data were

translated into daily energy expenditure (kJ/d) as the sum of metabolic equivalents

corresponding to activity intensity (ml/kg per min) x body weight (kg) x duration (min)

[142].

Blood pressure

During the physical examination blood pressure and heart rate were measured in

triplicates on a comfortable chair after 10 min resting time in an air-conditioned room (M8

Comfort, Omron, Japan). Blood pressure was determined as the mean of all three

measurements.

STUDY POPULATION AND METHODS 27

Plasmodium falciparium infection

Detection of P. falciparum was performed by primer-specific PCR [143].

Blood collection and biomarker measurement

Fasting blood samples were collected into fluoride and serum tubes from each participant.

In fluoride whole blood, FPG was measured immediately after blood collection by

photometry (Glucose 201+ Analyzer, HemoCue, Ångelholm, Sweden). FPG is presented

as plasma equivalents. The inter-assay coefficient of variation (CV) ranged between 1.7

and 6.1%. Serum tubes were centrifuged at 8,000 rpm for 10 min and aliquots were frozen

at -20°C. On dry ice, samples were transferred to the German Institute of Human Nutrition

Potsdam-Rehbruecke, where following biomarkers were measured using standard

techniques: triglycerides, HDL-cholesterol and total cholesterol were measured by

colorimetric assays (ABX Pentra400, Horiba Medical, Reichenbach, Germany). The inter-

assay CVs were 4.5%, 1.8% and 3.0%, respectively. Low-density lipoprotein (LDL)

cholesterol was calculated according to the Friedewald formula [144]. If triglycerides were

> 3.0 mmol/L, LDL-cholesterol was quantified directly. Total adiponectin concentration

was measured using a commercially available ELISA with intra- and inter-assay CVs of

4.9% and 6.7% (BioVendor, Heidelberg, Germany). Serum C-reactive protein (CRP) was

quantified by turbidimetric immunoprecipitation (ABX Pentra400, Horiba Medical,

Germany) with an inter-assay CV of 2.2%. The Homeostatic Model Assessment (HOMA)

is a method used to quantify insulin resistance [145]. HOMA-IR was calculated according

to the formula = Fasting Insulin (µU/ml) x Fasting plasma glucose (mmol/L) / 22.5 [145].

Selection of response variables for RRR

The three response variables adiponectin, HDL-cholesterol and triglycerides were chosen

because they are affected by diet and are related to the pathophysiology of type 2

diabetes: Randomized clinical trials showed that a diet high in complex carbohydrates,

mono-unsaturated fatty acid (MUFA), and fiber and fish intake increased the adiponectin

concentrations [146, 147]. In addition, moderate alcohol consumption [148] and a diet with

a low glycemic load [149] raised the adiponectin levels. Furthermore, many studies have

reported that adiponectin has an insulin-sensitizing effect and anti-inflammatory properties

and higher adiponectin levels are associated with a lower type 2 diabetes risk [150].

A meta-analysis of experimental studies which investigated the effect of moderate alcohol

intake on lipids found that a dose of 30g ethanol per day increased HDL-cholesterol and

triglyceride concentrations [151]. Furthermore, a diet low in carbohydrates [152] or high in

MUFA [153] increased the concentrations of HDL-cholesterol and decreased triglycerides.

In contrast, a high intake of trans-fatty acids decreased the concentrations of HDL-

cholesterol and raised triglyceride concentrations [154]. HDL-cholesterol plays an

STUDY POPULATION AND METHODS 28

important role in glucose metabolism; it modulates mechanisms including insulin

sensitivity, insulin secretion and glucose uptake by skeletal muscles [155]. In addition,

type 2 diabetes is associated with decreased HDL-cholesterol concentrations and

increased triglyceride concentrations [156].

2.1.4 Analytical study population

For the present thesis data from the KDH study were used. Participants without diabetes

(controls + hypertensive participants) were defined as controls for this work. Figure 2

shows the exclusion criteria and respective number of participants of the analytical study

population within the KDH study. From the 1466 participants, that were initially included in

the KDH study, 245 were excluded due to missing information on nutrition (141),

anthropometry (39), socio-economic status (SES, 31) and genetic polymorphisms (34).

Thus, 1221 individuals (679 controls, 542 diabetes cases) remained for the analyses

(Figure 2). The number of excluded participants was similar between diabetes cases and

controls. The baseline characteristics did not differ between excluded and included

participants. For example: age, gender and BMI were fairly comparable between the

excluded participants and those remaining in the analyses (mean age: 52.0 ± 15.1 vs.

50.4 ± 15.3 years; gender: 77.5% vs. 75.5% women; mean BMI: 26.0 ± 5.3 vs. 25.8 ± 5.2

kg/m2).

The analytical study population of 1221 participants was considered for the evaluation of

the associations between anthropometric measures and type 2 diabetes, characterization

of the dietary behavior and for the evaluation of the associations between dietary patterns

derived by exploratory factor analysis and type 2 diabetes. For the evaluation of the

association between a dietary pattern derived by RRR analysis and type 2 diabetes,

further 15 individuals were excluded due to missing data on biomarkers. Hence, this

analysis comprised 1206 individuals (668 controls, 538 diabetes cases).

STUDY POPULATION AND METHODS 29

Figure 2: Flow diagram of analytical study population

FFQ: food frequency questionnaire, 24HDR: 24 hour dietary recall, SNPs: single nucleotide

polymorphism

2.2 Statistical analysis

All statistical analyses were performed using SAS statistical software (version 9.4, SAS

Institute, Cary, NC, USA).

The general characteristics of the study population were compared between type 2

diabetes cases and controls. Arithmetic means and respective standard deviation or

median with interquartile range were computed for continuous variables and participant

number and respective proportion were computed for categorical variables. To test for

significant differences in baseline characteristics between type 2 diabetes cases and

controls the non-parametric Mann-Whitney U test was used for continuous and χ²-test for

categorical variables. Generally, all statistical test were two-sided with a significance level

at < 0.05.

STUDY POPULATION AND METHODS 30

2.2.1 Anthropometry

All analyses were performed separately for women and men, because of sex-specific

differences in anthropometry.

Anthropometric characteristics

Anthropometric characteristics and prevalences of overweight and obesity between

diabetes cases and controls were compared by Mann-Whitney-U-test for continuous

variables and by χ²-test for categorical variables. Age-adjusted Spearman correlations

were used to assess the relationship between anthropometric measures.

Associations between anthropometric measures and type 2 diabetes

To evaluate associations between anthropometric measures and type 2 diabetes, the

measures of interest (BMI, WC, HC, WHR and WHtR) were categorized into quintiles

based on their distribution among the controls. Odds ratios (OR) and corresponding 95%

confidence intervals (CI) for the association between various anthropometric measures

and type 2 diabetes were evaluated across quintiles and per 1 standard deviation (SD)

difference in anthropometric measures using multivariate-adjusted logistic regression. The

lowest quintile was used as the reference category among women and, for sample size

reasons, the two lowest quintiles among men. The significance of a linear trend across the

categories was tested by assigning each participant the median of a category and by

modeling this value as a continuous variable.

When evaluating the associations between anthropometric measures and type 2 diabetes,

various confounders were considered: the basic model was adjusted for age (model 1),

the second model was further adjusted for smoking status (current or never/ex-smoker),

family history of diabetes (yes/no), educational attainment (any/none), fat and fiber intake

(g/1000kcal) (model 2). An alternative second model was adjusted for age, smoking status

(current or never/ex-smoker), family history of diabetes (yes/no), SES sum score, fat and

fiber intake (g/1000kcal) (model 2a). The final model was adjusted for age, smoking status

(current or never/ex-smoker), family history of diabetes (yes/no), SES sum score, fat and

fiber intake (g/1000kcal) and BMI to test whether the associations are independent of

general obesity (model 3).

Sensitivity analyses

Sensitivity analyses were performed to test the robustness of the results: To test whether

the associations between anthropometric measures and type 2 diabetes was confounded

by systolic and diastolic blood pressure and P. falciparum infection [129], the second

model (model 2a) was further adjusted for these factors. Interactions of the association

STUDY POPULATION AND METHODS 31

between anthropometric measures and type 2 diabetes with the SES sum score (very low

SES vs. low SES vs. moderate SES) were tested by performing stratified analyses,

evaluating the significance of cross-product terms.

Discrimination of anthropometric measures

Receiver operating characteristic (ROC) curve analysis was used to compare the

discriminative abilities of anthropometric measures for identifying diabetes cases. A ROC

curve is a graphical plot of sensitivity vs. false-positive rate (1-specificity) over all possible

cut-off points for classifying patients as positive vs. negative as it is exemplified in

Figure 3. The area under the ROC curve (ROC-AUC) discriminates between diseased

and non-diseased persons and estimates the probability that the predicted risk for a

diseased person is higher than that for a non-diseased person [157]. A ROC-AUC of 0.5

reflects an uninformative model such as tossing a coin, whereas an ROC-AUC of 1.0

represents perfect discrimination [158]. The ROC-AUC estimated the discriminative

capabilities of those anthropometric measures associated with diabetes. ROC-AUCs and

95% CI were calculated and compared using the method by DeLong et al. [159]. For the

ROC-AUC analysis the multivariate-adjusted model (model 2a) was applied, including

age, diabetes family history, SES sum score, smoking status, fiber intake, fat intake and

energy expenditure.

Figure 3: Example of a receiver operating characteristic (ROC) curve

The area under the curve (AUC) is 0.785 [95%CI: 0.756-0.815]. The Youden index indicates the

point that is located nearest to the upper left corner as the ‘optimal’ cut-point with highest sensitivity

and specificity at the same time (red dot).

STUDY POPULATION AND METHODS 32

Examination of cut-offs

Sensitivity and specificity of sex-specific cut-off points for BMI, WC and WHR

recommended by the WHO [24] were estimated using ROC curve analysis. The Youden

index was computed to identify population-specific cut-off points of these measures for the

optimal differentiation between cases and controls. The Youden index is the point on the

ROC curve that is located nearest to the upper right corner with having highest sensitivity

and specificity at the same time (Figure 3). This maximum cut-point refers to the optimal

cut-point. The Youden index is derived from maximum (sensitivity + specificity – 1) and

ranges from 0 to 1 [160].

In sensitivity analysis diabetes cases with a poor glycemic control (FPG ≥ 7 mmol/L) were

excluded to investigate whether those cases had an impact on the cut-off points.

2.2.2 Factor analysis

An exploratory factor analysis was applied by using thirty-three food items or food groups

from the FFQ for the identification of dietary patterns. Of the fifty-one original food items of

the FFQ, some food items were collapsed or excluded as described in Table S1

(appendix). Briefly, “alcoholic beverages” were excluded because the majority of

participants (> 90%) never consumed such beverages as well as “water”, “tomatoes”, and

“pepper” which were consumed on a daily basis by all participants and thus did not

contribute to variation in the usual diet. Furthermore, “spices” were excluded as they did

not considerably contribute to energy and macronutrient intake. The single fruit items

(orange, mango, papaya, pineapple, banana and avocado) were grouped into one food

group “fruits” to avoid overrepresentation of fruit intake in the pattern analysis. Also,

“chocolate”, “ice cream” and “toffee” were grouped into one food group “sweets”, because

the consumption of these food items were rare.

To identify underlying dietary patterns, factor analysis was performed using the PROC

FACTOR procedure in SAS. The original food items were collapsed into latent factors

explaining the maximum of the total variance of these 33 food-item variables. Factors

were derived in decreasing order of importance: The first factor accounts for as much as

possible variation in the food items. The second factor accounts for as much as possible

of the remaining variation, and so on. The amount of this variation is reflected by the

eigenvalue. An orthogonal rotation (Varimax) was applied to ensure that the factors

remained uncorrelated and to improve interpretability. To detect the optimal number of

factors to be extracted, the criteria of an eigenvalue ≥ 1, the scree plot and plausibility of

the factors were used. A scree plot is a graph of the explained variance (eigenvalues)

STUDY POPULATION AND METHODS 33

plotted against the number of achieved factors and is useful to determine the optimal

number of factors to retain. A break between the factors allows separating factors with

large eigenvalues from those with small eigenvalues. The factors before the break are

assumed to be meaningful and are retained for analysis.

The factor score for each pattern was calculated as the sum of the z-standardized intakes

(mean = 0 and SD = 1) of 33 food items multiplied by an individual weight. Each

participant received a factor score for each identified dietary pattern. These scores were

used to rank participants according to the degree to which they conformed to each dietary

pattern.

Factors were derived among the total study population. Quintiles of dietary pattern scores

were constructed based on the distribution among the control group. Baseline

characteristics and frequencies of food intake were calculated across the quintiles of each

dietary pattern score among the control group.

Associations between dietary patterns and type 2 diabetes

Logistic regression analysis was applied to evaluate the associations between dietary

patterns and type 2 diabetes. ORs and 95% CI were calculated across the quintiles and

per 1 SD of the factor score. The significance of a linear trend across the categories was

tested by assigning each participant the median of a category and by modeling this value

as a continuous variable.

When evaluating the associations between dietary patterns and type 2 diabetes, different

confounders were considered: the basic model was adjusted for age and sex (model 1),

the second model was further adjusted for family history of diabetes (yes or no),

unemployment (yes or no), educational attainment (any or none), literacy (able or unable),

smoking status (current or never/ex-smoker) and daily energy expenditure (kcal/d) (model

2). An alternative second model was adjusted for age, smoking status (current or

never/ex-smoker), family history of diabetes (yes/no), SES sum score and daily energy

expenditure (kcal/d) (model 2a). The final model was adjusted for age, smoking status

(current or never/ex-smoker), family history of diabetes (yes/no), SES sum score, daily

energy expenditure (kcal/d), BMI and WHR (model 3).

Sensitivity analyses

In sensitivity analyses, different dietary pattern solutions (three to five factors) were

examined to identify meaningful dietary patterns (appendix: Table S5-7). Furthermore,

two-factor solutions were examined with single fruit and single vegetable items (appendix:

Table S8) or with various fruit groups (citrus fruits: orange, classical fruits: banana,

STUDY POPULATION AND METHODS 34

avocado, and exotic fruits: mango, papaya, and pineapple) and vegetable groups (leafy

vegetable: green leafy vegetables and lettuce, and vegetables: garden egg, okra, and

cucumber) (appendix: Table S9).

Finally, interactions of the association between the dietary pattern and type 2 diabetes

with age (<51 vs. ≥51 years), sex (female vs. male), BMI (<25.0 vs. 25.0-29.9 vs. ≥30

kg/m²), central obesity (<88/≥88 cm waist circumference in women and <102/≥102 cm

waist circumference in men) and SES sum score (very low SES vs. low SES vs. moderate

SES) were tested by performing stratified analyses, evaluating the significance of cross-

product terms.

2.2.5 Reduced rank regression

Spearman correlations were used to assess the relationship between the diabetes-related

biomarkers (response variables) among the control group.

The RRR approach was applied to derive a dietary pattern predictive of diabetes risk.

RRR uses two sets of variables called the predictors (X1,..., Xn) and responses (Y1,..., Ym).

The aim of the RRR method is to determine a linear combination of predictors (e.g. food

groups) that explain as much as possible variation in the responses (e.g. biomarkers) [95].

The RRR starts from the eigenvalues of the covariance matrix of the responses. Thereby,

a response score (Y-score) is created with weights from eigenvectors of the covariance

matrix of responses predicted by ordinary least squares regression. The numbers of Y-

scores is equal to the number of responses included in RRR. For this present thesis

adiponectin, HDL-cholesterol and triglycerides were chosen as response variables,

because they are affected by diet and are related to type 2 diabetes [146-156]. Hence

three response scores were created with the following linear equation:

Response score=

1Y1+

2Y2+

3Y3,

where Yi, i=1,2,3 are the standardized concentration of biomarkers with mean=0 and

standard deviation=1. The parameters

I, i=1,2,3 can be considered as score-specific

weights of the biomarkers.

In the next step, the Y-scores are projected onto the space of predictors forming a factor

score that is a linear combination of predictors. For this present thesis 35 food items were

used as predictors and therefore the factor scores are called dietary pattern scores in this

work. Three dietary pattern scores were created based on the following linear equation:

STUDY POPULATION AND METHODS 35

Dietary pattern score=

1Y1+

2Y2+

3Y3,

where Yi, i=1,…,35 are the standardized intake of the food items with mean=0 and

standard deviation=1

and

I, i=1,…,35 are the score-specific weights of the food items.

The three response scores and three dietary pattern scores form three pairs in which each

pair reflect the same unobserved or latent variable in different sets of original variables

(biomarker and food items) [95].

Of the 51 original food items of the FFQ, some food items were collapsed or excluded as

described in Table S1 (appendix). Thus, 35 food items were used as predictor variables

and serum concentrations of HDL-cholesterol, adiponectin and triglycerides as response

variables. Adiponectin and triglycerides were log-transformed because they were not

normally distributed. The RRR approach was applied to the total study population. The

number of extracted dietary pattern scores was equal to the number of response variables

(n=3). Only the first factor was considered for subsequent analyses because it explained

the largest amount of variation among the biomarkers and was biological plausible. Each

participant received a factor score for the identified dietary pattern. These scores were

used to rank participants according to the degree to which they conformed to the dietary

pattern. Based on the distribution among the control group quintiles of the dietary pattern

score were constructed. Socio-demographic, anthropometric and biomarker

characteristics as well as frequencies of food intake were calculated across the quintiles

of the dietary pattern score among the control group. The differences among categorical

variables (χ²-test) and linear trends among continuous parameters (trend test) were

assessed.

Associations between dietary pattern and type 2 diabetes

Logistic regression analysis was applied to evaluate the associations between the dietary

pattern and type 2 diabetes. ORs and corresponding 95% CI were calculated across the

quintiles and per 1 SD of the pattern score. The significance of a linear trend across the

categories was tested by assigning each participant the median of a category and by

modeling this value as a continuous variable.

When evaluating the association between the RRR-derived dietary pattern and type 2

diabetes, various confounders were considered: the basic model was adjusted for age

and sex (model 1), the second model was further adjusted for family history of diabetes

(yes or no), unemployment (yes or no), educational attainment (any or none), literacy

(able or unable), smoking (current or never/ex-smoker) and daily energy expenditure

STUDY POPULATION AND METHODS 36

(kcal/d) (model 2). An alternative second model was adjusted for age, sex, smoking

(current or never/ex-smoker), family history of diabetes (yes/no), SES sum score and daily

energy expenditure (kcal/d) (model 2a). The final model was adjusted for age, sex,

smoking (current or never/ex-smoker), family history of diabetes (yes/no), SES sum score,

daily energy expenditure (kcal/d), BMI and WHR (model 3).

Sensitivity analyses

Several sensitivity analyses were applied. First, interactions of the association between

the dietary pattern and type 2 diabetes with sex, general obesity (BMI <30/≥30kg/m²) or

central adiposity (waist circumference <102/≥102cm [men], <88/≥88cm [women]) were

tested by performing stratified analyses, evaluating the significance of cross-product

terms. The robustness of the results was examined by excluding participants with lipid-

lowering and anti-inflammatory drug intake. Also, the association between biomarkers of

glucose metabolism (FPG and HOMA-IR) and the dietary pattern score were examined

across quintiles among the controls. Finally, the dietary pattern score was simplified.

Therefore a stepwise linear regression with the first response score as dependent variable

and all 35 food items as independent variable was applied. Only food items that were

significantly associated with the response score were considered for the simplified score.

This score was calculated by summing up the unweighted standardized intake of these

nine food items, while retaining the direction of the factor loadings. Socio-demographic,

anthropometric and biomarker characteristics as well as frequencies of food intake were

calculated across the quintiles of the simplified dietary pattern score among the control

group. The differences among categorical variables (χ²-test) and linear trends among

continuous parameters (trend test) were assessed. Logistic regression analysis was

applied to evaluate the associations between the simplified dietary pattern and type 2

diabetes. Multivariate-adjusted ORs and corresponding 95% CI were calculated across

the quintiles and per 1 SD of the simplified pattern score. The significance of a linear trend

across the categories was tested by assigning each participant the median of a category

and by modeling this value as a continuous variable. To assess the importance of

individual food components of the dietary pattern score for type 2 diabetes each

component were sequentially removed from the simplified dietary pattern score. The

change in estimate (CIE) was calculated as the difference between the ORs divided by

the OR from the simplified score multiplied by 100.

RESULTS 37

3 RESULTS

3.1 Characterization of study population

The characteristics of 1221 participants of the KDH study stratified by diabetes status are

presented in Table 7. The study population was mainly female, middle-aged and of low

SES. Type 2 diabetes cases (n=542) were on average older, less likely to live in the

metropolitan area of Kumasi and were more frequently unemployed, illiterate and without

formal education compared to the controls (n=679). In addition, diabetes cases were more

often in the lowest group of the SES sum score compared to the controls. They had also

higher anthropometric measures as seen in higher values of waist and hip circumference,

WHR and WHtR. Furthermore, they had more often diabetes in the family, tended to

smoke more often, had a higher anti-inflammatory drug intake and higher energy

expenditure than controls. As expected, diabetes cases had a poor metabolic profile

compared to controls, as observed by higher concentrations of FPG, HOMA-IR, blood

lipids, CRP and systolic blood pressure and lower concentrations of adiponectin.

Of the 542 cases, 97% have already been known to the diabetes center or the

hypertension clinic (mean time since diagnosis, 6.5 ± 5.8 years). The majority was on

metformin-based medication (80%) and sulfonylureas (60%), in addition to glitazones and

insulin (both 23%).

RESULTS 38

Table 7: Descriptive characteristics of 1221 urban Ghanaian participants of the KDH study

Characteristics

Controls (n=679)

Diabetes cases (n=542)

Sex (female)

523 (77.0)

399 (73.6)

Socio-demographic data

Age (years)

46.8 ± 15.8

54.8 ± 13.4 *

Residence (Kumasi metropolitan)

517 (76.1)

379 (69.9) *

Ethnic group (Akan)

584 (86.0)

474 (87.5)

Formal education (none)

113 (16.6)

191 (35.2) *

Literacy (unable to read and write)

177 (26.1)

249 (45.9) *

Occupation (unemployed)

110 (16.2)

192 (35.4) *

SES sum score

Very low SES (0-4 points)

120 (17.7)

202 (37.3) *

Low SES (5-8 points)

360 (53.0)

280 (51.7)

Moderate SES (9-10 points)

199 (29.3)

60 (11.0)

Anthropometric data

BMI (kg/m²)

25.8 ± 5.4

25.8 ± 5.1

WC (cm)

86.5 ± 13.1

90.8 ± 12.0 *

HC (cm)

100.2 ±10.8

99.7 ± 10.7

WHR

0.86 ± 0.09

0.91 ± 0.07 *

WHtR

0.54 ± 0.08

0.56 ± 0.08 *

History and activity

Family history of diabetes

171 (25.2)

317 (58.5) *

Smoking (ever)

29 (4.3)

41 (7.6) *

Lipid lowering drug intake

13 (1.9)

16 (3.0)

Anti-inflammatory drug intake

6 (0.9)

18 (3.3) *

Energy expenditure (kcal/d)

1,210 (845-1,628)

1,408 (815-2,000) *

Clinical data

FPG (mmol/L)

4.5 (4.1-4.9)

6.9 (5.3-10.3) *

HOMA-IR

1.37 (0.85-2.13)

2.00 (1.17-3.40) *

Adiponectin (mg/ml)

8.63 (6.5-11.63)

7.42 (5.36-9.98) *

HDL-cholesterol (mmol/L)

1.37 (1.13-1.62)

1.27 (1.04-1.54) *

Triglycerides (mmol/L)

1.19 (0.87-1.64)

1.36 (1.02-1.87) *

CRP (mg/L)

1.21 (0.12-4.48)

1.57 (0.23-4.62) *

Systolic blood pressure (mmHg)

130.9 ± 21.8

137.9 ± 23.2 *

Diastolic blood pressure (mmHg)

84.4 ± 12.9

84.6 ± 11.9

Prevalent hypertension (≥ 140/90 mmHg)

344 (51.5)

332 (61.7) *

Values are presented as means ± standard deviation or median (interquartile range) for continuous variables

and participant number (%) for categorical variables. 1 SES sum score is based on education, occupation and

income, ranging from 0 to 12 points. *p-value <0.05 comparing diabetes cases and controls; BMI: body mass

index; WC: waist circumference (cm); HC: hip circumference (HC); WHR: waist-to-hip ratio; WHtR: waist-to-

height ratio; FPG: fasting plasma glucose; HOMA-IR: homeostatic model assessment for insulin resistance;

HDL-cholesterol: high-density lipoprotein; CRP: C-reactive protein

RESULTS 39

3.2 Anthropometry

3.2.1 Anthropometric characteristics

Table 8 shows the anthropometric characteristics among women and men of the KDH

study. The mean BMI among women (26.6 ± 5.4 kg/m²) and men (23.3 ± 3.8 kg/m²) did

not differ between diabetes cases and controls. WC, WHR and WHtR were all higher in

women with type 2 diabetes than controls. In contrast, only WHR was higher among men

with diabetes compared to controls. In general, women showed higher prevalences of

overweight (34.6%), general obesity (24.6%) and central obesity (56.4%) than men

(24.8%, 4.7% and 8.0%, respectively). Especially the prevalence of central obesity was

higher among women with type 2 diabetes compared to controls (65.7 vs. 49.3%).

Table 8: Anthropometric characteristics among women and men of the KDH study

Women (922)

Men (299)

Anthropometric data

Controls

(523)

Diabetes cases

(399)

Controls

(156)

Diabetes cases

(143)

BMI

26.5 ± 5.6

26.8 ± 5.1

23.5 ± 3.9

23.1 ± 3.7

Prevalent overweight

164 (31.4)

155 (38.9)

40 (25.6)

34 (23.8)

Prevalent obesity

131 (25.1)

93 (23.3)

7 (4.5)

7 (4.9)

87.0 ± 13.4

92.3 ± 12.0*

84.6 ± 11.9

86.6 ± 11.0

Prevalent central obesity

258 (49.3)

262 (65.7)*

13 (8.3)

11 (7.7)

101.6 ± 11

101.8 ± 10.7

95.5 ± 8.6

93.9 ± 8.2

WHR

0.85 ± 0.07

0.91 ± 0.07*

0.88 ± 0.08

0.92 ± 0.07*

WHtR

0.55 ± 0.08

0.58 ± 0.07*

0.50 ± 0.07

0.51 ± 0.06

Values are presented as means ± standard deviation for continuous variables and participant number (%) for

categorical variables. *p-value<0.05 comparing diabetes cases and controls; 1 BMI 25.0-29.9 kg/m²; 2 BMI ≥ 30

kg/m²; 3 waist circumference ≥ 102cm for men and ≥ 88cm for women; BMI: body mass index; WC: waist

circumference (cm); HC: hip circumference; WHR: waist-to-hip ratio; WHtR: waist-to-height ratio

Sex-specific partial correlation coefficients for different anthropometric measures,

controlling for age, are given in Table 9. BMI, WC, HC and WHtR were generally strongly

correlated in both genders (r ≥ 0.72). In contrast, WHR showed a comparatively weaker

correlation with the other anthropometric measures in both genders.

Table 9: Age-adjusted Spearman correlation coefficients for anthropometric measures

among women and men

BMI

WHR

WHtR

BMI

1.00

0.87

0.88

0.39

0.87

0.89

1.00

0.81

0.67

0.96

0.83

0.82

1.00

0.17

0.76

WHR

0.54

0.75

0.28

1.00

0.67

WHtR

0.89

0.94

0.72

0.76

1.00

Correlation coefficients for men are presented in the lower left side of the table (blue), while those for women

are in the upper right side (red) (all coefficients are significant at p<0.05); BMI: body mass index; WC: waist

circumference (cm); HC: hip circumference; WHR: waist-to-hip ratio; WHtR: waist-to-height ratio

RESULTS 40

3.2.2 Associations between anthropometric measures and type 2 diabetes

ORs and corresponding 95% CI for the association between different anthropometric

measures and type 2 diabetes across quintiles and per 1SD are shown in Table 10 for

women and in Table 11 for men.

In women, BMI and HC were not associated with type 2 diabetes. The OR for the highest

quintile compared to the lowest quintile was 1.06 [95% CI: 0.60-1.86], p for trend = 0.80

and 1.02 [0.59-1.75], p for trend = 0.86, respectively in the multivariate-adjusted model

(model 2a). In contrast, WC, WHR and WHtR were all positively associated with type 2

diabetes. Comparing the highest with the lowest quintile, the multivariate-adjusted ORs for

type 2 diabetes were 2.42 [1.31-4.47], p for trend <0.001 for WC, 4.58 [2.44-8.60], p for

trend <0.001 for WHR and 3.14 [1.64-6.00], p for trend = 0.002 for WHtR (model 2a). The

strength of association generally increased after adjustment for BMI. Similarly to the

quintile-based analysis, WHR showed the strongest association with type 2 diabetes per 1

SD difference (1.95 [1.60-2.39] followed by WC (1.43 [1.18-1.73]) and WHtR (1.36 [1.13-

1.63]) (model 2a). Further adjustment for BMI strengthened these associations (model 3).

In men, BMI and WC were not associated with type 2 diabetes. The multivariate-adjusted

OR for the highest quintile compared to the lowest quintile was 0.75 [95% CI: 0.32-1.80], p

for trend=0.69 for BMI and 1.01 [0.40-2.53], p for trend = 0.54 for WC (model 2a). In

contrast, HC were inversely and WHR positively associated with type 2 diabetes.

Comparing the highest with the lowest quintile, the multivariate-adjusted ORs for type 2

diabetes were 0.46 [0.19–1.10], p for trend = 0.02 for HC and 3.50 [1.44-8.49], p for trend

<0.001 for WHR (model 2a). BMI adjustment strengthened these associations (model 3).

Similarly to the quintile-based analysis, WHR showed a positive association with type 2

diabetes per 1 SD difference (1.85 [1.27-2.69]) and HC was inversely related (0.66 [0.44-

0.97] (model 2a). Further BMI adjustment strengthened these associations (model 3). In

men, WHtR was positively associated with type 2 diabetes in the multivariate-adjusted

model (model 2a) in quintile 3 and 4. After adjustment for BMI (model 3), the association

became significant across all quintiles (5th vs. 1st quintile: 6.35 [1.21-33.46]), p for trend =

0.002) and per 1 SD difference (3.10 [1.34-7.15]).

RESULTS 41

Table 10: Multivariate-adjusted ORs (95% CI) for type 2 diabetes by different anthropometric measures among 922 women

Quintile

p for trend1

for 1 SD

BMI (kg/m²)

Median

19.9

23.2

25.7

28.7

33.7

No. of cases/controls

65/105

71/104

95/104

94/105

74/105

Model 1

1.00

1.07 (0.68-1.68)

1.29 (0.84-2.00)

1.27 (0.82-1.96)

1.02 (0.66-1.60)

0.85

1.01 (0.88-1.15)

Model 2

1.00

1.14 (0.67-1.93)

1.34 (0.81-2.22)

1.21 (0.73-2.00)

0.80 (0.46-1.38)

0.38

0.92 (0.77-1.09)

Model 2a

1.00

1.14 (0.67-1.96)

1.57 (0.94-2.63)

1.44 (0.86-2.43)

1.06 (0.60-1.86)

0.80

1.01 (0.84-1.20)

WC (cm)

Median

69.5

79.8

87.6

94.0

104.3

No. of cases/controls

32/109

64/102

71/103

122/106

110/103

Model 1

1.00

1.68 (0.99-2.83)

1.63 (0.97-2.74)

2.65 (1.61-4.36)

2.41 (1.46-3.98)

<0.001

1.35 (1.17-1.57)

Model 2

1.00

1.69 (0.93-3.05)

1.60 (0.88-2.91)

2.56 (1.44-4.53)

2.18 (1.19-4.00)

0.006

1.33 (1.11-1.60)

Model 2a

1.00

1.52 (0.84-2.76)

1.58 (0.87-2.88)

2.77 (1.56-4.92)

2.42 (1.31-4.47)

<0.001

1.43 (1.18-1.73)

Model 3

1.00

2.03 (1.09-3.78)

2.81 (1.43-5.54)

6.08 (2.98-12.43)

8.63 (3.45-21.55)

<0.001

3.33 (2.30-4.83)

HC (cm)

Median

88.0

96.0

101.0

106.5

115.0

No. of cases/controls

72/105

76/105

94/106

71/103

86/104

Model 1

1.00

0.87 (0.56-1.36)

1.12 (0.73-1.72)

0.84 (0.54-1.31)

1.02 (0.66-1.57)

0.97

0.98 (0.85-1.12)

Model 2

1.00

0.79 (0.47-1.32)

1.06 (0.65-1.73)

0.74 (0.44-1.26)

0.79 (0.47-1.34)

0.39

0.88 (0.74-1.04)

Model 2a

1.00

0.84 (0.49-1.42)

1.21 (0.73-2.00)

0.91 (0.53-1.57)

1.02 (0.59-1.75)

0.86

0.95 (0.80-1.14)

Model 3

1.00

0.84 (0.49-1.47)

1.23 (0.69-2.20)

0.94 (0.48-1.85)

1.06 (0.46-2.47)

0.84

0.81 (0.57-1.15)

WHR

Median

0.76

0.81

0.86

0.90

0.95

No. of cases/controls

23/104

42/106

67/104

87/104

180/105

Model 1

1.00

1.40 (0.78-2.54)

2.15 (1.22-3.80)

2.60 (1.48-4.57)

5.01 (2.89-8.70)

<0.001

1.95 (1.64-2.31)

Model 2

1.00

1.33 (0.69-2.56)

2.37 (1.25-4.46)

3.03 (1.59-5.76)

5.06 (2.70-9.46)

<0.001

1.98 (1.63-2.42)

Model 2a

1.00

1.20 (0.62-2.34)

2.15 (1.13-4.06)

2.79 (1.46-5.32)

4.58 (2.44-8.60)

<0.001

1.95 (1.60-2.39)

Model 3

1.00

1.26 (0.65-2.46)

2.34 (1.23-4.47)

3.19 (1.64-6.19)

5.34 (2.78-10.26)

<0.001

2.10 (1.69-2.60)

RESULTS 42

Table 10 continued

Quintile

p for trend1

for 1 SD

WHtR

Median

0.44

0.50

0.55

0.59

0.65

No. of cases/controls

25/104

81/104

72/105

112/105

109/105

Model 1

1.00

2.44 (1.41-4.20)

1.88 (1.08-3.27)

2.87 (1.68-4.92)

2.64 (1.54-4.55)

0.002

1.29 (1.12-1.50)

Model 2

1.00

2.97 (1.61-5.49)

2.16 (1.15-4.07)

3.19 (1.73-5.88)

2.77 (1.46-5.25)

0.015

1.27 (1.06-1.51)

Model 2a

1.00

2.76 (1.49-5.13)

2.20 (1.16-4.17)

3.46 (1.87-6.43)

3.14 (1.64-6.00)

0.002

1.36 (1.13-1.63)

Model 3

1.00

3.80 (1.99-7.26)

3.91 (1.90-8.06)

7.74 (3.56-16.78)

10.50 (4.05-27.21)

<0.001

2.95 (2.07-4.28)

Model 1: adjusted for age; Model 2: adjusted for age (years), diabetes family history, educational attainment, smoking status, fiber intake (g/1000kcal), fat intake (g/1000kcal) and

energy expenditure (kcal/d); Model 2a: adjusted for age (years), diabetes family history, SES sum score, smoking status, fiber intake (g/1000kcal), fat intake (g/1000kcal) and

energy expenditure (kcal/d); Model 3: Model 2a + BMI, 1 p-value represents linear trend across quintiles; CI: confidence interval, OR: odds ratio; SD: standard deviation; BMI: body

mass index; WC: waist circumference (cm); HC: hip circumference; WHR: waist-to-hip ratio; WHtR: waist-to-height ratio

Table 11: Multivariate-adjusted ORs (95% CI) for type 2 diabetes by different anthropometric measures among 299 men

Quintile

Men (n=299)

p for trend1

for 1 SD

BMI (kg/m²)

Median

18.9

20.7

23.2

28.9

No. of cases/controls

61/63

30/30

32/32

20/31

Model 1

1.00

0.99 (0.52-1.89)

0.77 (0.41-1.47)

0.51 (0.25-1.01)

0.07

0.74 (0.52-1.04)

Model 2

1.00

1.23 (0.57-2.66)

0.79 (0.37-1.72)

0.64 (0.28-1.47)

0.37

0.88 (0.58-1.34)

Model 2a

1.00

1.28 (0.59-2.79)

0.89 (0.40-1.96)

0.75 (0.32-1.80)

0.69

0.97 (0.63-1.50)

WC (cm)

Median

70.9

77.0

82.5

92.3

100.0

No. of cases/controls

40/63

36/31

44/31

23/31

Model 1

1.00

1.29 (0.66-2.49)

1.25 (0.64-2.45)

0.66 (0.32-1.37)

0.55

0.92 (0.68-1.23)

Model 2

1.00

1.59 (0.73-3.47)

1.86 (0.82-4.21)

0.85 (0.35-2.09)

0.82

1.02 (0.71-1.47)

Model 2a

1.00

1.66 (0.75-3.63)

2.07 (0.90-4.76)

1.01 (0.40-2.53)

0.54

1.12 (0.76-1.62)

Model 3

1.00

1.81 (0.76-4.28)

2.49 (0.81-7.65)

1.37 (0.29-6.38)

0.22

1.76 (0.81-3.85)

RESULTS 43

Table 11 continued

Quintile

Men (n=299)

p for trend1

for 1 SD

HC (cm)

Median

84.2

91.0

95.7

100.0

105.5

No. of cases/controls

72/64

30/30

19/33

22/29

Model 1

1.00

0.78 (0.41-1.47)

0.44 (0.22-0.88)

0.51 (0.26-1.01)

0.01

0.69 (0.50-0.95)

Model 2

1.00

0.84 (0.38-1.82)

0.27 (0.12-0.64)

0.41 (0.17-0.98)

0.009

0.65 (0.44-0.97)

Model 2a

1.00

0.87 (0.40-1.89)

0.29 (0.12-0.69)

0.46 (0.19-1.10)

0.02

0.66 (0.44-0.97)

Model 3

1.00

0.59 (0.26-1.38)

0.15 (0.05-0.43)

0.14 (0.04-0.53)

0.001

0.35 (0.17-0.70)

WHR

Median

0.79

0.84

0.88

0.92

0.97

No. of cases/controls

29/63

23/30

37/31

54/32

Model 1

1.00

1.22 (0.59-2.55)

1.75 (0.87-3.53)

2.01 (0.97-4.16)

0.015

1.40 (1.01-1.94)

Model 2

1.00

1.36 (0.56-3.29)

2.21 (0.96-5.13)

3.12 (1.31-7.38)

0.002

1.71 (1.19-2.46)

Model 2a

1.00

1.29 (0.53-3.18)

2.24 (0.96-5.20)

3.50 (1.44-8.49)

<0.001

1.85 (1.27-2.69)

Model 3

1.00

1.40 (0.56-3.47)

2.95 (1.21-7.19)

5.72 (2.06-15.94)

<0.001

2.21 (1.44-3.40)

WHtR

Median

0.41

0.45

0.49

0.55

0.60

No. of cases/controls

33/63

40/31

43/30

27/32

Model 1

1.00

1.81 (0.93-3.51)

1.66 (0.84-3.28)

0.87 (0.42-1.81)

0.97

0.92 (0.67-1.25)

Model 2

1.00

3.20 (1.41-7.26)

3.11 (1.35-7.16)

1.50 (0.61-3.69)

0.20

1.14 (0.78-1.65)

Model 2a

1.00

3.42 (1.49-7.82)

3.58 (1.52-8.42)

1.83 (0.72-4.64)

0.09

1.25 (0.84-1.84)

Model 3

1.00

4.85 (1.93-12.16)

7.29 (2.26-23.56)

6.35 (1.21-33.46)

0.002

3.10 (1.34-7.15)

Model 1: adjusted for age; Model 2: adjusted for age (years), diabetes family history, educational attainment, smoking status, fiber intake (g/1000kcal), fat intake (g/1000kcal) and

energy expenditure (kcal/d); Model 2a: adjusted for age (years), diabetes family history, SES sum score, smoking status, fiber intake (g/1000kcal), fat intake (g/1000kcal) and

energy expenditure (kcal/d); Model 3: Model 2a + BMI, 1 p-value represents linear trend across quintiles; CI: confidence interval, OR: odds ratio; SD: standard deviation; BMI: body

mass index; WC: waist circumference (cm); HC: hip circumference; WHR: waist-to-hip ratio; WHtR: waist-to-height ratio

RESULTS 44

Sensitivity analyses

Sensitivity analysis was performed by evaluating whether additional adjustment for blood

pressure and P. falciparum infection confounded the associations between anthropometric

measures and type 2 diabetes (Table 12). Further adjustment slightly strengthened the

associations between the anthropometric measures and type 2 diabetes. Nevertheless,

BMI was not associated with diabetes in both sexes.

Table 12: Multivariate-adjusted1 ORs (95% CI) for type 2 diabetes per 1SD by different

anthropometric measures among women and men

OR (95%CI) for 1 SD

Women (n=922)

Men (n=299)

BMI (kg/m²)

1.16 (0.98-1.36)

1.08 (0.73-1.62)

WC (cm)

1.63 (1.37-1.94)

1.24 (0.87-1.77)

HC (cm)

0.90 (0.76-1.08)

0.65 (0.44-0.97)

WHR

2.21 (1.81-2.71)

1.92 (1.33-2.78)

WHtR

1.54 (1.30-1.83)

1.35 (0.94-1.95)

1adjusted for model 2a: age (years), diabetes family history, smoking status, SES sum score, fiber intake

(g/1000kcal), fat intake (g/1000kcal), energy expenditure (kcal/d), systolic and diastolic blood pressure and P.

falciparum; CI: confidence interval, OR: odds ratio; SD: standard deviation; BMI: body mass index; WC: waist

circumference (cm); HC: hip circumference; WHR: waist-to-hip ratio; WHtR: waist-to-height ratio

As BMI was not associated with type 2 diabetes, a hypothesis was that diabetic atrophy in

cases with a poor glycaemic control (medication + FPG ≥ 7 mmol/L) may be responsible.

The prevalence of a poor glycaemic control in diabetes cases was 44% among women

and 50% among men. Among women, mean BMI (± SD) was significantly lower in

diabetes cases with a poor glycaemic control (26.1 ± 5.4 kg/m2) as compared to diabetes

cases with a good glycaemic control (27.3 ± 4.8 kg/m2; p = 0.01). In men, these figures

were 22.5 ± 3.9 kg/m2 and 23.7 ± 3.4 kg/m2, respectively (p = 0.04). The same pattern

was observed for mean body cell mass, assessed by BIA, in women (20.1 ± 3.3 vs. 21.0 ±

3.4 kg; p = 0.02) and in men (25.0 ± 5.4 vs. 27.5 ± 5.2 kg; p = 0.005). After exclusion of

diabetes cases with a poor glycaemic control, the association between BMI and type 2

diabetes strengthened among women and men. The OR per 1SD in the multivariate-

adjusted model (model 2a) was 1.10 [0.89-1.36] among women and 1.05 [0.61-1.79]

among men. However, the lack of association between BMI and type 2 diabetes persisted.

This was similar when using FPG ≥ 7.8 mmol/l as the cut-off value (women: 1.12 [0.91-

1.37], men: 1.13 [0.67-1.90]).

To test whether the association between BMI and type 2 diabetes was modified by SES,

interaction analyses were performed. The mean BMI increased with the SES sum score

among women (very low SES: 25.1 ± 4.5 kg/m², low SES: 27.3 ± 5.5 kg/m² and moderate

RESULTS 45

SES: 27.2 ± 5.7 kg/m²) and men (very low SES: 21.4 ± 3.4 kg/m², low SES: 23.4 ± 3.7

kg/m² and moderate SES: 24.7 ± 3.7 kg/m²). Nevertheless, the association between BMI

and type 2 diabetes was not modified by SES in both sexes (Table 13).

Table 13: Effect modification of the association between BMI and type 2 diabetes by

socioeconomic status

OR (95% CI) for 1 SD

Women

p for

interaction

Men

p for

interaction

SES sum score1

Very low SES

269

0.95 [0.66-1.36]

0.74

2.78 [0.65-11.94]

0.62

Low SES

453

1.01 [0.81-1.27]

187

0.71 [0.41-1.25]

Moderate SES

200

0.70 [0.41-1.21]

1.11 [0.39-3.13]

1 adjusted for age (years), diabetes family history, smoking status, fiber intake (g/1000kcal), fat intake

(g/1000kcal) and energy expenditure (kcal/d); CI: confidence interval, OR: odds ratio; SD: standard deviation;

SES: socioeconomic status

3.2.3 Discrimination of type 2 diabetes cases and controls

In a next step the discriminative power of selected anthropometric measures for

identifying type 2 diabetes by ROC-AUC comparisons was assessed among women and

men (Figure 4). For this analysis, only the anthropometric measures were used that were

associated with type 2 diabetes in logistic regression (model 2a). Thus, WC, WHR and

WHtR were used among women and HC and WHR among men.

For the ROC-AUC analysis the multivariate-adjusted model (model 2a) was applied,

including age, diabetes family history, SES sum score, smoking status, fiber intake, fat

intake and energy expenditure. In women, WHR (ROC-AUC: 0.811 [95% CI: 0.783-

0.840]) was the best obesity measure for discriminating type 2 diabetes, followed by WC

(0.794 [0.764-0.824], p = 0.004) and WHtR (0.792 [0.762-0.822], p = 0.002). In men, HC

and WHR had similar discriminative power: HC (0.806 [0.755-0.857]) and WHR (0.818

[0.769-0.868]) p = 0.29).

RESULTS 46

Figure 4: Receiver operating characteristic (ROC) curves of various anthropometric

measures for discriminating type 2 diabetes cases and controls among women and men

For the ROC-AUC analysis the multivariate-adjusted model (model 2a) was applied, including age,

diabetes family history, SES sum score, smoking status, fiber intake, fat intake and energy

expenditure. Among women, WHR showed the highest areas under the ROC curve (ROC-AUC)=

0.811 [95% CI: 0.783-0.840], followed by WC=0.794 [0.764-0.824] and WHtR 0.792 [0.762-0.822]

to predict the risk of type 2 diabetes. Among men, the ROC-AUCs for identifying diabetes cases

were 0.806 [0.755-0.857] for HC and 0.818 [0.769-0.868] for WHR.

RESULTS 47

3.2.4 Examination of cut-offs for obesity measures

Finally, the sensitivity and specificity of recommended cut-off points for the identification of

type 2 diabetes were calculated and the optimal cut-offs for BMI, WC and WHR with the

Youden Index assessed (Table 14). Using BMI ≥ 25 kg/m2, 62% of the cases and 44% of

the controls were correctly classified in women. Sensitivity was considerably lower among

men (29%), at higher specificity (70%). For BMI ≥ 30 kg/m2, only 23% of the cases, but

75.0% of the controls were correctly classified in women. In men, this cut-off showed a

low sensitivity (5%), at high specificity (96%). Recommended WC cut-off points identified

type 2 diabetes well in women (high sensitivity), but not in men. The sensitivity and

specificity for WHR were > 60% and identified diabetes cases and controls well in both

sexes.

Compared to the recommended cut-off points, the optimal BMI cut-offs were slightly

higher in both women (26.2 vs. 25.0/30.0) and men (26.7 vs. 25.0/30.0). The optimal cut-

off point for WC was higher than the recommended cut-offs for both, overweight and

obesity, in women (91.7 vs. 80.0/88.0 cm) and lower in men (83.4 vs. 94.0/102.0 cm). For

WHR, the population-specific cut-off point in women (0.88) exceeded the reference value

(0.85). These values were identical for men (0.90).

Table 14: Sensitivity and specificity of diabetes cases identified using sex-specific cut-off-

points and Youden index for BMI, WC and WHR

Women (n=922)

Men (n=299)

Variable

Sensitivity

Specificity

Sensitivity

Specificity

BMI (kg/m²)

≥ 25.0

62.2

43.6

28.7

69.9

≥ 30.0

23.3

75.0

4.9

95.5

optimal cut-off:

26.2

54.6

53.9

26.7

86.0

21.8

WC (cm)

≥ 80.0

86.0

30.6

≥ 88.0

65.7

50.7

optimal cut-off:

91.7

54.1

64.4

83.4

60.8

54.5

≥ 94.0

25.9

73.1

≥ 102.0

7.7

91.7

WHR

≥ 0.85

78.7

46.7

≥ 0.90

65.0

59.6

optimal cut-off:

0.88

65.7

62.0

0.90

63.6

61.5

BMI: body mass Index; WC: waist circumference (cm); WHR: waist-to-hip ratio; 1 optimal cut-offs were

identified by using the Youden index, which is derived from maximum (sensitivity + specificity – 1)

RESULTS 48

Sensitivity analyses

To investigate whether diabetes cases with a poor glycaemic control (medication + FPG ≥

7 mmol/L) have influenced these results, those cases were excluded in a sensitivity

analysis (Table 15). The results were comparable; however, the optimal cut-off point for

BMI changed markedly, decreasing from 26.2 to 25.6 kg/m2 among women and from 26.7

to 20.7 kg/m2 among men. In addition, the optimal cut-off point for WHR decreased from

0.90 to 0.88 among men.

Table 15: Calculation of sensitivity and specificity of diabetes cases identified using sex-

specific cut-off-points and Youden index for BMI, WC and WHR in participants with a good

glycaemic control (FPG < 7mmol/L)

Women (n=732)

Men (n=225)

Variable

Sensitivity

Specificity

Sensitivity

Specificity

BMI (kg/m²)

≥ 25.0

66.5

43.6

31.9

69.9

≥ 30.0

24.4

75.0

5.8

95.5

optimal cut-off:

25.6

65.1

48.8

20.7

85.5

30.8

WC (cm)

≥ 80.0

90.4

30.6

≥ 88.0

71.3

50.7

optimal cut-off:

91.7

59.3

64.4

83.6

71.0

55.1

≥ 94.0

26.1

73.1

≥ 102.0

8.7

91.7

WHR

≥ 0.85

81.8

46.7

≥ 0.90

69.6

59.6

optimal cut-off:

0.88

71.3

60.8

0.88

81.2

50.0

BMI: body mass index; WC: waist circumference (cm); WHR: waist-to-hip ratio; 1 optimal cut-offs were

identified by using the Youden index, which is derived from maximum (sensitivity + specificity – 1)

RESULTS 49

3.3 Nutrition

3.3.1 Intake of energy, macronutrients and food groups

The median daily energy intake of the study population was 1,966 kcal/d (interquartile

range: 1,531-2,515 kcal/d). The dietary energy supply by macronutrients is shown in

Figure 5. Carbohydrates provided the highest contribution to the total energy intake

(53 ± 13%), followed by fat (28 ± 11%) and protein (19 ± 8%).

Figure 5: Dietary energy supply by macronutrients of the total study population

The median energy and macronutrient intakes of controls and diabetes cases are

presented in Table 16. The median energy intake was slightly higher in the controls

compared to the cases. As for the macronutrients, the intake in the control group was

higher for carbohydrates and lower for total fat and protein compared with the diabetes

group. Dietary fiber intake was similar between the groups. Almost one-quarter of the

study population consumed two meals per day and 76% took three daily meals.

Table 16: Median energy and macronutrient intake and number of meals per day in the study

population

Controls (n=679)

Diabetes cases (n=542)

Energy intake (kcal)

2,014 (1,576-2,673)

1,893 (1,488-2,380) *

Protein (g/1000kcal)

38 (32-48)

43 (35-55) *

Fat (g/1000kcal)

27 (20-36)

29 (20-39) *

Carbohydrate (g/1000kcal)

130 (109-150)

117 (95-141) *

Fibre (g/d)

15 (11-19)

16 (12-20)

Count of meals

5 (0.7)

1 (0.2) *

170 (25.0)

75 (13.9)

494 (72.8)

451 (83.5)

10 (1.5)

12 (2.2)

1 (0.2)

Values are presented as median (interquartile range) for continuous variables and participant number (%) for

categorical variables. Differences between diabetes cases and controls were compared by Mann-Whitney-U-

test for continuous variables and by χ²-test for categorical variables (*p-value <0.05)

53%

28%

19%

Carbohydrates Fat Protein

RESULTS 50

With respect to the liquids of the FFQ all participants consumed water on a daily basis, but

the majority of the participants never consumed alcoholic beverages (92%) or coffee

(85%). In general, the intake of juice, chocolate drink (milo) and soft drinks was not

pronounced among the diabetes cases and the intake of those food items were low

(median intake: 0.5-1.5 servings/week) among the controls. The milk intake was slightly

higher among controls compared to diabetes cases (median intake: 1.5 vs 0.5

servings/week). In contrast, all participants consumed salt on a daily basis. Fish was

consumed on a daily basis among both groups. Banku, rice and palm oil (3.5

servings/week) and cassava, yam, millet, and legumes, nuts and oilseeds (1,5

servings/week) were also equally consumed (Figure 6). The consumption of vegetable oil,

fruits, red meat and sweets was higher among the controls compared to the diabetes

cases; however the intake of those food items was low (median intake: 0.5-3.5

servings/week). In contrast, the consumption of plantain, bread, and vegetables was

higher among the type 2 diabetes cases.

Figure 6: Food intake (servings/week) among controls and type 2 diabetes cases (FFQ)

01234567

Porridge

Sweets

Cocoyam

Eggs

Poultry

Red meat

Legumes, nuts & oilseeds

Millet

Yam

Cassava

Fruits

Vegetable oil

Palmoil

Rice

Banku

Vegetables

Bread

Plantain

Fish

Servings/week

Type 2 diabetes cases Controls

RESULTS 51

3.3.2 Dietary patterns derived by factor analysis

Initially factor analysis of the 33 food items was applied in sex-stratified analysis. The

scree plots of all factors with eigenvalues ≥ 1.0 (ten factors) were nearly identical between

women and men (Figure S3, appendix). In a two-factor solution, the resulting rotated

factor loadings of highly loading food items (rotated factor loadings ≥ 0.35) were similar

between men and women (Table S2, appendix). The first factor was characterized by high

intakes of sweets and sweet drinks, rice, red meat, poultry, eggs, milk, vegetable oil,

margarine, fruits and vegetables, and low intake of plantain in both sexes. The second

factor was characterized by high intakes of plantain, green leafy vegetables, beans,

garden egg, fish, fermented maize products (banku), palm oil, okra and fruits (Table S2).

Also, dietary patterns were calculated separately in the control group and in the total study

population and these results were nearly identical (Figure S4 and Table S3, appendix).

Hence, dietary patterns were derived among the total study population.

The scree plot of all factors with eigenvalues ≥ 1.0 among the total study population is

shown in Figure 7. Ten factors exhibited an eigenvalue ≥ 1.0. However, the scree plot

showed no clear differences between the fourth to the tenth factor.

Figure 7: Scree plot of eigenvalues ≥ 1.0 among the total study population

0,0

0,5

1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

5,0

012345678910 11

Eigenvalue

Factor Number

RESULTS 52

Based on the criteria to detect the optimal number of factors two dietary patterns were

identified that explained 22.2% of the variation in the food items. The first dietary pattern

explained 13.7% of the variance among food items and the second 8.5%. The rotated

factor loadings of the two identified dietary pattern with the 33 food items are presented in

Table S4 (appendix). The two dietary patterns with factor loadings ≥ 0.35 are illustrated in

Figure 8. The first dietary pattern was labeled “purchase” dietary pattern and was

characterized by high intakes of sweets and sweet drinks, rice, foods rich in protein (red

meat, poultry, eggs and milk), plant oils (vegetable oil and margarine), fruits and

vegetables (carrot, lettuce and cucumber), and low intake of plantain. The second dietary

pattern was labeled “traditional” dietary pattern and was positively correlated with the

intake of traditional food items as plantain, green leafy vegetables, beans, garden egg,

fish, maize (banku), palm oil, okra and fruits.

Figure 8: Spider graph of the two dietary patterns identified by factor analysis among the

total study population

-0,8

-0,6

-0,4

-0,2

0,2

0,4

0,6

0,8

Juice

Sweets

Rice

Soft drinks

Vegetable oil

Chocolate drink

Red meat

Eggs

Margarine

Fruits

Carrot

LettuceMilk

Poultry

Cucumber

Plantain

Green leaves

Beans

Gardenegg

Smoked fish

Banku

Palm oil

Okro

"Purchase" dietary pattern "Traditional" dietary pattern

RESULTS 53

The characteristics across quintiles of the “purchase” dietary pattern among the 679

controls are shown in Table 17, and those of the “traditional” dietary pattern in Table 18.

Participants in the highest quintile of the “purchase” dietary pattern were on average

younger, leaner and exhibited lower energy expenditure than those in the lower quintiles.

The frequency of participants with a very low SES sum score decreased across the

quintiles of the pattern score. In addition, the mean intakes in the highest quintile of the

“purchase” dietary pattern score were low for sweets, eggs, margarine, vegetables,

poultry and plantain (1.5 servings/week), moderate for sweet drinks, red meat, fruits and

milk (3.5 servings/week) and high for rice (7.0 servings/week) compared to those in the

lower quintiles (Table 17).

As for the “traditional” dietary pattern, participants in the highest quintile were

characterized by higher age, increased measures of obesity and higher energy

expenditure when compared with those in the lower quintiles (Table 18). The frequency of

participants with a very low SES sum score increased across the quintiles of the pattern

score. In addition, the mean intakes in the highest quintile of the “traditional” dietary

pattern score were low for juice, soft drinks, sweets, red meat, poultry, margarine, and

milk (0.5-1.5 servings/week), moderate for fruits, beans and okra (3.5 servings/week), and

high for plantain, green leafy vegetables, garden egg, fish, banku, and palm oil (5.5-7.0

servings/week).

RESULTS 54

Table 17: Characteristics by quintiles of the “purchase” dietary pattern among 679 controls of the KDH study

Quintile of the "purchase" dietary pattern

Characteristics

135

136

Sex (female)

106 (79.0)

107 (79.0)

103 (76.0)

118 (87.0)

89 (65.0)

Age (years)

57.9 ± 12.5

52.6 ± 13.8

47.5 ± 13.3

43.3 ± 13.8

32.9 ± 12.9

BMI (kg/m²)

25.7 ± 5.7

25.8 ± 4.7

27.0 ± 6.1

26.4 ± 5.0

24.1 ± 4.9

WHR

0.88 ± 0.08

0.88 ± 0.06

0.87 ± 0.07

0.86 ± 0.07

0.82 ± 0.07

Family history of diabetes

38 (28.1)

29 (21.3)

33 (24.3)

32 (23.5)

39 (28.7)

Smoking (ever)

8 (5.9)

6 (4.4)

5 (3.7)

3 (2.2)

7 (5.1)

SES sum score (very low SES)

52 (38.5)

32 (23.5)

19 (14.0)

11 (8.1)

6 (4.4)

Energy expenditure (kcal/d)

1,447 (965-2,027)

1,384 (962-1,856)

1,423 (968-1,931)

1,226 (981-1,618)

1,211 (961-1,645)

Food intake (servings/week)

Juice

0 (0-0)

0 (0-0.5)

0.5 (0-1.5)

1.0 (0-1.5)

3.5 (1.0-3.5)

Sweets

0 (0- 0.5)

0 (0-0.5)

0.5 (0-0.5)

0.5 (0.3-0.5)

1.5 (0.5-1.5)

Rice

1.5 (0.5-3.5)

3.5 (1.5-3.5)

4.5 (3.5-7.0)

5.5 (3.5-7.0)

7.0 (5.5-7.0)

Soft drinks

0.5 (0-0.5)

0.5 (0.5-1.5)

1.5 (0.5-1.5)

1.5 (0.5-3.5)

3.5 (1.5-3.5)

Vegetable oil

0.5 (0.5-1.5)

1.5 (1.5-3.5)

3.5 (1.5-3.5)

3.5 (1.5-5.5)

5.5 (3.5-7.0)

Milo

0.5 (0-1.5)

1.5 (0-1.5)

1.5 (0.5-2.5)

1.5 (0.5-3.5)

3.5 (1.5-5.5)

Red meat

0.5 (0-1.5)

0.5 (0.5-1.5)

1.5 (0.5-3.5)

3.5 (1.5-7.0)

Eggs

0.5 (0-0.5)

0.5 (0.5-1.5)

1.5 (0.5-1.5)

1.5 (0.5-3.5)

Margarine

0 (0-0)

0 (0-0.5)

0.5 (0-0.5)

1.5 (0-3.5)

Fruits

1.5 (0.5-3.5)

1.5 (1.5-3.5)

3.5 (1.5-3.5)

3.5 (3.5-3.5)

Carrot

0.5 (0-0.5)

0.5 (0-1.5)

0.5 (0.5-1.5)

1.5 (0.5-3.5)

Lettuce

0 (0-0.5)

0.5 (0-1.5)

0.5 (0.5-1.5)

1.5 (0.5-3.5)

Milk

0.5 (0-0.5)

0.5 (0-1.5)

0.5 (0.5-3.5)

3.5 (0.5-7.0)

3.5 (1.5-7.0)

Poultry

0.5 (0.5-0.5)

0.5 (0.5-1.5)

1.5 (0.5-3.5)

Cucumber

0 (0-0)

0 (0-0.5)

0.5 (0-0.5)

0.5 (0-1.5)

Values are expressed as mean ± standard deviation, participant number (%) or median (interquartile range)

RESULTS 55

Table 17 continued

Quintile of the "purchase" dietary pattern

Food intake (servings/week)

Plantain

7.0 (3.5-7.0)

3.5 (1.5-7.0)

3.5 (1.5-5.5)

1.5 (0.5-3.5)

Green leaves

1.5 (0.5-3.5)

3.5 (1.5-4.5)

1.5 (1.5-3.5)

3.5 (1.5-5.5)

3.5 (1.5-7.0)

Beans

0.5 (0.5-1.5)

1.5 (0.5-3.5)

1.5 (1.5-3.5)

Garden egg

7.0 (5.5-7.0)

7.0 (3.5-7.0)

7.0 (1.5-7.0)

7.0 (3.5-7.0)

Fish

7.0 (7.0-7.0)

7.0 (3.5-7.0)

Banku

3.5 (1.5-7.0)

Palm oil

3.5 (1.5-5.5)

3.5 (1.5-3.5)

3.5 (1.5-5.5)

3.5 (1.5-7.0)

Okro

0.5 (0-1.5)

1.5 (0.5-3.5)

1.5 (0.5-7.0)

Values are expressed as median (interquartile range)

Table 18: Characteristics by quintiles of the “traditional” dietary pattern among 679 controls of the KDH study

Quintile of the "traditional" dietary pattern

Characteristics

135

136

Sex (female)

113 (83.7)

100 (73.5)

109 (80.1)

104 (76.5)

97 (71.9)

Age (years)

42.0 ± 16.5

46.0 ± 15.8

45.3 ± 15.0

49.1 ± 15.6

51.9 ± 14.3

BMI (kg/m²)

25.5 ± 5.5

25.2 ± 5.2

26.1 ± 5.3

26.3 ± 5.4

26.1 ± 5.4

WHR

0.84 ± 0.08

0.86 ± 0.09

0.86 ± 0.07

0.88 ± 0.06

Family history of diabetes

40 (29.6)

31 (22.8)

34 (25.0)

36 (26.5)

30 (22.2)

Smoking (ever)

2 (1.5)

6 (4.4)

8 (5.9)

5 (3.7)

8 (5.9)

SES sum score (very low SES)

15 (11.1)

24 (17.7)

19 (14.0)

32 (23.5)

30 (22.2)

Energy expenditure (kcal/d)

1,231 (861-1,720)

1,448 (1,067-1,883)

1,348 (1,005-1,800)

1,339 (966-1,846)

1,408 (962-2,009)

Values are expressed as mean ± standard deviation, participant number (%) or median (interquartile range)

RESULTS 56

Table 18 continued

Quintile of the "traditional" dietary pattern

Food intake (servings/week)

Juice

0.5 (0-1.5)

0 (0-1.5)

0.5 (0-1.5)

Sweets

0.5 (0-0.5)

Rice

3.5 (1.5-7.0)

5.5 (3.5-7)

Soft drinks

1.5 (0.5-1.5)

0.5 (0.5-1.5)

1.5 (0.5-3.5)

0.5 (0.5-1.5)

1.5 (0.5-3.5)

Vegetable oil

3.5 (1.5-5.5)

3.5 (1.5-3.5)

1.5 (0.5-3.5)

3.5 (1.5-3.5)

Milo

1.5 (0.5-3.5)

1.5 (0-3.5)

1.5 (0.5-3.5)

Red meat

1.5 (0.5-3.5)

Eggs

0.5 (0.5-1.5)

Margarine

0 (0-0.5)

0.5 (0-1.5)

Fruits

1.5 (1.5-3.5)

3.5 (1.5-3.5)

3.5 (3.5-3.5)

Carrot

0.5 (0-1.5)

1.5 (0.5-1.5)

0.5 (0.5-1.5)

1.5 (0.5-3.5)

Lettuce

0.5 (0-0.5)

0.5 (0-1.5)

1.5 (0.5-1.5)

Milk

1.5 (0.5-3.5)

1 (0.5-3.5)

1.5 (0.5-3.5)

0.5 (0.5-3.5)

1.5 (0.5-5.5)

Poultry

0.5 (0.5-1.5)

Cucumber

0.5 (0-0.5)

0 (0-0.5)

0.5 (0-1.5)

Plantain

1.5 (0.5-3.5)

3.5 (1.5-5.5)

3.5 (1.5-7.0)

3.5 (3.5-7)

7.0 (3.5-7.0)

Green leaves

1.5 (0.5-1.5)

1.5 (1.5-3.5)

1.5 (0.5-3.5)

3.5 (1.5-5.5)

7.0 (3.5-7.0)

Beans

1.5 (0.5-1.5)

1.5 (0.5-3.5)

3.5 (1.5-7.0)

Gardenegg

1.5 (1.5-3.5)

3.5 (3.5-7.0)

7.0 (3.5-7.0)

7.0 (7.0-7.0 )

7.0 (7.0-7.0)

Fish

3.5 (3.5-7.0)

7.0 (4.5-7.0)

7.0 (7.0-7.0)

7.0 (7.0-7.0 )

7.0 (7.0-7.0)

Banku

1.5 (1.5-3.5)

3.5 (1.5-7.0)

7.0 (3.5-7.0)

Palm oil

1.5 (1.5-3.5)

3.5 (1.5-3.5)

3.5 (3.5-7.0)

5.5 (3.5-7.0)

Okro

0.5 (0.5-1.5)

1.5 (0.5-3.5)

3.5 (1.5-7.0)

Values are expressed as median (interquartile range)

RESULTS 57

3.3.3 Associations between dietary patterns and type 2 diabetes

ORs and corresponding 95% CI for the associations between the two identified dietary

patterns and type 2 diabetes across quintiles and per 1SD are shown in Table 19. The

“purchase” dietary pattern was inversely associated with type 2 diabetes (highest quintile

compared to the lowest quintile: 0.10 [95% CI 0.06-0.17], p for trend <0.001) in the age-

and sex-adjusted model. This association was similar after further adjustment for family

history of diabetes, SES sum score, smoking, daily energy expenditure, BMI and WHR in

the fully adjusted model (model 3). In contrast, the “traditional” dietary pattern was

positively associated with type 2 diabetes (highest quintile compared to the lowest quintile:

3.14 [95% CI 2.09-4.73], p for trend <0.001) in the age- and sex-adjusted models. This

positive association was slightly attenuated after further adjustments (highest quintile

compared to the lowest quintile: 2.98 [95% CI 1.83-4.85], p for trend <0.001, model 3).

Similarly to the quintile-based analysis, the “purchase” dietary pattern decreased the odds

of type 2 diabetes per 1 SD difference of the pattern score by 57% (47-65%), whereas the

“traditional” dietary pattern increased the odds of type 2 diabetes by 53% (32-78%) in the

fully adjusted model (model 3).

RESULTS 58

Table 19: Multivariate-adjusted ORs (95% CI) for type 2 diabetes per quintiles and per 1 SD of dietary pattern scores

Quintile

p for trend

OR per 1 SD

"Purchase" dietary pattern

Median

-0.91

-0.29

0.21

0.71

1.61

No. of cases/controls

272/135

140/136

69/136

40/136

21/136

Model 1

1.00

0.56 (0.40-0.76)

0.29 (0.20-0.41)

0.18 (0.12-0.27)

0.10 (0.06-0.17)

<0.001

0.37 (0.31-0.44)

Model 2

1.00

0.70 (0.49-1.01)

0.31 (0.20-0.47)

0.22 (0.13-0.35)

0.10 (0.06-0.19)

<0.001

0.39 (0.32-0.48)

Model 2a

1.00

0.73 (0.50-1.05)

0.34 (0.22-0.51)

0.24 (0.15-0.49)

0.12 (0.06-0.21)

<0.001

0.42 (0.34-0.51)

Model 3

1.00

0.69 (0.48-1.01)

0.36 (0.23-0.55)

0.24 (0.15-0.41)

0.13 (0.07-0.23)

<0.001

0.43 (0.35-0.53)

"Traditional" dietary pattern

Median

-1.41

-0.77

-0.26

0.31

1.12

No. of cases/controls

48/135

64/136

118/136

127/136

185/136

Model 1

1.00

1.19 (0.75-1.88)

2.17 (1.42-3.31)

2.21 (1.45-3.37)

3.14 (2.09-4.73)

<0.001

1.52 (1.34-1.72)

Model 2

1.00

1.21 (0.71-2.06)

2.52 (1.53-4.14)

2.79 (1.70-4.56)

3.63 (2.25-5.85)

<0.001

1.60 (1.38-1.84)

Model 2a

1.00

1.13 (0.66-1.91)

2.36 (1.44-3.87)

2.58 (1.58-4.22)

3.36 (2.09-5.42)

<0.001

1.57 (1.36-1.81)

Model 3

1.00

1.07 (0.62-1.85)

2.06 (1.23-3.44)

2.40 (1.45-3.98)

2.98 (1.83-4.85)

<0.001

1.53 (1.32-1.78)

Model 1: adjusted for age and sex; Model 2: adjusted for age (years), sex, diabetes family history, educational attainment, literacy, unemployment, smoking status and energy

expenditure (kcal/d); Model 2a: adjusted for age (years), sex, diabetes family history, SES sum score, smoking status and energy expenditure (kcal/d); Model 3: Model 2a + BMI

and WHR; 1 p-value represents linear trend across quintiles; CI: confidence interval, SD: standard deviation, OR: odds ratio

RESULTS 59

Sensitivity analyses

With respect to the nutrition transition in SSA, this study found unexpected associations of

the “purchase” and the “traditional” dietary pattern with type 2 diabetes. Hence, sensitivity

analyses were performed to test whether the associations between the dietary patterns

and type 2 diabetes were modified by age, sex, BMI, central obesity and SES (Table 20).

Only age and the SES sum score showed a significant interaction with the “purchase”

dietary pattern: The inverse association was stronger in participants who were younger

(age < 51 years) compared to their older counterparts (age ≥ 51 years). This was also

observed in participants with moderate SES compared to participants with very low and

low SES. With respect to the association between the traditional dietary pattern and type 2

diabetes, no significant effect modification by age, sex, BMI and central obesity and SES

were found.

Table 20: Effect modification of the associations between dietary patterns and type 2

diabetes by age, sex, BMI, central obesity and SES sum score

ORs (95% CI) for 1 SD

„purchase“ dietary

pattern

p for

interaction

„traditional“

dietary pattern

p for

interaction

Age1

< 51 years

585

0.35 (0.26-0.48)

0.001

1.59 (1.28-1.97)

0.37

≥ 51 years

636

0.55 (0.41-0.73)

1.51 (1.23-1.87)

Sex2

Men

299

0.44 (0.30-0.64)

0.77

1.44 (1.09-1.91)

0.60

Women

922

0.41 (0.32-0.53)

1.59 (1.33-1.89)

BMI3

< 25 kg/m²

590

0.33 (0.23-0.46)

0.75

1.66 (1.32-2.10)

0.21

25.0-29.9 kg/m²

393

0.58 (0.42-0.80)

1.43 (1.19-1.82)

≥ 30 kg/m²

238

0.40 (0.24-0.65)

1.47 (1.04-2.07)

Central obesity4

<88/<102 cm5

677

0.41 (0.31-0.54)

0.28

1.58 (1.29-1.93)

0.20

≥88/≥102 cm

544

0.43 (0.31-0.60)

1.49 (1.19-1.87)

SES sum score6

Very low SES

322

0.49 (0.31-0.77)

0.005

1.19 (0.87-1.63)

0.14

Low SES

640

0.44 (0.33-0.58)

1.64 (1.34-2.00)

Moderate SES

259

0.29 (0.17-0.50)

1.63 (1.12-2.38)

1 subgroups were created by taking the median of age; adjusted for sex, diabetes family history, SES sum

score smoking status, energy expenditure (kcal/d), BMI and WHR

2 adjusted for age (years), diabetes family history, SES sum score, smoking status, energy expenditure

(kcal/d), BMI and WHR

3 adjusted for age (years), sex, diabetes family history, SES sum score, smoking status, energy expenditure

(kcal/d), BMI and WHR

4 adjusted for age (years), sex, diabetes family history, SES sum score, smoking status, energy expenditure

(kcal/d), BMI and WHR; 5 <88/≥88cm waist circumference for women and <102/≥102 cm waist circumference

for men

6 adjusted for age (years), sex, diabetes family history, smoking status, energy expenditure (kcal/d), BMI and

WHR; CI: confidence interval, SD: standard deviation, OR: odds ratio

RESULTS 60

Also, further dietary pattern solutions (3–5 factors) were examined (Table S5-7,

appendix), but did not reveal meaningful dietary patterns. The results of these analyses

highlight that the first pattern (“purchase” pattern) remained (characterized by high intake

of soft drinks, juice, sweets, chocolate drink, red meat, vegetable oil, margarine, rice,

eggs) in all other factor solutions. Also, in all other solutions, a potential second pattern

consisted of three food items (carrot, cucumber, lettuce) only, which was highly unlikely to

reflect a true dietary pattern. Similarly, in a three-factor solution, the “traditional” pattern

remained as a third pattern (Table S5), whereby it separated into a “starchy food” pattern

(characterized by high intake of plantain, cassava, garden egg, fish, green leafy

vegetables) and a “groundnut and beans” pattern (characterized by a high intake of

beans, groundnut, maize (banku), millet) in a four-factor solution (Table S6). Lastly, a five-

factor solution, revealed a fifth pattern characterized by a high intake of bread and milk –

only two food items (Table S7). This lack of plausibility supports the two-factor solution.

Finally, the two-factor solutions were examined with single fruit and single vegetable items

or with various fruit and vegetable groups (Table S8-9, appendix). However, the single

fruit items (banana, mango, orange, pineapple and avocado) didn’t reveal into a clear

separation between the two identified dietary patterns (Table S8). Also, different fruit

(classical fruits, citrus fruits and exotic fruits) and vegetable groups (leafy vegetables and

vegetables) didn’t separate the two identified dietary patterns (Table S9). This supports

the use of the 33 food items and groups for the factor analysis as shown in Table S4

(appendix).

3.3.4 Dietary pattern derived by reduced rank regression

The aim of the RRR method is to identify dietary patterns that explain as much as possible

variation in response variables. In the first step, response scores are extracted that are a

linear combination of the response variables. In this analysis concentrations of

adiponectin, HDL-cholesterol and triglycerides were used as response variables in RRR.

Concentrations of these response variables differed significantly between diabetes cases

and controls, with lower values of adiponectin and HDL-cholesterol, but higher values of

serum triglycerides in the diabetes group (see Table 7). Among the control group,

adiponectin correlated positively with HDL-cholesterol (r = 0.16, p<0.001) and negatively

with triglycerides (r = -0.17, p<0.001). HDL-cholesterol and triglycerides were weakly

correlated with each other (r = 0.08, p = 0.03).

Table 21 shows the explained variation in the three biomarkers by the three response

scores. The first response score explained 83.5% of the variation in triglycerides, followed

RESULTS 61

by HDL-cholesterol with 10.9% and adiponectin with 5.6%. The second and third

response scores were largely driven by the explained variation in adiponectin and HDL-

cholesterol and only marginally by triglycerides.

Table 21: Explained biomarker variation of the three response scores

Explained variation of biomarker concentration (%)

Adiponectin

HDL-Cholesterol

Triglycerides

Response score 1

5.6

10.9

83.5

Response score 2

51.6

42.8

5.5

Response score 3

45.7

49.4

4.9

1 The biomarker variation explained by the three response scores is calculated by multiplication of

standardized score parameter (obtained from multiple linear regression of biomarkers on original response

scores) and Pearson’s correlation coefficient (of biomarkers with response score) x100.

The weights of the biomarkers for the three response scores are presented in Table 22.

These weights are important for the interpretation of the biological plausibility of the

biomarkers in each response score. Adiponectin and HDL-cholesterol were both inversely

and triglycerides positively associated with the first response score, whereas all three

biomarkers were positively associated with the second response score. With the third

response score, adiponectin showed an inverse and HDL-cholesterol and triglycerides a

positive association.

Table 22: Weight of biomarkers in response scores derived by RRR

Weights of biomarkers

Adiponectin

HDL-Cholesterol

Triglycerides

Response score 1

-0.15

-0.33

0.93

Response score 2

0.72

0.61

0.33

Response score 3

-0.68

0.72

0.15

In the second step, the response scores are projected into the space of the predictors to

produce factor scores that are a linear combination of the predictor variables. In this

analysis the intake frequencies of 35 food items were used as predictors and therefore the

factor scores represent dietary patterns. The number of derived dietary patterns is always

equal to the number of response variables in RRR. The explained biomarker variations of

the three dietary pattern scores are shown in Table 23. The first dietary pattern score

explained 3.8% of the total variation in all three biomarkers and was largely driven by the

explained variation in triglycerides (9.9%) and only marginally by HDL-cholesterol (1.3%)

RESULTS 62

and adiponectin (0.3%). The second and third pattern score explained 1.4% and 1.0% of

the total biomarker variation, respectively (Table 23).

Table 23 : Explained biomarker variation of the three dietary pattern scores

Explained variation (%) of biomarker concentration

Adiponectin

HDL-cholesterol

Triglycerides

Total biomarkers

Dietary pattern score 1

0.3

1.3

9.9

3.8

Dietary pattern score 2

2.1

1.5

0.4

1.4

Dietary pattern score 3

1.7

2.9

0.1

1.0

In summary, only the first response score showed a biological plausible biomarker profile

as seen by the direction of the biomarker weights (Table 22) and also the first dietary

pattern score explained the highest biomarker variation (Table 23). Therefore only the first

dietary pattern score was considered for further analyses.

The first dietary pattern explained 8.0% of the total variation in foods and was

characterized by a high consumption of plantain, garden egg and cassava and a low

intake of juice, sweets, rice, hot chocolate, soft drinks, vegetable oil, red meat, milk and

eggs (Table 24). The food items with factor loadings > 0.20 (marked in bold in Table 24)

were considered to be the main contributors to the pattern score.

RESULTS 63

Table 24: Factor loadings1 of all 35 food items derived by reduced rank regression

Food item

Factor Loading

Plantain

0.38

Garden egg

0.27

Cassava

0.24

Juice

-0.34

Sweets

-0.31

Vegetable oil

-0.29

Rice

-0.27

Milo (hot chocolate)

-0.26

Soft drinks

-0.24

Eggs

-0.23

Red meat

-0.21

Margarine

-0.18

Milk

-0.15

Groundnut

-0.14

Carrot

-0.12

Fruits

-0.09

Lettuce

-0.09

Crab

0.08

Sweet potato

-0.06

Cocoyam

0.06

Beans

-0.06

Cucumber

-0.05

Green leaves

0.05

Agushie (pumpkin seeds)

-0.05

Yam

-0.04

Palm oil

-0.04

Fish

0.04

Porridge

-0.04

Alcoholic drinks

-0.03

Millet

0.02

Okro

-0.02

Bread

0.01

Coffee

0.01

Banku (fermented maize product)

-0.01

Poultry

0.003

1 Factor loadings correspond to correlation coefficients between food intake and the dietary pattern score; the

food items with factor loadings > 0.20 are marked in bold

RESULTS 64

Table 25 shows the characteristics and biomarker concentrations and the median intake

of the 35 food items across dietary pattern quintiles among the controls (n=668).

Participants in the highest quintile of the pattern were older, heavier, and of lower SES

than those in the lower quintiles. No linear trend across quintiles was observed for

adiponectin (p for trend = 0.09). Participants in the highest quintile had lower HDL-

cholesterol concentrations (p for trend = 0.05) and higher triglycerides (p for trend <0.001)

compared to the lower quintiles. With respect to the food intake, five food items showed a

positive association with the dietary pattern score, whereas 18 food items were inversely

associated and 12 food items were not related with the pattern score. The intake of these

35 food items varied clearly across the quintiles of the dietary pattern score. Participants

in the highest quintile consumed plantain 5.5-times more frequently than those in the

lowest quintile. In contrast, participants in the highest quintile consumed juice 3.5-times

less frequently than those in the lowest quintile.

3.3.5 Association between RRR-derived dietary pattern and type 2 diabetes

ORs and corresponding 95% CI for the association between the dietary pattern and type 2

diabetes across quintiles and per 1SD are shown in Table 26. The age and sex-adjusted

OR for type 2 diabetes in the highest quintile compared to the lowest was 11.90 [95% CI:

6.25-22.67], p for trend <0.001. Adjustment for family history of diabetes, smoking,

educational attainment, literacy, unemployment and energy expenditure (model 2) only

slightly attenuated the association (OR for highest quintile: 10.63 [5.40-20.93], p for trend

<0.001). Adjustment for the SES sum score instead of educational attainment, literacy and

unemployment in the multivariate-adjusted model (model 2a) further attenuated the

association between the dietary pattern and type 2 diabetes (OR for highest quintile: 9.45

[4.81-18.57], p for trend <0.001). The dietary pattern remained also strongly associated

with diabetes risk after additional adjustment for BMI and WHR (OR for highest quintile:

7.99 [4.00-15.94], p for trend <0.001). Similarly to the quintile-based analysis, the dietary

pattern increased the odds of type 2 diabetes per 1 SD difference of the pattern score by

65% [43-89%] in the fully adjusted model.

RESULTS 65

Table 25: Characteristics, biomarker concentrations and food intake by quintiles of dietary pattern score among 668 controls

Quintile of the dietary pattern score

Characteristics

p for trend1

134

133

134

132

Sex (female)

106 (79.0)

101 (77.0)

96 (72.0)

107 (80.0)

0.48

Age (years)

32.2 ± 12.9

43.8 ± 14.1

48.5 ± 12.6

52.6 ± 13.5

57.2 ± 12.6

<0.001

BMI (kg/m²)

23.8 ± 4.5

25.8 ± 5.4

26.7 ± 5.1

26.0 ± 5.0

26.7 ± 6.3

<0.001

WHR

0.81 ± 0.07

0.85 ± 0.08

0.88 ± 0.07

0.89 ± 0.07

<0.001

SES sum score (very low SES)

5 (3.7)

18 (13.4)

25 (18.9)

29 (21.6)

41 (30.6)

<0.001

Family history of diabetes

29 (21.6)

34 (25.4)

37 (28.0)

37 (27.6)

30 (22.4)

0.66

Smoking (ever)

4 (3.0)

7 (5.2)

5 (3.8)

4 (3.0)

7 (5.2)

0.78

Energy expenditure (kcal/d)

1,086 (896-1,477)

1,210 (861-1,588)

1,326 (959-1,698)

1,195 (755-1,751)

1,295 (649-1,701)

0.52

Biomarker

Adiponectin (mg/ml)

8.81 (6.37-11.55)

8.36 (6.08-11.16)

8.08 (6.21-11.26)

9.29 (6.89-11.98)

8.76 (6.98-11.76)

0.09

HDL-cholesterol (mmol/L)

1.38 (1.12-1.63)

1.39 (1.14-1.68)

1.42 (1.15-1.66)

1.37 (1.11-1.56)

1.32 (1.09-1.54)

0.05

Triglycerides (mmol/L)

0.92 (0.69-1.22)

1.17 (0.82-1.54)

1.23 (0.96-1.72)

1.27 (0.93-1.75)

1.46 (1.07-1.89)

<0.001

Food intake (servings/week)

positive association

Plantain

1.5 (0.5-3.5)

3.5 (1.5-3.5)

3.5 (1.5-7.0)

5.5 (3.5-7)

7.0 (5.5-7.0)

<0.001

Garden egg

3.5 (1.5-7.0)

7.0 (3.5-7.0)

7.0 (7.0-7.0)

<0.001

Cassava

1.5 (0.5-1.5)

1.5 (1.5-3.5)

3.5 (1.5-5.5)

5.5 (3.5-7.0)

<0.001

Crab

0 (0-0.5)

0.5 (0-0.5)

0 (0-0.5)

0.5 (0-0.5)

<0.001

Cocoyam

0.5 (0-0.5)

0.5 (0-1.5)

0.5 (0.5-1.5)

0.5 (0-1.5)

<0.001

inverse association

Juice

3.5 (0.5-3.5)

0.5 (0-1.5)

0 (0-0.5)

<0.001

Sweets

0.5 (0.5-1.5)

0.5 (0-0.5)

0 (0-0.5)

<0.001

Vegetable oil

3.5 (3.5-7.0)

3.5 (1.5-5.5)

3.5 (1.5-3.5)

1.5 (1.5-3.5)

1.5 (0.5-3.5)

<0.001

Rice

7.0 (3.5-7.0)

3.5 (3.5-7.0)

3.5 (2.5-7.0)

3.5 (1.5-5.5)

1.5 (0.5-3.5)

<0.001

Hot chocolate

3.5 (1.5-5.5)

1.5 (0.5-3.5)

0.5 (0-1.5)

<0.001

Soft drinks

1.5 (0.5-3.5)

1.5 (0.5-1.5)

0.5 (0.5-1.5)

0.5 (0-1.5)

<0.001

Eggs

1.5 (0.5-3.5)

0.5 (0.5-1.5)

0.5 (0-1.5)

0.5 (0-0.5)

<0.001

Red meat

3.5 (1.5-7.0)

1.5 (0.5-3.5)

0.5 (0.5-1.5)

<0.001

Margarine

0.5 (0-1.5)

0 (0-0.5)

<0.001

RESULTS 66

Table 25 continued

Quintile of dietary pattern score

Food intake (servings/week)

p for trend1

inverse association

Milk

3.5 (0.5-7.0)

1.5 (0.5-5.5)

0.5 (0.5-4.5)

0.5 (0-3.5)

0.5 (0-1.5)

<0.001

Groundnut

1.5 (1.5-3.5)

1.5 (0.5-3.5)

1.5 (0.5-2.5)

1.5 (0.5-1.5)

0.5 (0-1.5)

<0.001

Carrot

1.5 (0.5-3.5)

0.5 (0.5-1.5)

0.5 (0-1.5)

<0.001

Fruits

3.5 (1.5-3.5)

1.5 (1.5-3.5)

3.5 (1.5-3.5)

1.5 (1.5-3.5)

<0.001

Lettuce

0.5 (0.5-1.5)

0.5 (0-1.5)

0.5 (0-0.5)

<0.001

Sweet potato

0 (0-0.5)

0 (0-0)

0.004

Beans

1.5 (1.5-3.5)

1.5 (0.5-3.5)

0.002

Cucumber

0.5 (0-1.5)

0.5 (0-0.5)

0.3 (0-0.5)

0 (0-0.5)

0.001

Agushie (pumpkin seeds)

0.5 (0.5-1.5)

0.5 (0-1.5)

0.5 (0.5-1.5)

0.5 (0-1.5)

0.008

no significant association

Green leaves

1.5 (1.5-3.5)

3.5 (1.5-5.5)

3.5 (0.5-3.5)

0.68

Yam

1.5 (1.5-3.5)

1.5 (0.5-3.5)

0.07

Palm oil

3.5 (1.5-5.5)

3.5 (1.5-3.5)

3.5 (1.5-5.5)

0.18

Fish

7.0 (5.5-7.0)

7.0 (7.0-7.0)

0.27

Porridge

0.5 (0-1.5)

0.5 (0-0.5)

0 (0-0.5)

0.5 (0-0.5)

0.43

Alcoholic drinks

0 (0-0)

0.41

Millet

1.5 (0.5-3.5)

1.0 (0.5-3.5)

1.5 (0.5-3.5)

1.5 (0-3.5)

0.23

Okro

1.5 (0.5-3.5)

0.36

Bread

4.5 (1.5-7.0)

3.5 (1.5-7.0)

5.5 (1.5-7.0)

3.5 (1.5-7.0)

0.57

Coffee

0 (0-0.5)

0 (0-0)

0.32

Banku (fermented maize product)

3.5 (1.5-7.0)

3.5 (1.5-5.5)

3.5 (1.5-7)

3.5 (1.5-7.0)

0.91

Poultry

1.5 (0.5-1.5)

0.5 (0.5-1.5)

0.91

Values are expressed as mean ± standard deviation, participant number (%) or median (interquartile range); 1 p-value represents linear trend across quintiles;

RESULTS 67

Table 26: Multivariate-adjusted ORs (95% CI) for type 2 diabetes by quintiles and per 1SD of the dietary pattern score

Quintile

p for trend

OR per 1 SD

Median

-2.05

-1.07

-0.31

0.38

1.30

No. of cases/controls

13/134

45/134

84/132

174/134

222/134

Model 1

1.00

2.85 (1.46-5.59)

5.18 (2.71-9.89)

9.89 (5.23-18.69)

11.90 (6.25-22.67)

<0.001

1.81 (1.60-2.06)

Model 2

1.00

2.78 (1.37-5.66)

4.57 (2.31-9.02)

8.64 (4.43-16.88)

10.63 (5.40-20.93)

<0.001

1.78 (1.55-2.03)

Model 2a

1.00

2.66 (1.31-5.39)

4.43 (2.25-8.72)

7.90 (4.05-15.38)

9.45 (4.81-18.57)

<0.001

1.71 (1.50-1.96)

Model 3

1.00

2.42 (1.17-5.02)

3.98 (1.99-7.97)

6.78 (3.43-13.40)

7.99 (4.00-15.94)

<0.001

1.65 (1.43-1.89)

Model 1: adjusted for age and sex; Model 2: adjusted for age (years), sex, diabetes family history, educational attainment, literacy, unemployment, smoking status and energy

expenditure (kcal/d); Model 2a: adjusted for age (years), sex, diabetes family history, SES sum score, smoking status and energy expenditure (kcal/d); Model 3: Model 2a + BMI

and WHR; 1 p-value represents linear trend across quintiles; CI: confidence interval; SD: standard deviation; OR: odds ratio

RESULTS 68

Sensitivity analyses

Several sensitivity analyses were applied to assess the robustness of the results. First,

interaction analyses were used to examine whether the association between the dietary

pattern score and type 2 diabetes was modified by sex, general obesity or central

adiposity. The association between the dietary pattern and type 2 diabetes was consistent

for women and men (women: OR for 1SD in the full-adjusted model = 1.66 [1.41-1.95],

men: OR = 1.68 [1.28-2.22], p for interaction = 0.58), for BMI-categories (< 30 kg/m²: OR

for 1SD = 1.66 [1.42-1.94], ≥ 30kg/m²: OR = 1.72 [1.26-2.35], p for interaction = 0.32) and

for categories of waist circumference (< 88 cm (women), < 102 cm (men): OR for 1SD =

1.63 [1.36-1.97], ≥ 88/102cm: OR = 1.68 [1.36-2.08], p for interaction = 0.57).

Additionally, the RRR analysis was repeated after excluding participants with lipid-

lowering (2.3%) or anti-inflammatory drug intake (2.0%). The results were virtually

identical (OR for 1SD in the full-adjusted model=1.67 [1.45-1.92]).

The concentrations of adiponectin, HDL-cholesterol and triglycerides chosen as response

variables in the RRR might have changed during the course of diabetes. Although we can

not examine this among the diabetes cases, we can assess whether the positive

association of the dietary pattern score remain with FPG concentrations and HOMA-IR in

the apparently healthy control group. Indeed, FPG concentrations slightly increased from

the lowest quintile to the fourth quintile of the pattern score, however, in the fifth quintile

FPG decreased (mean ± standard deviation: 4.50 ± 0.67, 4.57 ± 0.76, 4.59 ± 0.78, 4.63 ±

0.71 and 4.51 ± 0.60 mmol/L, p for trend = 0.70). No clear trend was seen for HOMA-IR:

median (interquartile range): 1.47 (0.94-2.27), 1.49 (1.06-2.26), 1.46 (0.95-2.21), 1.27

(0.84-1.92) and 1.34 (0.79-2.16), p for trend = 0.08.

Finally the dietary pattern was simplified. The first step for the simplification was to select

the main contributors for the pattern score. Therefore a stepwise linear regression with the

first response score as dependent variable and all 35 food items as independent variable

were applied. Thus, only food items that were significantly associated with the response

score were considered for the simplified score. Table 27 shows the selected food items in

the order of their stepwise selection with regression coefficients and p-values. The food

items plantain, cassava, garden egg and poultry were positively associated, whereas fish,

red meat, vegetable oil, juice and sweets were inversely associated with the response

score.

RESULTS 69

Table 27: Selected food items identified by linear stepwise regression

Food item

Regression coefficient

p-value

Cassava

0.033

0.008

Plantain

0.042

0.002

Fish

-0.041

0.028

Red meat

-0.028

0.047

Poultry

0.047

0.018

Garden egg

0.043

0.002

Vegetable oil

-0.038

0.010

Juice

-0.068

0.005

Sweets

-0.099

0.030

The response score was used as dependent and all 35 food items as independent variable in linear stepwise

regression analysis

In the next step of the simplification of the dietary pattern score only the nine selected

food items were used to calculate the simplified dietary pattern score. The simplified score

represents the sum of the unweighted standardized intake of the nine food items. The

simplified score was strongly correlated with the original dietary pattern score (r = 0.91,

p≤0.001) and showed similar associations with sociodemographic characteristics and the

three response variables (Table 28). The characteristics, biomarker concentrations and

the median intake of the nine food items across the simplified dietary pattern quintiles

among the controls (n = 668) are shown in Table 28. Participants in the highest quintile of

the pattern were older, heavier, and of lower SES than those in the lower quintiles. No

linear trend across quintiles was observed for adiponectin (p for trend = 0.09). Participants

in the highest quintile had lower HDL-cholesterol concentrations (p for trend = 0.05) and

higher triglycerides (p for trend <0.001) compared to the lower quintiles. Furthermore, the

strength of the association between the simplified score and type 2 diabetes was nearly

identical to the association between the original dietary pattern score and diabetes. The

OR [95%CI] per 1SD difference was 1.66 [1.41-1.96] in the fully-adjusted model (model 3)

(Table 29). The strength of association between the simplified score and type 2 diabetes

across quintiles was somewhat lower as for the original score. In the full-adjusted model

the OR for the highest compared to the lowest quintile was 4.02 [2.32-6.96].

The importance of individual components for type 2 diabetes was examined by

sequentially subtracting each component from the simplified score (Table 30). The

removal of juice, plantain, red meat, sweets, vegetable oil and garden egg weakened the

association for type 2 diabetes. The subtraction of fish and poultry showed rather minor

changes. Surprisingly, the subtraction of cassava, which was positively associated with

the dietary pattern score, increased the association for type 2 diabetes.

RESULTS 70

Table 28: Characteristics, biomarker concentrations and food intake by quintiles of the simplified dietary pattern score among 668 controls

Quintile of the simplified dietary pattern score

Characteristics

p for trend1

131

123

135

148

131

Sex (female)

108 (82.4)

95 (77.0)

102 (75.6)

111 (75.0)

100 (76.3)

0.61

Age (years)

34.8 ± 14.2

42.2 ± 14.6

47.4 ± 13.3

52.8 ± 13.0

55.9 ± 14.1

<0.001

BMI (kg/m²)

24.4 ± 4.9

25.8 ± 5.4

26.4 ± 5.1

26.7 ± 5.5

25.5 ± 5.8

<0.001

WHR

0.82 ± 0.07

0.86 ± 0.07

0.87 ± 0.07

0.88 ± 0.07

<0.001

SES sum score (very low SES)

10 (37.6)

14 (11.4)

22 (16.3)

29 (19.6)

43 (32.8)

<0.001

Family history of diabetes

37 (28.2)

29 (23.6)

37 (27.4)

38 (25.7)

26 (19.9)

0.53

Smoking (ever)

3 (2.3)

7 (5.7)

3 (2.2)

8 (5.4)

6 (4.6)

0.43

Energy expenditure (kcal/d)

1,146 (876-1,535)

1,198 (949-1,687)

1,309 (890-1,764)

1,235 (738-1,642)

1,252 (679-1,719)

0.52

Biomarker

Adiponectin (mg/ml)

8.99 (6.67-11.32)

7.78 (5.88-11.46)

8.12 (6.12-11.91)

8.83 (6.64-11.98)

8.83 (7.00-12.68)

0.09

HDL-cholesterol (mmol/L)

1.36 (1.09-1.69)

1.38 (1.16-1.61)

1.46 (1.18-1.73)

1.37 (1.13-1.56)

1.30 (1.07-1.55)

0.05

Triglycerides (mmol/L)

0.88 (0.72-1.23)

1.10 (0.86-1.47)

1.25 (0.94-1.75)

1.29 (0.95-1.74)

1.46 (0.99-1.86)

<0.001

Food intake (servings/week)

positive association

Plantain

1.5 (0.5-1.5)

1.5 (1.5-3.5)

3.5 (1.5-5.5)

5.5 (3.5-7)

7.0 (7.0-7.0)

<0.001

Garden egg

3.5 (1.5-7.0)

7.0 (3.5-7.0)

7.0 (7.0-7.0)

<0.001

Cassava

1.5 (0.5-1.5)

1.5 (1.5-3.5)

3.5 (1.5-5.5)

7.0 (3.5-7.0)

<0.001

inverse association

Juice

1.5 (0.5-3.5)

0.5 (0-1.5)

0 (0-0.5)

<0.001

Sweets

0.5 (0.5-1.5)

0.5 (0-1.5)

0.5 (0-0.5)

0 (0-0.5)

<0.001

Vegetable oil

5.5 (3.5-7.0)

3.5 (1.5-3.5)

1.5 (1.5-3.5)

1.5 (0.5-3.5)

<0.001

Poultry

1.5 (0.5-1.5)

0.5 (0.5-1.5)

0.5 (0-1.5)

0.84

Fish

7.0 (5.5-7.0)

7.0 (7.0-7.0)

0.11

Red meat

3.5 (1.5-7.0)

1.5 (0.5-3.5)

0.5 (0.5-1.5)

<0.001

Values are expressed as mean ± standard deviation, participant number (%) or median (interquartile range); 1 p-value represents linear trend across quintiles

RESULTS 71

Table 29: Multivariate-adjusted ORs (95% CI) for type 2 diabetes by quintiles and per 1SD of the simplified dietary pattern score

Quintile

p for trend

OR per 1 SD

No. of cases/controls

27/131

43/123

92/135

158/148

218/131

Median

-1.71

-0.87

-0.24

0.32

1.02

Model 1

1.00

1.41 (0.82-2.45)

2.57 (1.55-4.25)

3.63 (2.22-5.93)

5.35 (3.26-8.76)

<0.001

1.86 (1.61-2.15)

Model 2

1.00

1.53 (0.84-2.78)

2.36 (1.37-4.07)

3.59 (2.11-6.10)

5.24 (3.08-8.93)

<0.001

1.84 (1.57-2.15)

Model 2a

1.00

1.51 (0.83-2.72)

2.33 (1.35-4.00)

3.30 (1.95-5.60)

4.72 (2.77-8.02)

<0.001

1.75 (1.49-2.05)

Model 3

1.00

1.36 (0.74-2.51)

2.11 (1.21-3.70)

2.91 (1.69-5.02)

4.02 (2.32-6.96)

<0.001

1.66 (1.41-1.96)

Model 1: adjusted for age and sex; Model 2: adjusted for age, sex, diabetes family history, educational attainment, literacy, unemployment, smoking status and energy expenditure;

Model 2a: adjusted for age, sex, diabetes family history, SES sum score, smoking status and energy expenditure; Model 3: adjusted for age, sex, diabetes family history, SES sum

score, smoking status and energy expenditure, BMI and WHR; 1 p-value represents linear trend across quintiles; CI: confidence interval; SD: standard deviation; OR: odds ratio

Table 30: Importance of individual food components of the dietary pattern score

OR (95% CI) per 1SD

CIE (%)

Simplified dietary pattern score

1.66 (1.41-1.96)

Simplified dietary pattern score without juice

1.46 (1.25-1.70)

- 12.0

Simplified dietary pattern score without plantain

1.47 (1.25-1.72)

- 11.4

Simplified dietary pattern score without red meat

1.54 (1.32-1.80)

- 7.2

Simplified dietary pattern score without sweets

1.55 (1.33-1.82)

- 6.6

Simplified dietary pattern score without vegetable oil

1.56 (1.33-1.83)

- 6.0

Simplified dietary pattern score without garden egg

1.57 (1.34-1.85)

- 5.4

Simplified dietary pattern score without poultry

1.70 (1.44-2.01)

+ 2.4

Simplified dietary pattern score without fish

1.72 (1.46-2.03)

+ 3.6

Simplified dietary pattern score without cassava

2.15 (1.80-2.56)

+ 29.5

Simplified dietary pattern score: sum of unweighted standardized intake of nine food items, which were significantly associated with the response score (plantain + cassava +

gardenegg – fish – red meat – poultry – juice – sweets – vegetable oil; CIE: Change in estimate: difference between ORs divided by OR of original factor score and multiplied by

100 (%);CI: confidence interval; SD: standard deviation; OR: odds ratio; Model 3 adjustment were applied: adjusted for age, sex, diabetes family history, SES sum score, smoking

status, energy expenditure, bmi and whr

DISCUSSION 72

4 DISCUSSION

4.1 Discussion of results

4.1.1 Anthropometric characteristics

In this mainly female, middle-aged, and of low SES study population from urban Ghana,

prevalences of overweight and obesity were higher among women than men. In particular,

central obesity is common among the majority of the women. This is consistent with other

studies that reported high rates of obesity in urban Ghana [33, 161]. A cross-sectional

study from urban Ghana did also find a higher prevalence of general obesity (BMI ≥ 30

kg/m²) among women (36%) than men (10%) [162]. Also in the Women’s Health Study of

Accra, consisting of 2,814 women, aged ≥18 years and living in the Accra Metropolitan

Area, 79% of the women exhibited a central obesity (WC ≥ 88cm). Furthermore, the

majority of these Ghanaian women (72%) were not satisfied with their body size, and 30%

of them preferred a heavier body size. In addition, 61% of the women who were satisfied

with their body size were overweight or obese [163]. One explanation for the high obesity

rates is the regional perception of obesity as a marker of affluence [1] and obesity is

associated with beauty and health in women [38].

4.1.2 Associations between anthropometric measures and type 2 diabetes

The present thesis investigated for the first time various anthropometric measures and

their relationships with type 2 diabetes in an urban Ghanaian population. In this study,

measures of central obesity but not of general obesity were positively associated with type

2 diabetes in both women and men [164]. Specifically, BMI was not associated with the

risk of type 2 diabetes, while WHR showed the strongest relationship in both sexes.

So far, only a few studies evaluated the association between obesity measures and type 2

diabetes in SSA (Table 4). These studies have been conducted in various regions of SSA,

applied varying obesity measurements and diabetes ascertainment, and included a wide

range of sample sizes. Furthermore, interpretation of the findings is further complicated by

high levels of ethnic diversity possibly resulting in different body compositions in these

African studies. Consequently, the heterogeneity of these studies complicates the

comparison of their results.

DISCUSSION 73

Results from the few prior existing studies are inconsistent, suggesting either a positive

[44, 71-73] or no association [45, 46] between BMI and type 2 diabetes. Indeed, BMI was

neither associated with type 2 diabetes among women nor among men in the present

study. A possible explanation for the lack of association could be the high prevalence of

type 2 diabetes cases with a poor glycaemic control (46%) in this population. Performing a

sensitivity analysis among the poorly controlled diabetes cases (medication + FPG ≥

7mmol/L) showed a reduced mean BMI and mean body cell mass. This argues for

diabetic muscle atrophy as an underlying mechanism [165]. However, after exclusion of

poorly controlled diabetes cases, the ORs for type 2 diabetes increased slightly, but were

still not significant. Thus, a poor glycaemic control is unlikely to explain the difference in

the associations observed for BMI and central obesity measures in this Ghanaian study

population. Furthermore, another hypothesis for the lack of association is that SES could

have modified the association between BMI and type 2 diabetes. In fact, the mean BMI

rose with increasing SES sum score and was higher in women and men with moderate

SES compared to very low SES. Nevertheless, no interaction for the association between

BMI and type 2 diabetes with SES was found in both sexes.

The lack of a clear association between BMI and type 2 diabetes in this study is in

contrast to findings of studies in other ethnic populations [20]. However, general body fat

mass may play a substantially different role among populations with African ancestry

compared to Caucasians. This is supported by the observation that associations between

the fat mass and obesity-related (FTO) gene (a major obesity risk locus), and diabetes are

weaker or even contrary in populations with African ancestry compared to European

populations [166-168].

With regard to central obesity, findings are more consistent across SSA populations: WC

[44-46] or WHR [44, 72, 73] were more strongly associated with diabetes than BMI. This is

in line with the present study, where WHR showed the strongest relationship with type 2

diabetes independent of BMI. WHR was less strongly correlated with BMI than WC, and is

therefore a more specific surrogate for fat distribution. This indicates the importance of the

body fat distribution in the diabetes development compared to generalized obesity.

Visceral fat is highly metabolically active, and its accumulation, reflected by larger WHR,

causes an increased delivery of free fatty acids to the liver, resulting in insulin resistance,

hyperinsulinaemia and hypertriglyceridaemia [169] and hence greatly increases the risk of

type 2 diabetes.

DISCUSSION 74

HC was inversely associated with type 2 diabetes among men. This is in line with prior

studies that have reported a protective association of larger HC on diabetes in South

Africans [45], Asian-Indian men [170] and four non-Caucasian ethnic groups [171].

Regarding a potential biological mechanism of this relation, increased muscle mass or

femoral fat mass accumulation reflected by large hips may be responsible. It has been

suggested, that the subcutaneous femoral fat tissue is protective, because of high

lipoprotein lipase activity and due to the lower lipolysis rate in this region compared to the

visceral region [172]. These fat depots are therefore more likely to protect liver and

muscle by taking up and storing increased concentrations of free fatty acids [173].

4.1.3 Discrimination of type 2 diabetes cases and controls

This is the first study that compared the discriminative power of different obesity measures

for type 2 diabetes in SSA [164]. WHR showed the best discriminatory ability for diabetes

in this Ghanaian population. A recent study among Cameroonians suggested that central

obesity measures are better predictors for diabetes than BMI [46]. In this study WC

showed the best discriminatory ability to identify diabetes. Further, prior studies from SSA

on cardiovascular diseases observed that central obesity measures were better predictors

of disease risk than BMI. A cross-sectional study in Ethiopia evaluated the relation

between measures of obesity and cardiovascular disease risk [174]: WC in women and

WHtR in men were the measures most strongly associated. Overall, comparisons of ROC

curves identified that WC was the best predictor of cardiovascular disease risk among this

population. Furthermore, several other non-Caucasian indigenous populations, such as

Taiwanese [175], South Indians [176], Australian Aboriginal people and Torres Strait

Islanders [177] consistently observed that WHR was a better predictor of diabetes risk

than BMI.

Even though data from African Americans are not directly comparable to this present

study, they also point to a particularly important role of central (rather than general)

obesity as a risk factor for diabetes among blacks. In the prospective Insulin Resistance

Atherosclerosis Study, measures of central and general adiposity similarly predicted type

2 diabetes, except in African Americans, in whom central obesity measures were more

predictive [40]. In line with this, the Atherosclerosis Risk in Communities Study, showed

that WC was a better predictor for incident diabetes than BMI in African Americans

compared to Caucasians [178].

DISCUSSION 75

4.1.4 Examination of cut-offs for obesity measures

There is an ongoing discussion, whether the recommended cut-off points for obesity

measures, which have been mainly derived among Caucasian populations, are

appropriate for other populations [52, 54]. As for the population-specific anthropometric

cut-off points, data from Asian populations have revealed that lower values of general and

central obesity measures might be meaningful for the identification of individuals at risk of

diabetes [52, 55, 56]. However, there is insufficient evidence for specific cut-offs for the

identification of type 2 diabetes in SSA populations.

This is the first study, which evaluated population-specific cut-off points for BMI, WC and

WHR to identify type 2 diabetes in an SSA population: WHR was the best obesity

measure to identify diabetes cases at the recommended cut-offs with high sensitivity and

specificity in both sexes. The optimal WHR cut-off point in women (0.88) exceeded the

reference value (0.85), but was identical for men (0.90). With respect to WC,

recommended cut-off points identified type 2 diabetes well in women (high sensitivity), but

not in men. The optimal cut-off point for WC was higher than the recommended cut-offs

for both, overweight and obesity, in women (91.7 vs. 80.0/88.0 cm) and lower in men

(83.4 vs. 94.0/102.0 cm). This is in line with previous findings that West African men have

generally lower WC values than Western populations [68]. This cross-sectional study from

SSA recommended WC cut-off points of 75.6 and 80.5 cm for men and 71.5 and 81.5 cm

for women of Nigerian and Cameroon origin, respectively, for the identification of

hypertension [68]. In addition, a study from urban South Africa reported that the WC cut-

off point for the identification of the metabolic syndrome among women was higher (91

cm) compared to the cut-off recommended by the IDF (80 cm) [70]. Also a study from

rural South Africa recommended an optimal WC cut-off point of 86 cm in men and 92 cm

in women for the identification of the presence of at least two components of the metabolic

syndrome [69]. With regard to BMI, the recommended cut-off for overweight (≥ 25 kg/m²)

identified type 2 diabetes moderate, but cut-off for obesity (≥ 30 kg/m²) showed a low

sensitivity and specificity in both sexes. Furthermore, the optimal BMI cut-offs were

slightly higher compared to the recommended cut-off points for overweight and obesity, in

both women (26.2 vs. 25.0/30.0) and men (26.7 vs. 25.0/30.0). With respect to the limited

studies from SSA, further research is needed to increase the evidence for specific cut-offs

among SSA populations.

DISCUSSION 76

4.1.5 Intake of energy, macronutrients and food groups

The diet of this Ghanaian study population was rich in carbohydrates and fat. The diet

largely relied on energy dense food such as plantain, banku, bread, rice, fish (mostly fried)

and palm oil. Main dishes are mainly served as soups or stew including plantain, fish and

vegetables (tomato, pepper, onions). The majority of this Ghanaian study population

consumed three meals per day. These findings are comparable with results from a cross-

sectional study among 400 rural Ghanaian women [79]: Women consumed three meals

per day and the most frequently consumed food items were the starchy staples, maize,

fish, pepper, onion and tomato. Fish was the main source of animal protein and meat and

milk were less frequently consumed. This is in line with the low consumption of meat and

milk in the present study.

4.1.6 Dietary patterns derived by factor analysis

Two dietary patterns were identified by using factor analysis in this Ghanaian population

[179]: The “purchase” dietary pattern was characterized by a high consumption of sweets,

rice, protein-rich foods (red meat, poultry, eggs and milk), fruits and vegetables and low

consumption of plantain. The “traditional” dietary pattern was characterized by a high

intake of plantain, green leafy vegetables, beans, garden egg, fruits, fish, fermented maize

products and palm oil. While participants in the “purchase” dietary pattern were younger,

leaner and of better SES, participants in the “traditional” pattern were older, heavier and of

lower SES.

Exploratory derived dietary patterns in SSA

Only five cross-sectional studies in comparatively small and specific populations have

attempted to identify dietary patterns in SSA by factor analysis or cluster analysis (see

Introduction, Table 5). These studies had diverse settings, used different criteria to

identify dietary patterns and also the diets per se differed strongly from the Ghanaian diet

in the present study. This makes the findings difficult to compare to dietary patterns

among other SSA populations. However, some differences and similarities have to be

discussed: A cross-sectional study from Cameroon identified two dietary patterns by factor

analysis [98]. The “fruit and vegetable” pattern (characterized by high consumption of

fruits, green and dark yellow vegetables, tubers, oils and fats, fish, rice, milk, pasta, soft

drinks, sweets and meat) was similar to the mixed “purchase” dietary pattern of this study.

In contrast, the “meat” pattern showed high factor loadings for bush meat, poultry, and red

meat, but low factor loadings for sweets, cakes and sugar. Among 1086 elderly people in

DISCUSSION 77

urban and rural Botswana, five dietary patterns were identified using factor analysis [99]:

A “beer”, “meat and fruit”, “vegetable and bread”, “seasonal produce”, and “milk, tea and

candy” pattern, whereby only the “meat and fruit” pattern exhibited some similarities to the

“purchase” dietary pattern of this study. Furthermore, among 1072 urban Burkinans, a

“snacking” pattern (fried foods, sugar-sweetened products, cereals and dairy products)

and a “modern foods” dietary pattern (processed meats, eggs, low in nuts, seeds, cereals,

and beans) were revealed by principal component analysis [78]. In a cross-sectional study

from rural Tanzania, five dietary patterns were identified by applying principal component

analysis: A “purchase” dietary pattern (characterized by bread and cakes, sugar, and

black tea), a “traditional-coast” (characterized by fruits, nuts, starchy plants, and fish), a

“‘traditional-inland” (characterized by cereals, oils and fats, and vegetables), a “pulses”

(characterized mainly by pulses, with few or no vegetables) and an “animal products”

pattern (characterized by a high consumption of meat, eggs and/or milk) [97]. However,

these patterns showed no similarities with the dietary patterns of this study. The best

comparable dietary patterns to this study were identified among 200 urban Beninese

residents by cluster analysis [96]: A “traditional” dietary pattern characterized by high

intakes of grains, fruits, fish, and green leafy vegetables and a “transitional” pattern

characterized by high intakes of bread, pasta, roots, nuts, meat, eggs, dairy, fats, and

sweets. As a mixed pattern, the “transitional” pattern included imported and traditional

foods and is similar to the “purchase” dietary pattern of the present study. These data

highlight the difficulties to compare, not to say transfer, established dietary patterns in

African populations with other SSA regions.

Characteristics of exploratory derived dietary patterns in SSA

It is not trivial to find common characteristics of the identified dietary patterns in SSA. For

example, the pattern with high consumption of meat in Cameroon predominated in

participants of low educational level [98], whereas in Benin participants of high SES

adhered to such pattern [96]. With respect to a pattern loading high on vegetables, this

was characterized by individuals of low SES in Benin [96], while in Botswana, such

pattern prevailed in households with children [99]. In part, this is in line with the

observations of the present study. Adherence to dietary patterns was associated with

SES. Participants with a high score of the “purchase” dietary pattern were characterized

by a better SES, whereas those with a high score of the ‘traditional’ dietary pattern were

more deprived. Indeed, the “purchase” dietary pattern included food items (sweets, red

meat and poultry) that are expensive for the majority of the population in this area. It

seems that particularly people above the average income level and with better knowledge

of healthy food adhere to this. In contrast, low income and poor education might favor the

DISCUSSION 78

adherence to the “traditional” dietary pattern. Furthermore, the study from Benin observed,

that the proxy indicator for SES “birthplace” was associated with the “transitional” and

“traditional” dietary pattern [96]. Participants in the transitional cluster were more often

born in urban areas and those in the traditional cluster mainly stemmed from rural areas.

4.1.7 Associations between dietary patterns and type 2 diabetes

The “purchase” dietary pattern was associated with a reduced odds of type 2 diabetes,

whereas the “traditional” dietary pattern was related to an increased odds of type 2

diabetes. With respect to SSA, no previous study evaluated the association between

dietary patterns and type 2 diabetes. However, a few investigated the relationship

between dietary patterns and various health outcomes. In the aforementioned study from

Benin, no clear associations of dietary patterns (characterized by urban vs. rural

citizenship and high vs. low SES), with self-reported health status, were observed [96].

Nevertheless, urbanization and the epidemic of type 2 diabetes are paralleling in SSA [15,

31]. With respect to Ghana, the proportion of the population living in urban areas has

increased from 23 to 51% during the past 50 years, with the second highest proportion of

urban citizens (61%) in the Ashanti Region [136]. Thus, it is justified to hypothesize that

peasants from rural Ghana moved to the cities, where they faced lower income and

altered food availability. In this situation, well-known traditional foods that are satiating and

inexpensive, such as carbohydrate-dense foods of the ‘traditional’ dietary pattern, appear

preferable. This could be a possible explanation for the observed positive association

between the “traditional” dietary pattern and type 2 diabetes. However, this study has no

information about movements from rural to urban areas and therefore cannot investigate

this hypothesis so far. Of note, such associations between SES and the adherence to

dietary patterns have been observed in Western populations, too. So-called prudent or

healthier patterns were associated with increased income, better education and older age

[180]. With respect to diet–disease relationships in the African region, the scarce data

remain inconclusive. The Cameroon study was designed to identify patterns associated

with hypertension and observed that the fourth quartile of the ‘fruit and vegetable’ pattern

compared with the first quartile reduced the risk of hypertension by 60% [98]. The

Botswanian study [99] did not investigate associations between dietary patterns and

health outcomes. In Burkina Faso, obesity was positively associated with the “snacking”

pattern, but not with the “modern foods” dietary pattern [78]. Similarly, in Tanzania, the

“purchase” dietary pattern (characterized by bread and cakes, sugar, and black tea)

showed the strongest positive association with BMI [97]. Clearly, these findings warrant

DISCUSSION 79

further investigations to understand the determinants of adherence to dietary patterns in

SSA.

Until now, the present study is the first to investigate associations between dietary

patterns and the risk of type 2 diabetes in SSA. Prior studies evaluating associations

between dietary patterns and type 2 diabetes are difficult to compare to the herein

identified dietary patterns due to differences in food availability, processing and

consumption in SSA. However, the inverse association of the “purchase” dietary pattern in

the present study, which was rich in fruits and vegetables, appears to be consistent with

previous studies reporting an inverse association for patterns sometimes called “prudent”

or “healthy” characterized by higher intakes of fruits and vegetables [103, 104]. However,

there are important differences in pattern structure observed in the present study

compared with patterns evaluated in European [103, 110], US [104, 109] or Asian

populations [106, 107, 181]. The “purchase” dietary pattern was also characterized by a

high intake of red meat. Previous studies identified a “Western” dietary pattern with high

intake of red meat; this pattern was associated with an increased risk of type 2 diabetes in

other populations [104, 109, 181]. This is also true for the consumption of sweets in such

“Western” patterns, which is positively associated with type 2 diabetes in Western

populations [101, 105, 109], but inversely in the present study. While the consumption of

sweets was low in this Ghanaian study population and the types of red meat differed from

those in Western populations which might explain that the pattern was – in contrast to so-

called “Western” patterns – inversely associated with the risk of type 2 diabetes, the

present study highlights that dietary patterns derived by exploratory methods are specific

for this African study population. An alternative explanation for the inverse association of

the “purchase” pattern with type 2 diabetes could be that the combined consumption of

fruits, rice, meat and other food groups [179] may represent dietary diversity, which is

inversely associated with biomarkers of type 2 diabetes [182].

A local specificity might be postulated and even more pronounced for the “traditional”

dietary pattern, which was characterized by high intakes of fermented maize products,

palm oil and other traditional foods which are merely absent in diets in Western and Asian

regions. The unexpected positive association between the “traditional” dietary pattern and

type 2 diabetes in the present study is in contrast to findings from a Lebanese case-

control study [183]. This study identified a “traditional” dietary pattern that was

characterized by high intake of olives oil, fruits and vegetables, whole wheat bread, and

traditional dishes and was inversely associated with type 2 diabetes (OR [95%CI] = 0.46

[0.22-0.97]) [183]. However, studies among other non-white populations, showed also

harmful associations between the traditional foods and different health outcomes [184-

187]. A cross-sectional study among 2,374 Inuit in Greenland observed that the traditional

DISCUSSION 80

pattern was positively associated with type 2 diabetes, impaired fasting glucose and

fasting plasma glucose and negatively related to ß-cell function [185]. Eilat-Adar et al.

found higher HDL-cholesterol levels, but also higher triglycerides, BMI and HOMA-IR in

American-Indians who adhered to the traditional pattern [184]. In a prospective Swedish

study, slightly higher hazard ratios (HR per 1-point increase in score: 1.04 [95% CI 1.01-

1.07], p<0.018) for all-cause mortality in men have been found for higher scores of the

traditional “Sami” diet [187]. Adherence to a traditional pattern high in beans and legumes,

rice and oil, and low in high-fat dairy, condiments and nuts and seeds was associated with

lower HDL-cholesterol and a higher odds of metabolic syndrome (OR [95%CI] = 1.7 [1.04-

2.7]) among Puerto Rican elders living in Massachusetts [186]. The authors attempt to

find possible explanations for the association between the traditional dietary patterns and

poor health outcomes. Eilat-Adar et al. hypothesized that the preparation methods have

changed from cooking to more frying the foods [184]. This could also be a possible

explanation for the positive association between the “traditional” dietary pattern and type 2

diabetes in the present study. The explanation for the detrimental association of the

traditional rice and beans pattern with the metabolic syndrome in the study by Noel et al.

was that a diet rich in total carbohydrates with high glycemic load foods could promote

lower HDL-cholesterol and higher triglyceride concentrations [186]. Another explanation

given by Jeppesen et al. was that the traditional Inuit diet could contain contaminants such

as mercury and persistent organic pollutants contributing to the decreased ß-cell function

[185]. With respect to the traditional “Sami” diet, a limited FFQ and residual confounding

by unmeasured factors were the explanation for the positive association with all-cause

mortality in the Swedish study [187].

An alternative explanation for the detrimental association of the “traditional” pattern in the

present study could be that this pattern is not diverse. In other studies, dietary diversity

was inversely associated with biomarkers of type 2 diabetes [182], obesity [188],

cardiovascular risk factors [189] and the metabolic syndrome [190].

These findings highlight that dietary patterns are population-specific, especially in SSA

where food availability and consumption may be substantially different due to distinctions

in climate, agriculture, food production and processing, and cultural habits compared to

other regions [191, 192].

DISCUSSION 81

4.1.8 Dietary pattern derived by reduced rank regression

The present thesis identified for the first time a dietary pattern among an SSA population

by using 35 food items as predictor variables and adiponectin, HDL-cholesterol and

triglycerides as response variables with the RRR approach. Participants in the highest

quintile of this dietary pattern were older, heavier, and of lower SES than those in the

lower quintiles.

The identified dietary pattern was characterized by a high consumption of traditional foods

(plantain, garden egg and cassava) and low consumption of purchase foods (juice,

sweets, vegetable oil, rice, milo, soft drinks, eggs and red meat). This pattern was related

to higher concentrations of triglycerides and lower concentrations of HDL-cholesterol.

The results of the RRR analysis confirm the previous findings of the exploratory factor

analysis in this study population. The food items positively associated with the dietary

pattern score in the RRR analysis (plantain and garden egg) were also included in the

“traditional” dietary pattern. In addition, those food items inversely associated with the

pattern score in RRR (rice, juice, eggs, milo, sweets and red meat) were also included in

the “purchase” dietary pattern. Moreover, previous analysis in the KDH study has revealed

a strong association between high levels of triglycerides (≥ 1.695 mmol/L) and type 2

diabetes in this study population: multivariate-adjusted OR= 1.83 [95% CI: 1.13-2.97]

[130]. Indeed, triglycerides were the main drivers of the association between the RRR-

derived dietary pattern and type 2 diabetes.

4.1.9 Association between RRR-derived dietary pattern and type 2 diabetes

This is the first study in SSA that evaluated the association between a RRR-derived

dietary pattern and type 2 diabetes. The advantage of the RRR approach is that it

incorporates information on biological pathways and thus derives dietary patterns

predictive of a disease [95].

In the present study, a high dietary pattern score was associated with an increased risk of

type 2 diabetes. Also the simplified dietary pattern showed a positive association with type

2 diabetes in this study. The simplified dietary pattern score (nine food items) correlated

highly with the more complex original score (including 35 food items), but has the

advantage that it is easier to calculate and interpret [193]. This simplification approach is

often used in factor analysis [100, 101, 105] or RRR analysis [110-112, 194]. Some

studies used only food items with high factor loadings for the simplification of the dietary

patterns [100, 105, 106, 194, 195]. However, applying a cut-point is a subjective decision

DISCUSSION 82

of the investigator and can vary between factor loadings of ≥ 0.20 to ≥ 0.30 in studies

[100, 105, 106, 194, 195]. Other studies used only those food items that contributed most

to the explained interindividual variation in the dietary pattern score to simplify their dietary

patterns [196-198]. Another method to simplify a dietary pattern is to test for significant

associations between the food groups and the response score in RRR [199]. In the

present thesis, this objective method was used by applying a stepwise linear regression

with the first response score as dependent variable and all 35 food items as independent

variable. Thus, only the nine food items that were significantly associated with the

response score were considered for the simplified score. The importance of individual

components for type 2 diabetes was examined by sequentially subtracting each

component from the simplified pattern score. However, no single food item was

responsible for the positive association with type 2 diabetes.

Overall, few epidemiological studies have investigated the association between dietary

patterns derived by RRR and type 2 diabetes in Western populations. Although these

studies used various intermediate markers as response variables including inflammatory

biomarkers [111, 113], HOMA-IR [112] and HbA1c, HDL-cholesterol, C-reactive protein

(CRP), and adiponectin [110], the explained variation in biomarkers of these studies is

comparable to findings of this study.

Two studies investigated the generalizability of the associations between RRR-derived

dietary patterns and type 2 diabetes among independent European [114] and US

populations [115]. While the EPIC-InterAct study found a good generalizability for the

three RRR dietary pattern scores based on the American NHS, the German EPIC-

Potsdam Study and the British WHS [114], the American Framingham Offspring Study

reported a good generalizability for the American NHS derived dietary pattern, but not for

the European derived dietary patterns (EPIC-Potsdam Study and WHS) [115]. Thus, the

generalizability of dietary patterns associated with diabetes risk may be better in

populations with comparable dietary intakes. Indeed, the transferability of previous RRR

dietary patterns established in European and US populations to an urban Ghanaian

population is complicated by the differences in the diet per se. However, all previous

prospective studies found strong associations between dietary patterns obtained by RRR

and type 2 diabetes [110-113]. The strong relationships observed with the RRR method

can partly be attributed to the use of disease-related biomarkers. In the Whitehall II Study,

a diet high in low calories/soft drinks, onions, sugar-sweetened beverages, burgers and

sausages, crisps and other snacks and white bread and low in medium-/high-fiber

breakfast cereals, jam, French dressing/vinaigrette and whole meal bread was associated

with a two-to threefold increase in diabetes risk by using HOMA-IR as the response

DISCUSSION 83

variable [112]. The Insulin Resistance Atherosclerosis Study found a threefold to more

than fourfold increased odds of diabetes associated with a dietary pattern high in red

meat, low-fiber bread and cereal, dried beans, fried potatoes, tomato vegetables, eggs,

cheese, and cottage cheese and low wine by using the inflammatory markers

plasminogen activator inhibitor-1 and fibrinogen as the response variable [111]. In the

Nurses’ Health Study, Schulze et al. observed a pattern high in sugar-sweetened soft

drinks, refined grains, diet soft drinks and processed meat but low in wine, coffee,

cruciferous vegetables, and yellow vegetable; this was related to a two- to threefold

increase in diabetes risk by using inflammatory markers as the response variables [113].

Heidemann et al. found a dietary pattern among Caucasians which was positively

associated with HDL-cholesterol and adiponectin and inversely with HbA1c and CRP in

the EPIC-Potsdam Study. This pattern was characterized by a high intake of fresh fruits

and low intake of high-caloric soft drinks, beer, red meat, poultry, processed meat,

legumes and bread (except wholegrain bread) and was inversely associated with type 2

diabetes risk [110].

The inverse association of red meat with the pattern score that was positively associated

with type 2 diabetes in this Ghanaian study population is an important difference to dietary

patterns derived among Western populations. Although not all previous meta-analyses

found a positive association between red meat intake and type 2 diabetes [200], there is a

large amount of epidemiological evidence for this positive association among Western

populations [201, 202]. Possibly, the different types and preparation methods of red meat

in urban Ghana compared to those among Western populations partially explain the

inverse association with type 2 diabetes in our study. With regard to plantain – a major

staple food in Ghana – we observed the highest contribution to the dietary pattern score

and a positive association with type 2 diabetes. Plantain features a high glycemic index,

and the content of simple sugars increases continuously during the ripening process

[203]. Evidence from large epidemiological studies showed, that the glycemic index and

the glycemic load are associated with a higher risk of type 2 diabetes [204]. Furthermore,

frequent intake of carbohydrates has been related to an increase of fasting triglyceride

concentrations and a reduction of HDL-cholesterol [205-207]. As for cassava, it seemed

contra-intuitive that its intake was positively associated with the pattern score, but

inversely with type 2 diabetes. However, an inverse association between cassava flour

and incident diabetes was also observed in a Brazilian study [208]. In this Ghanaian study

population, plantain and cassava were frequently consumed and the preparation methods

were diverse including cooking, frying and pounding. Lacking a plausible biological

explanation, novel methods for the preparation of the traditional foods may explain the

different associations of plantain and cassava with type 2 diabetes.

DISCUSSION 84

4.2 Discussion of methods

4.2.1 Study design and study population

For the present work an unmatched hospital-based case-control study was used to

evaluate the anthropometric measures and the nutritional behavior and their associations

with type 2 diabetes in an urban Ghanaian population. This study design is well suited for

rare diseases or as a preliminary study where little is known about the association

between the risk factor and the disease of interest. Case-control studies are comparatively

quick, relatively inexpensive and easy to implement. In case-control studies, exposures

and outcome are determined simultaneously and, thus, provide no information about the

chronology of the exposure and the disease. Therefore, the present study cannot

investigate the temporal relationship of the associations between anthropometric

measures and dietary patterns with type 2 diabetes. The presence of reverse causation

cannot be excluded in the present work. Still, in the context of scarce epidemiological data

from SSA, this case-control study is useful to establish hypotheses on the associations

between anthropometric measures and dietary patterns with type 2 diabetes in SSA.

Case-control studies are prone to bias and confounding. To minimize such bias, care

must be taken in the selection of cases and controls. All people in the source population

who develop the disease of interest are presumed to be included as cases in a case-

control study [209]. In addition, controls should be selected from the same source

population, from which the cases were drawn and the selection should be independently

of their exposure status [209]. In the KDH study cases were recruited from the diabetes

center and the hypertension clinic at the KATH. The preliminary controls were selected

from the outpatient department, hospital staff and friends, neighbors and community

members. Clearly, the hospital-based selection of controls helped to make the comparison

groups more similar in terms of potential confounders, which are difficult to measure, such

as socioeconomic background. However, it is possible that the hospital-based selection of

controls has led to a comparison population supposedly heterogeneous and not fully

representative of the source population from which the cases were drawn from. While this

could have masked an association between BMI and type 2 diabetes in this study, it

seems implausible that it might also explain the higher risk observed with higher WHR.

Thus, selection bias seems an unlikely explanation for the stronger effect of WHR

compared to BMI observed in this study. With respect to the findings of the nutritional

behavior, the selection of hospital staff may lead to a control group that is younger, and

possess a higher socio-economic position through better education and financial

capability leading to food choices from the “purchase” pattern compared to the rest of the

DISCUSSION 85

controls. This selection bias would rather overestimate the associations between the

“purchase” and also the “traditional” dietary pattern with type 2 diabetes in this population.

4.2.2 Data quality

The retrospective determination of exposures by self-report or interview bears the risk of

interviewer or recall bias. Interviewer bias arises when the interviewer consciously or

unconsciously acquires inaccurate information from study subjects [210]. However,

standardized interviews by trained nurses of the same cultural background and language

are expected to keep such information bias to a minimum. Recall bias occur when cases

and controls recall exposures differently. After the diagnosis, cases may spend more time

thinking about reasons for their disease and may be more likely to remember exposures

more readily than controls. Recall bias is not problematic for the anthropometric

measures, because they were objectively measured, but it could have biased the

associations between the dietary patterns and type 2 diabetes.

Exposure assessments

Anthropometric measures such as BMI and WC are easy and relatively inexpensive to

measure, are associated with different health outcomes and are therefore useful in the

clinical assessment of disease status [211]. BMI and WC differ in body composition: While

BMI reflects lean and total fat mass but not fat distribution (general obesity), WC

represents total and abdominal fat (central obesity) and both are highly correlated with

each other [212]. Ratios such as WHR and WHtR that represent also central obesity are

not so strongly correlated with BMI and are therefore alternatives to WC. Nevertheless,

ratios are more prone to measurement error because they require two measurements.

Still, all anthropometric measurements were carried out under standardized procedures by

trained hospital stuff in this study, which is expected to reduce the impact of measurement

error.

With respect to the assessment of the nutritional behavior, a culturally sensitive 24HDR

and a FFQ were applied. 24HDR are useful to describe the short-term and current diet

and can be applied easily and quickly [213]. A single 24HDR is useful to describe the

average dietary intake in a population; however multiple recalls are needed to describe

the usual dietary intake [214]. The general limitation of 24HDR, i.e. forgotten foods (recall

bias) and under - or over-reporting of food items, seem less problematic in SSA than in

Europe or North America [215]. Furthermore, public knowledge about associations

DISCUSSION 86

between specific food items and type 2 diabetes may not be pronounced in the study

area. Only since 2012, a national policy for the prevention and control of chronic non-

communicable diseases exists in Ghana, including health promotions for a healthy diet

[216]. Thus, at the time of study conduct, nutritional counseling was not part of the routine

diabetes management at the hospital and thus, does not apply to this study population. As

a consequence, awareness of type 2 diabetes is not expected to be accompanied by a

change in the nutritional behavior and also reporting bias in noting of specific foods seems

unlikely in this population. Specifically, under-reporting might have occurred in participants

with diabetes and/or overweight, who may have tended to give socially desirable answers,

leading to biased dietary data and biased associations with type 2 diabetes. Nevertheless,

in the SSA region, obesity is however perceived as a marker of affluence [1] and is linked

to health and beauty in Ghanaian women [38]. Furthermore, the determinants of food

choice are less influenced by social desirability than more by convenience, availability and

price [217]. Thus, under-reporting may not occur in this study population.

24HDR perform well in regions with high rates of illiteracy and of low SES, particularly

when applied by interviewers of the same cultural background speaking the local

language. Clearly, the inter- and intra-individual variance of a 24HDR limits information on

the actual, individual diet. However, 24HDR are useful to compare energy and nutrient

intakes between population groups [218], specifically when applying local household

measures and food composition tables. In contrast to 24HDR, FFQ depict the long-term

usual diet and are useful to rank participants in accordance to their dietary intake [214].

FFQs are commonly used in epidemiological studies, because they are easy, quick and

cost-efficient [214]. As with all retrospective assessment methods, FFQ bears the risk of

recall bias and under/overestimation of portion sizes; their quantitative precision may be

limited. The application of a locally specific FFQ by trained nurses of the same cultural

background and language helped to keep these information biases to a minimum.

Nevertheless, they are substantially cheaper than 24HDR [218] and feasible to measure

nutrition exposition in case control or cross-sectional studies. If the food list is culturally

sensitive as in the present study, FFQ exhibit excellent properties to assess nutritional

behavior in this study setting. However, the FFQ of the KDH study has yet to be validated.

The assessment of SES in a resource-limited country, such as Ghana, is difficult due the

lack of standardized economic data of income and tax. The use of education, occupation

and literacy as proxy markers for SES is also common among other studies from SSA [4,

36, 219-221]. However, the use of single indicators bears the risk of residual confounding.

To minimize such confounding, the SES sum score were developed in the KDH study in

DISCUSSION 87

2013 [140] to improve the description of SES in this Ghanaian population. A systematic

review and meta-analysis reported that a low socio-economic position (measured by

educational level, occupation and income) was associated with an increased risk of type 2

diabetes in high-, middle- and low-income countries [222]. While the associations were

consistent in high-income countries, available data from middle- and low-income countries

are limited. Thus, education, occupation, literacy and the SES sum score were used as

important confounder in logistic regression analyses in this thesis.

In the present study population of low SES, the obesity measures and dietary patterns

were strongly associated with proxy markers of SES (education, literacy and

unemployment). Surprisingly, the strength of associations only slightly attenuated after

adjusting for these factors. Also, adjustment of the SES sum score instead of the proxy

markers did not change the strong associations. Thus, there is the possibility that SES is

imperfectly measured in the present study, and that residual confounding by SES may

partly explain the observed associations.

With respect to the reliability of the biomarkers, blood samples were collected from all

study participants after 10 hours of fasting and biomarker measurements were done under

standardized procedures. It is possible, that the long-term storage and blood processing

may have affected biomarker concentrations. With respect to the case-control design, it is

possible that the concentrations of selected biomarkers (triglycerides, HDL-cholesterol

and adiponectin) chosen as response variables in the RRR might have changed during

the course of diabetes. Although the investigation of this issue is not possible among the

diabetes cases, we assessed whether the positive associations of the dietary pattern

score with FPG concentrations and HOMA-IR remained in the apparently healthy control

group. Indeed, FPG concentrations increased across quintiles with the exception of the

highest quintile, while this association was less clear for HOMA-IR.

Outcome assessment

The definition of type 2 diabetes was based on one fasting glucose measurement ≥ 126

mg/dl and documented anti-diabetic medication according to the WHO classification [135].

Compared to the use of an oral-glucose-tolerance test or HbA1C, this definition is sub-

optimal and might have overestimated type 2 diabetes cases. However, of the 542 cases,

97% have already been known to the hospital for years (mean time since diagnosis, 6.5 ±

5.8 years) and had a well-documented medical history. The majority were on metformin-

based medication or on combination therapy with sulphonylureas. Only 3.7% of cases

were on insulin monotherapy [130]. Furthermore, the combination of a single

measurement with medication as definition criteria follows the general practice in a

DISCUSSION 88

resource-poor setting like Ghana. The practicability and interpretation of using HbA1c for

the diagnosis of type 2 diabetes in SSA is questionable, because of high costs and high

prevalence of hemoglobinopathies [12].

4.2.3 Statistical methods

ROC-curve analysis

ROC-curve analysis is often used to assess the discriminatory ability of diagnostic or

screening markers. It is a graphical plot of sensitivity vs. false-positive rate (1-specificity)

for all possible cut-off values. The ROC-AUC is an objective measure to quantify the

accuracy of a diagnostic test and has the advantage that is independent of specific cut-off

values and the prevalence of disease [223]. This analysis allows the comparison between

different diagnostic tests. ROC curve analysis is also useful to evaluate the optimal cut-off

point to be used in clinical practice. The determination of the optimal cut-off depends on

the intended use of this cut-off: For clinical practice (e.g. diabetes prevention program) a

high sensitivity would be necessary to identify all diseased persons, whereas criteria for a

potential harmful treatment to prevent diabetes may need a high specificity. However, in

this study, the optimal cut-offs for obesity measures were identified by the use of the

Youden index that maximize sensitivity and specificity [160]. In this approach, equal

weight is given to sensitivity and specificity.

Dietary pattern analyses

In this thesis, an exploratory factor analysis and the RRR approach were applied to

identify dietary patterns in an urban Ghanaian population. In contrast to dietary indexes

(“a priori”) that were originally created and tested among Caucasian populations [91, 92]

and thus are not applicable for this Ghanaian population, “a posteriori” methods identify

combinations of foods and drinks as they are consumed in reality in a specific population

[224].

The factor analysis is a useful method to describe dietary patterns in a study population

independent of their relevance to health outcomes. It identifies dietary patterns relying on

the combined consumption of foods and drinks only and aggregates food items or food

groups based on the degree of their inter-correlations [88]. The factor analysis contains

several arbitrary but important choices: First, the investigator has to decide about

collapsing food items into food groups for entering into the analysis. It is possible that the

choices of food grouping may affect the patterns derived. In the present thesis, different

food groups were created for fruits and vegetables in sensitivity analysis (Table S8-9,

appendix), however the identified dietary patterns did not change markedly. It is also

DISCUSSION 89

clear, that a smaller number of input variables explain a higher amount of variance in

dietary intake compared with a larger number of input variables [94]. The next subjective

decision is the number of factors to be retained [225]. This decision should be based on

different criteria such as scree plot, eigenvalues and the interpretability of the patterns.

The use of cut points for eigenvalues and factor loadings, the method of rotation, and

even the labelling of the derived dietary patterns are additional important decisions. In the

present study the optimal number of retained dietary patterns was based on the criteria of

an eigenvalue ≥ 1, the scree plot and plausibility of the patterns; food items with factor

loadings of ≥ 0.35 were defined as the major contributors of the patterns. However, these

criteria are usually used in dietary pattern analysis [94]. In sensitivity analyses further

dietary pattern solutions (3–5 factors) were examined (Table S5-7, appendix), but did not

reveal meaningful dietary patterns. Exploratory dietary patterns account commonly only

for a small or moderate proportion of total variance in the diet, suggesting the existence of

other dietary patterns that could be more important in the prediction of a specific disease

[93]. Nevertheless, the two dietary patterns identified by exploratory factor analysis were

strongly associated with type 2 diabetes in this study population. In addition, the two

dietary patterns included food items that were also in the RRR-derived dietary pattern,

which is more predictive of a disease because of the use of diabetes related biomarkers.

Finally, Hu et al. reported that dietary patterns derived by factor analysis were valid and

reproducible over time and across different dietary assessment methods in the Health

Professionals Follow-up Study [226].

In contrast to factor analysis, RRR is a useful tool to derive health-related dietary patterns.

The main advantage of the RRR method is that it combines the strengths of the

exploratory and the hypothesis-oriented approach. It uses the dietary information of the

study population and the response variables (biomarkers) selected based on prior

scientific knowledge about the associations with the disease of interest and thus is not

purely data-driven such as factor analysis. The RRR is similar to factor analysis in its

mathematical foundation and technique to derive factors. The calculation of pattern scores

in both methods are based on the determination of eigenvalues and corresponding

eigenvectors of the covariance matrix of predictors (factor analysis) and responses (RRR)

[227]. However, the aims of both methods are different: While factor analysis aims to

identify linear combinations of predictors (food items) by maximizing the explained

variation in all predictor variables, RRR identifies linear functions of predictors (food items)

that explain a maximum of variation in the disease-related response variables

(biomarkers) [95].

DISCUSSION 90

The RRR is useful to identify biological pathways by which dietary intake may affect the

diabetes risk. Another advantage is that the use of response variables can support the

interpretation of the observed association between dietary patterns and the disease [90].

This method has also disadvantages: It is limited to existing studies with biomarkers or

intermediate variables and requires prior knowledge of the diet-disease association [93].

RRR focuses on response variables associated with a specific pathway from diet to

disease, therefore, other potential important pathways are not taken into consideration.

The selected biomarkers in RRR of this study are similar to those in studies among

Caucasian populations [110]. However, the selection of biomarkers was limited to

adiponectin, HDL-cholesterol and triglycerides, because they are all affected by the diet

[146, 147, 151, 152] and linked to the pathophysiology of type 2 diabetes [150, 155, 156].

Inflammatory markers, such as CRP, were not considered as response variables in this

analysis due to the high prevalence of infectious diseases, complicating the interpretation

of CRP as a risk factor for diabetes. Indeed, 13% of this Ghanaian study population had a

Plasmodium falciparum infection [129]. HOMA-IR was also disregarded as a response

variable because of its metabolic proximity to type 2 diabetes. HOMA-IR is useful to

determine pre-diabetic stages, but is not on the causal pathway for type 2 diabetes.

Finally, the reproducibility of the associations between RRR-derived dietary patterns and

type 2 diabetes needed to be confirmed in other populations. However, the application of

different dietary questionnaires, different food grouping and especially different dietary

habits among populations could make this investigation difficult [115].

PUBLIC HEALTH RELEVANCE 91

4.3 Public health relevance

The increasing burden of type 2 diabetes poses a major public health challenge in SSA,

where scarce financial resources causes a major barrier to adequate diabetes care

delivery and management of type 2 diabetes [3]. There is an urgent need to limit and

reduce the increasing diabetes prevalence in SSA and thus to prevent potentially harmful

and costly complications. Therefore, epidemiological surveillance and the development of

effective and sustainable primary and secondary prevention are necessary to tackle

SSA’s chronic disease epidemic [228]. Primary prevention of diabetes by lifestyle

interventions including weight loss, diet and exercise offer excellent opportunities to

reduce the burden of type 2 diabetes [229]. Secondary prevention is important to prevent

severe diabetes complications through the optimization of glycemic control and treatment

of coexisting risk factors and thus improve quality of life [230].

The high prevalence of overweight and obesity in this study indicates the need for

appropriate interventions for its prevention and treatment. This thesis highlights

particularly the potential importance of central obesity for the diabetes prevention in SSA.

With respect to public health interventions, women and men may equally benefit from the

prevention and reduction of central obesity in this region. In SSA, obesity is perceived

positively, as a sign of affluence, good health and beauty [1, 38]. Therefore, the

awareness of obesity as an important risk factor for type 2 diabetes must be enhanced in

the public to avert the ongoing burden of type 2 diabetes in SSA. Furthermore, in

resource-limited countries, such as Ghana, the use of WHR and WC as simple measures

of obesity is an economical method to identify individuals with an increased risk for

diabetes.

Another promising way to reduce the diabetes burden in SSA is to modify the nutritional

behavior. Therefore, identification of dietary patterns in SSA could be important for public

health implications, because dietary patterns are easy to understand for the public and

translate into practice [88]. Health promotion activities in Ghana should increase the

awareness about a healthy diet and dietary recommendations should include more

consumption of fruits and vegetables and less consumption of starchy, energy dense and

fatty foods. Furthermore, in this thesis a dietary pattern was identified by applying the

RRR approach that was associated with a diabetogenic risk marker profile and thus gives

information about the pathophysiological pathway. Triglycerides were the main drivers of

the association between the RRR-derived dietary pattern and type 2 diabetes. Although

widespread diabetes screening in the general population cannot be encouraged in SSA,

targeted screening to identify individuals with high-risk characteristics through the

PUBLIC HEALTH RELEVANCE 92

assessment of central obesity measures and blood lipid concentrations should be

undertaken in Ghana. The reduction of central obesity and triglyceride concentrations may

help to prevent the development of type 2 diabetes and diabetes related complications.

Furthermore, a nationwide effective and sustainable diabetic education program should be

developed and implemented at hospital and community settings in Ghana to increase the

knowledge about diabetes. This program should include a simple and easy-to-understand

approach to reach also persons with lower educational level. At the moment, the existing

national diabetes program in Ghana is currently in the process of enhancing by the

National Diabetes Management and Research Centre of Ghana’s Ministry of Health

through a partnership with the World Diabetes Foundation (project funded from 2009-

2015) [231]. The aim of this project is to improve the prevention and care for diabetes and

related NCD throughout Ghana by establishing an NCD program including training of

health professionals and health promotion activities [231]. A well-structured training of

health personnel in diabetes management, prevention and control could enhance the

knowledge about the disease in diabetes patients and thus may reduce severe diabetes

complications. Additionally, health promotion activities should be implemented in schools,

churches, mosques and other community gathering places to increase the awareness of

diabetes, its risk factors and the prevention of diabetes among the general population.

Furthermore, circulation of information about type 2 diabetes in Ghana could be

emphasized by using mass media via radio, television and newspapers.

Finally, the application of findings of this thesis could make a major, rapid, and cost-

effective contribution to the prevention and control of the diabetes epidemic in SSA.

CONCLUSION AND FURTHER PERSPECTIVES 93

5 CONCLUSION AND FURTHER PERSPECTIVES

Overall, the present thesis adds knowledge about two important modifiable risk factors for

type 2 diabetes in SSA - obesity and the nutritional behavior.

The following conclusions can be drawn from the observed results among this Ghanaian

population:

• Findings of this work indicate that central obesity (with WHR as strongest

predictor), rather than general obesity measures are associated with an increased

risk for type 2 diabetes among women and men. Thus, body fat distribution seems

to play an important role in the diabetes development in SSA.

• The current recommended cut-off points for obesity measures are inappropriate to

assess diabetes risk in this population. Country- or region-specific cut-off points for

anthropometric measures could be useful to identify individuals with type 2

diabetes in SSA populations.

• Dietary patterns are associated with type 2 diabetes in SSA. The “purchase”

dietary pattern, characterized by a high consumption of sweets, red meat, fruits

and vegetables and low consumption of plantain is related to a decreased risk for

type 2 diabetes. The “traditional” dietary pattern, characterized by a high intake of

plantain, green leafy vegetables, fish, fermented maize products, and palm oil is

associated with an increased risk for type 2 diabetes.

• Adherence to traditional food items and low preference for purchased foods relate

to increased serum triglycerides and decreased HDL-cholesterol, both risk factors

for type 2 diabetes, and as a consequence may increase the risk for type 2

diabetes

Given the limited studies from SSA, the present thesis evaluated for the first time the

potential importance of obesity and dietary patterns for the development of type 2

diabetes in a SSA population. This study established hypotheses on the relationship

between various anthropometric measures and type 2 diabetes in an urban Ghanaian

population. Nevertheless, prospective studies are needed to provide stronger evidence for

the associations between obesity measures, fat distribution and type 2 diabetes risk

among people from SSA. Furthermore, it remains to be examined whether preventive

strategies against type 2 diabetes should take into account WHR in addition to the

conventional measure of BMI. The results of this study also support the urgent need to

evaluate country or region-specific cut-off points for anthropometric measures to identify

CONCLUSION AND FURTHER PERSPECTIVES 94

individuals with type 2 diabetes in prospective studies in SSA. These studies could be

important, because if the evidence for ethnic differences in cut-off points for BMI, WC and

WHR as risk predictors is growing, ethnic-specific cut-off points should be recommended

by the public health guidelines to identify individuals with an increased risk for type 2

diabetes or other health outcomes.

In addition, this thesis showed the great potential of dietary pattern analysis in an SSA

population. However, the relationship between dietary patterns and type 2 diabetes are

still unclear in SSA and clearly require further verification in other regions of West Africa.

In addition, the determinants of adherence to dietary patterns in SSA should be examined.

Especially, the detrimental association between the traditional dietary pattern and type 2

diabetes warrant further investigations. Given the limited data from SSA, a case-control

design is useful to establish hypotheses on the relationship between dietary patterns and

type 2 diabetes. However, these hypotheses require verification in prospective studies

from SSA building a clear temporal relationship between dietary patterns and type 2

diabetes. Finally, the reproducibility of the association between the RRR-derived dietary

pattern and type 2 diabetes should be evaluated in independent populations.

REFERENCES 95

6 REFERENCES

1. Mbanya, J.C., et al., Diabetes in sub-Saharan Africa. Lancet, 2010. 375(9733): p. 2254-66.

2. IDF, International Diabetes Federation diabetes atlas, sixth edition, update 2014, available

online: http://www.idf.org/diabetesatlas (accessed 13th January 2015)

3. Mbanya, J.C., et al., Obesity and type 2 diabetes in Sub-Sahara Africa. Curr Diab Rep,

2014. 14(7): p. 501.

4. Ziraba, A.K., J.C. Fotso, and R. Ochako, Overweight and obesity in urban Africa: A

problem of the rich or the poor? BMC Public Health, 2009. 9: p. 465.

5. Marais, B.J., et al., Tuberculosis comorbidity with communicable and non-communicable

diseases: integrating health services and control efforts. Lancet Infect Dis, 2013. 13(5): p.

436-48.

6. Hu, F.B., Globalization of diabetes: the role of diet, lifestyle, and genes. Diabetes Care,

2011. 34(6): p. 1249-57.

7. American Diabetes Association, Diagnosis and classification of diabetes mellitus. Diabetes

Care, 2014. 37 Suppl 1: p. S81-90.

8. Joost, H.G., Pathogenesis, risk assessment and prevention of type 2 diabetes mellitus.

Obes Facts, 2008. 1(3): p. 128-37.

9. International Expert Comittee, International Expert Committee report on the role of the A1C

assay in the diagnosis of diabetes. Diabetes Care, 2009. 32(7): p. 1327-34.

10. American Diabetes Association, Diagnosis and classification of diabetes mellitus. Diabetes

Care, 2010. 33 Suppl 1: p. S62-9.

11. WHO, Use of Glycated Haemoglobin (HbA1c) in the Diagnosis of Diabetes Mellitus, 2011.

available online: http://www.who.int/diabetes/publications/diagnosis_diabetes2011/en/

(accessed 13th January 2015)

12. Smaldone, A., Glycemic control and Hemoglobinopathy: When A1C may not be reliable.

Diabetes spectrum, 2008. Volume 21, Number 1: p. 46-49.

13. Hall, V., et al., Diabetes in Sub Saharan Africa 1999-2011: epidemiology and public health

implications. A systematic review. BMC Public Health, 2011. 11: p. 564.

14. Dodu, S.R., The incidence of diabetes mellitus in Accra (Ghana); a study of 4,000 patients.

West Afr Med J, 1958. 7(3): p. 129-34.

15. Amoah, A.G., S.K. Owusu, and S. Adjei, Diabetes in Ghana: a community based

prevalence study in Greater Accra. Diabetes Res Clin Pract, 2002. 56(3): p. 197-205.

16. WHO, Obesity and overweight factsheets, 2015. available online:

http://www.who.int/mediacentre/factsheets/fs311/en/ (accessed 13th January 2015)

17. Hajer, G.R., T.W. van Haeften, and F.L. Visseren, Adipose tissue dysfunction in obesity,

diabetes, and vascular diseases. Eur Heart J, 2008. 29(24): p. 2959-71.

18. Haffner, S.M., Abdominal adiposity and cardiometabolic risk: do we have all the answers?

Am J Med, 2007. 120(9 Suppl 1): p. S10-6; discussion S16-7.

19. WHO, BMI Classification: Global Database on Body Mass Index, 2006. available online:

http://apps.who.int/bmi/index.jsp?introPage=intro_3.html. (accessed 13th January 2015)

20. Vazquez, G., et al., Comparison of body mass index, waist circumference, and waist/hip

ratio in predicting incident diabetes: a meta-analysis. Epidemiol Rev, 2007. 29: p. 115-28.

21. Rexrode, K.M., et al., Abdominal adiposity and coronary heart disease in women. JAMA,

1998. 280(21): p. 1843-8.

22. Lean, M.E., T.S. Han, and C.E. Morrison, Waist circumference as a measure for indicating

need for weight management. BMJ, 1995. 311(6998): p. 158-61.

23. IDF, The IDF consensus worldwide definition of the metabolic syndrome, 2006. available

online: http://www.idf.org/webdata/docs/IDF_Meta_def_final.pdf. (accessed 13th January

2015)

REFERENCES 96

24. WHO, waist circumference and waist-hip ratio: report of a WHO expert consultation,

Geneva,8-11 December 2008; 2011. available online:

http://whqlibdoc.who.int/publications/2011/9789241501491_eng.pdf. (accessed 13th

January 2015)

25. National Institutes of Health, Clinical Guidelines on the Identification, Evaluation, and

Treatment of Overweight and Obesity in Adults--The Evidence Report. . Obes Res, 1998. 6

Suppl 2: p. 51S-209S.

26. US Department of Agriculture, US Department of Health and Human Services, Dietary

guidelines for Americans. Washington, DC: US Department of Agriculture, Publication; pp.

261–495.pp. 20124. 1990.

27. Khan, N.A., et al., The 2009 Canadian Hypertension Education Program recommendations

for the management of hypertension: Part 2--therapy. Can J Cardiol, 2009. 25(5): p. 287-

98.

28. Graham, I., et al., European guidelines on cardiovascular disease prevention in clinical

practice: executive summary. Atherosclerosis, 2007. 194(1): p. 1-45.

29. Examination Committee of Criteria for 'Obesity Disease' in Japan, Japan Society for the

Study of Obesity, New criteria for 'obesity disease' in Japan. Circ J, 2002. 66(11): p. 987-

92.

30. Zhou, B.F. and Cooperative Meta-Analysis Group of the Working Group on Obesity in

China, Predictive values of body mass index and waist circumference for risk factors of

certain related diseases in Chinese adults--study on optimal cut-off points of body mass

index and waist circumference in Chinese adults. Biomed Environ Sci, 2002. 15(1): p. 83-

96.

31. Abubakari, A.R. and R.S. Bhopal, Systematic review on the prevalence of diabetes,

overweight/obesity and physical inactivity in Ghanaians and Nigerians. Public Health,

2008. 122(2): p. 173-82.

32. Tuakli-Wosornu, Y.A., M. Rowan, and J. Gittelsohn, Perceptions of physical activity,

activity preferences and health among a group of adult women in urban Ghana: a pilot

study. Ghana Med J, 2014. 48(1): p. 3-13.

33. Ng, M., et al., Global, regional, and national prevalence of overweight and obesity in

children and adults during 1980-2013: a systematic analysis for the Global Burden of

Disease Study 2013. Lancet, 2014. 384(9945): p. 766-81.

34. Mogre, V., R. Nyaba, and S. Aleyira, Lifestyle risk factors of general and abdominal obesity

in students of the school of medicine and health science of the university of development

studies, tamale, ghana. ISRN Obes, 2014. 2014: p. 508382.

35. Cohen, A.K., et al., Educational attainment and obesity: a systematic review. Obes Rev,

2013. 14(12): p. 989-1005.

36. Sodjinou, R., et al., Obesity and cardio-metabolic risk factors in urban adults of Benin:

relationship with socio-economic status, urbanisation, and lifestyle patterns. BMC Public

Health, 2008. 8: p. 84.

37. Fezeu, L., et al., Association between socioeconomic status and adiposity in urban

Cameroon. Int J Epidemiol, 2006. 35(1): p. 105-11.

38. Amoah, A.G., Sociodemographic variations in obesity among Ghanaian adults. Public

Health Nutr, 2003. 6(8): p. 751-7.

39. Huxley, R., et al., Ethnic comparisons of the cross-sectional relationships between

measures of body size with diabetes and hypertension. Obes Rev, 2008. 9 Suppl 1: p. 53-

61.

40. MacKay, M.F., et al., Prediction of type 2 diabetes using alternate anthropometric

measures in a multi-ethnic cohort: the insulin resistance atherosclerosis study. Diabetes

Care, 2009. 32(5): p. 956-8.

41. Nyamdorj, R., et al., BMI compared with central obesity indicators in relation to diabetes

and hypertension in Asians. Obesity (Silver Spring), 2008. 16(7): p. 1622-35.

REFERENCES 97

42. The Decoda Study Group, BMI compared with central obesity indicators in relation to

diabetes and hypertension in Asians. Obesity (Silver Spring), 2008. 16(7): p. 1622-35.

43. Lee, C.M.Y., et al., Indices of abdominal obesity are better discriminators of cardiovascular

risk factors than BMI: a meta-analysis. Journal of Clinical Epidemiology, 2008. 61(7): p.

646-653.

44. Balde, N.M., et al., Diabetes and impaired fasting glucose in rural and urban populations in

Futa Jallon (Guinea): prevalence and associated risk factors. Diabetes Metab, 2007. 33(2):

p. 114-20.

45. Motala, A.A., et al., Diabetes and other disorders of glycemia in a rural South African

community: prevalence and associated risk factors. Diabetes Care, 2008. 31(9): p. 1783-

88.

46. Mbanya, V.N., et al., Body mass index, waist circumference, hip circumference, waist-hip-

ratio and waist-height-ratio: Which is the better discriminator of prevalent screen-detected

diabetes in a Cameroonian population? Diabetes Research and Clinical Practice, 2015.

108(1): p. 23-30.

47. Wang, Y., et al., Comparison of abdominal adiposity and overall obesity in predicting risk of

type 2 diabetes among men. Am J Clin Nutr, 2005. 81(3): p. 555-63.

48. Meisinger, C., et al., Body fat distribution and risk of type 2 diabetes in the general

population: are there differences between men and women? The MONICA/KORA

Augsburg cohort study. Am J Clin Nutr, 2006. 84(3): p. 483-9.

49. Huerta, J.M., et al., Risk of type 2 diabetes according to traditional and emerging

anthropometric indices in Spain, a Mediterranean country with high prevalence of obesity:

results from a large-scale prospective cohort study. BMC Endocr Disord, 2013. 13: p. 7.

50. Qiao, Q. and R. Nyamdorj, The optimal cutoff values and their performance of waist

circumference and waist-to-hip ratio for diagnosing type II diabetes. Eur J Clin Nutr, 2010.

64(1): p. 23-9.

51. Calle, E.E., et al., Body-mass index and mortality in a prospective cohort of U.S. adults. N

Engl J Med, 1999. 341(15): p. 1097-105.

52. Misra, A., J.S. Wasir, and N.K. Vikram, Waist circumference criteria for the diagnosis of

abdominal obesity are not applicable uniformly to all populations and ethnic groups.

Nutrition, 2005. 21(9): p. 969-76.

53. Lear, S.A., et al., The use of BMI and waist circumference as surrogates of body fat differs

by ethnicity. Obesity (Silver Spring), 2007. 15(11): p. 2817-24.

54. Lear, S.A., et al., Appropriateness of waist circumference and waist-to-hip ratio cutoffs for

different ethnic groups. Eur J Clin Nutr, 2010. 64(1): p. 42-61.

55. Nishida, C., G.T. Ko, and S. Kumanyika, Body fat distribution and noncommunicable

diseases in populations: overview of the 2008 WHO Expert Consultation on Waist

Circumference and Waist-Hip Ratio. Eur J Clin Nutr, 2010. 64(1): p. 2-5.

56. Yoon, K.H., et al., Epidemic obesity and type 2 diabetes in Asia. Lancet, 2006. 368(9548):

p. 1681-8.

57. WHO Expert Consultation, Appropriate body-mass index for Asian populations and its

implications for policy and intervention strategies. Lancet, 2004. 363(9403): p. 157-63.

58. Katzmarzyk, P.T., et al., Ethnic-specific BMI and waist circumference thresholds. Obesity

(Silver Spring), 2011. 19(6): p. 1272-8.

59. Chiu, M., et al., Deriving ethnic-specific BMI cutoff points for assessing diabetes risk.

Diabetes Care, 2011. 34(8): p. 1741-8.

60. Huxley, R., et al., Waist circumference thresholds provide an accurate and widely

applicable method for the discrimination of diabetes. Diabetes Care, 2007. 30(12): p. 3116-

61. Gurrici, S., et al., Relationship between body fat and body mass index: differences

between Indonesians and Dutch Caucasians. Eur J Clin Nutr, 1998. 52(11): p. 779-83.

REFERENCES 98

62. Lear, S.A., et al., Visceral adipose tissue accumulation differs according to ethnic

background: results of the Multicultural Community Health Assessment Trial (M-CHAT).

Am J Clin Nutr, 2007. 86(2): p. 353-9.

63. Song, M.Y., et al., Prepubertal Asians have less limb skeletal muscle. J Appl Physiol

(1985), 2002. 92(6): p. 2285-91.

64. Katzmarzyk, P.T., et al., Racial differences in abdominal depot-specific adiposity in white

and African American adults. Am J Clin Nutr, 2010. 91(1): p. 7-15.

65. Wagner, D.R. and V.H. Heyward, Measures of body composition in blacks and whites: a

comparative review. Am J Clin Nutr, 2000. 71(6): p. 1392-402.

66. Okosun, I.S., et al., Abdominal adiposity values associated with established body mass

indexes in white, black and hispanic Americans. A study from the Third National Health

and Nutrition Examination Survey. Int J Obes Relat Metab Disord, 2000. 24(10): p. 1279-

85.

67. Okosun, I.S., et al., Predictive values of waist circumference for dyslipidemia, type 2

diabetes and hypertension in overweight White, Black, and Hispanic American adults. J

Clin Epidemiol, 2000. 53(4): p. 401-8.

68. Okosun, I.S., et al., Predictive value of abdominal obesity cut-off points for hypertension in

blacks from west African and Caribbean island nations. Int J Obes Relat Metab Disord,

2000. 24(2): p. 180-6.

69. Motala, A.A., et al., The prevalence of metabolic syndrome and determination of the

optimal waist circumference cutoff points in a rural South african community. Diabetes

Care, 2011. 34(4): p. 1032-7.

70. Crowther, N.J. and S.A. Norris, The current waist circumference cut point used for the

diagnosis of metabolic syndrome in sub-Saharan African women is not appropriate. PLoS

One, 2012. 7(11): p. e48883.

71. Fisch, A., et al., Prevalence and risk factors of diabetes mellitus in the rural region of Mali

(West Africa): a practical approach. Diabetologia, 1987. 30(11): p. 859-62.

72. Aspray, T.J., et al., Rural and urban differences in diabetes prevalence in Tanzania: the

role of obesity, physical inactivity and urban living. Trans R Soc Trop Med Hyg, 2000.

94(6): p. 637-44.

73. Giday, A., Hypertension, obesity and central obesity in diabetics and non diabetics in

Southern Ethiopia; Ethiopian Journal of Health Development, 2010. 24(2): p. 145-147;

available online: http://www.ajol.info/index.php/ejhd/article/viewFile/62964/50859.

(accessed 13th January 2015)

74. WHO, United Nations Children’s Fund, The World Bank. UNICEF-WHO-World Bank Joint

Child Malnutrition Estimates. (UNICEF, New York; WHO, Geneva; The World Bank,

Washington, DC; 2012). available online:

http://www.who.int/nutgrowthdb/jme_unicef_who_wb.pdf. (accessed 13th January 2015)

75. Abrahams, Z., Z. McHiza, and N.P. Steyn, Diet and mortality rates in Sub-Saharan Africa:

stages in the nutrition transition. BMC Public Health, 2011. 11: p. 801.

76. Bosu, W.K., An overview of the nutrition transition in West Africa: implications for non-

communicable diseases. Proc Nutr Soc, 2014: p. 1-12.

77. Amuna, P. and F.B. Zotor, Epidemiological and nutrition transition in developing countries:

impact on human health and development. Proc Nutr Soc, 2008. 67(1): p. 82-90.

78. Becquey, E., et al., Dietary patterns of adults living in Ouagadougou and their association

with overweight. Nutr J, 2010. 9: p. 13.

79. Nti, C.A., Household dietary practices and family nutritional status in rural Ghana. Nutr Res

Pract, 2008. 2(1): p. 35-40.

80. FAO, Food and Agriculture Organization, Ghana Nutrition Profile – Nutrition and Consumer

Protection Division, 2009. available online: ftp://ftp.fao.org/ag/agn/nutrition/ncp/gha.pdf.

(accessed 13th January 2015)

81. Barichella, M., et al., Nutritional status and dietary habits in Parkinson's disease patients in

Ghana. Nutrition, 2013. 29(2): p. 470-3.

REFERENCES 99

82. FAO, Food and Agriculture Organization of the United Nations Statistics Division: Food

Balance Fact Sheets, 2014. available online: http://faostat3.fao.org/download/FB/*/E

(accessed 13th January 2015)

83. Cappuccio, F.P., et al., A community programme to reduce salt intake and blood pressure

in Ghana [ISRCTN88789643]. BMC Public Health, 2006. 6: p. 13.

84. Kerry, S.M., et al., Rural and semi-urban differences in salt intake, and its dietary sources,

in Ashanti, West Africa. Ethn Dis, 2005. 15(1): p. 33-9.

85. Clausen, T., et al., Diverse alcohol drinking patterns in 20 African countries. Addiction,

2009. 104(7): p. 1147-54.

86. Martinez, P., et al., Alcohol abstinence and drinking among African women: data from the

World Health Surveys. BMC Public Health, 2011. 11: p. 160.

87. Minicuci, N., et al., Sociodemographic and socioeconomic patterns of chronic non-

communicable disease among the older adult population in Ghana. Glob Health Action,

2014. 7: p. 21292.

88. Hu, F.B., Dietary pattern analysis: a new direction in nutritional epidemiology. Curr Opin

Lipidol, 2002. 13(1): p. 3-9.

89. Trichopoulos, D. and P. Lagiou, Dietary patterns and mortality. Br J Nutr, 2001. 85(2): p.

133-4.

90. Schulze, M.B. and K. Hoffmann, Methodological approaches to study dietary patterns in

relation to risk of coronary heart disease and stroke. Br J Nutr, 2006. 95(5): p. 860-9.

91. Kennedy, E.T., et al., The Healthy Eating Index: design and applications. J Am Diet Assoc,

1995. 95(10): p. 1103-8.

92. Trichopoulou, A., et al., Adherence to a Mediterranean diet and survival in a Greek

population. N Engl J Med, 2003. 348(26): p. 2599-608.

93. Michels, K.B. and M.B. Schulze, Can dietary patterns help us detect diet-disease

associations? Nutr Res Rev, 2005. 18(2): p. 241-8.

94. Newby, P.K. and K.L. Tucker, Empirically derived eating patterns using factor or cluster

analysis: a review. Nutr Rev, 2004. 62(5): p. 177-203.

95. Hoffmann, K., et al., Application of a new statistical method to derive dietary patterns in

nutritional epidemiology. Am J Epidemiol, 2004. 159(10): p. 935-44.

96. Sodjinou, R., et al., Dietary patterns of urban adults in Benin: relationship with overall diet

quality and socio-demographic characteristics. Eur J Clin Nutr, 2009. 63(2): p. 222-8.

97. Keding, G.B., et al., Dietary patterns and nutritional health of women: the nutrition transition

in rural Tanzania. Food Nutr Bull, 2011. 32(3): p. 218-26.

98. Nkondjock, A. and E. Bizome, Dietary patterns associated with hypertension prevalence in

the Cameroon defence forces. Eur J Clin Nutr, 2010. 64(9): p. 1014-21.

99. Maruapula, S. and K. Chapman-Novakofski, Health and dietary patterns of the elderly in

Botswana. J Nutr Educ Behav, 2007. 39(6): p. 311-9.

100. Bauer, F., et al., Dietary patterns and the risk of type 2 diabetes in overweight and obese

individuals. Eur J Nutr, 2012.

101. Malik, V.S., et al., Dietary patterns during adolescence and risk of type 2 diabetes in

middle-aged women. Diabetes Care, 2012. 35(1): p. 12-8.

102. Erber, E., et al., Dietary patterns and risk for diabetes: the multiethnic cohort. Diabetes

Care, 2010. 33(3): p. 532-8.

103. Montonen, J., et al., Dietary patterns and the incidence of type 2 diabetes. Am J Epidemiol,

2005. 161(3): p. 219-27.

104. van Dam, R.M., et al., Dietary patterns and risk for type 2 diabetes mellitus in U.S. men.

Ann Intern Med, 2002. 136(3): p. 201-9.

105. Hodge, A.M., et al., Dietary patterns and diabetes incidence in the Melbourne Collaborative

Cohort Study. Am J Epidemiol, 2007. 165(6): p. 603-10.

106. Odegaard, A.O., et al., Dietary patterns and incident type 2 diabetes in chinese men and

women: the singapore chinese health study. Diabetes Care, 2011. 34(4): p. 880-5.

REFERENCES 100

107. Yu, R., et al., Relationship between dietary intake and the development of type 2 diabetes

in a Chinese population: the Hong Kong Dietary Survey. Public Health Nutr, 2011. 14(7): p.

1133-41.

108. Nettleton, J.A., et al., Dietary patterns and risk of incident type 2 diabetes in the Multi-

Ethnic Study of Atherosclerosis (MESA). Diabetes Care, 2008. 31(9): p. 1777-82.

109. Fung, T.T., et al., Dietary patterns, meat intake, and the risk of type 2 diabetes in women.

Arch Intern Med, 2004. 164(20): p. 2235-40.

110. Heidemann, C., et al., A dietary pattern protective against type 2 diabetes in the European

Prospective Investigation into Cancer and Nutrition (EPIC)--Potsdam Study cohort.

Diabetologia, 2005. 48(6): p. 1126-34.

111. Liese, A.D., et al., Food intake patterns associated with incident type 2 diabetes: the Insulin

Resistance Atherosclerosis Study. Diabetes Care, 2009. 32(2): p. 263-8.

112. McNaughton, S.A., G.D. Mishra, and E.J. Brunner, Dietary patterns, insulin resistance, and

incidence of type 2 diabetes in the Whitehall II Study. Diabetes Care, 2008. 31(7): p. 1343-

113. Schulze, M.B., et al., Dietary pattern, inflammation, and incidence of type 2 diabetes in

women. Am J Clin Nutr, 2005. 82(3): p. 675-84; quiz 714-5.

114. InterAct Consortium, Adherence to predefined dietary patterns and incident type 2 diabetes

in European populations: EPIC-InterAct Study. Diabetologia, 2014. 57(2): p. 321-33.

115. Imamura, F., et al., Generalizability of dietary patterns associated with incidence of type 2

diabetes mellitus. Am J Clin Nutr, 2009. 90(4): p. 1075-83.

116. Idemyor, V., Diabetes in sub-Saharan Africa: health care perspectives, challenges, and the

economic burden of disease. J Natl Med Assoc, 2010. 102(7): p. 650-3.

117. Levitt, N.S., Diabetes in Africa: epidemiology, management and healthcare challenges.

Heart, 2008. 94(11): p. 1376-82.

118. Mbanya, J.C. and E. Sobngwi, Diabetes in Africa. Diabetes microvascular and

macrovascular disease in Africa. J Cardiovasc Risk, 2003. 10(2): p. 97-102.

119. Burgess, P.I., et al., Epidemiology of diabetic retinopathy and maculopathy in Africa: a

systematic review. Diabet Med, 2013. 30(4): p. 399-412.

120. Akinboboye, O., et al., Trends in coronary artery disease and associated risk factors in

sub-Saharan Africans. J Hum Hypertens, 2003. 17(6): p. 381-7.

121. Kiawi, E., et al., Knowledge, attitudes, and behavior relating to diabetes and its main risk

factors among urban residents in Cameroon: A qualitative survey. Ethnicity & Disease,

2006. 16(2): p. 503-509.

122. Katchunga, P.B., et al., Knowledge of the general population about hypertension and

diabetes mellitus in South Kivu, Democratic Republic of Congo. Revue D Epidemiologie Et

De Sante Publique, 2012. 60(2): p. 141-147.

123. Foma, M.A., et al., Awareness of diabetes mellitus among diabetic patients in the Gambia:

a strong case for health education and promotion. BMC Public Health, 2013. 13: p. 1124.

124. Hjelm, K. and G. Nambozi, Beliefs about health and illness: a comparison between

Ugandan men and women living with Diabetes Mellitus. International Nursing Review,

2008. 55(4): p. 434-441.

125. Mufunda, E., et al., Level and determinants of diabetes knowledge in patients with diabetes

in Zimbabwe: a cross-sectional study. Pan Afr Med J, 2012. 13: p. 78.

126. Awah, P.K., N. Unwin, and P. Phillimore, Cure or control: complying with biomedical

regime of diabetes in Cameroon. BMC Health Serv Res, 2008. 8: p. 43.

127. Ovenseri-Ogbomo, G.O., et al., Knowledge of diabetes and its associated ocular

manifestations by diabetic patients: A study at Korle-Bu Teaching Hospital, Ghana. Niger

Med J, 2013. 54(4): p. 217-23.

128. Qiao, Q. and R. Nyamdorj, Is the association of type II diabetes with waist circumference or

waist-to-hip ratio stronger than that with body mass index? Eur J Clin Nutr, 2010. 64(1): p.

30-4.

REFERENCES 101

129. Danquah, I., G. Bedu-Addo, and F.P. Mockenhaupt, Type 2 diabetes mellitus and

increased risk for malaria infection. Emerg Infect Dis, 2010. 16(10): p. 1601-4.

130. Danquah, I., et al., Diabetes mellitus type 2 in urban Ghana: characteristics and associated

factors. BMC Public Health, 2012. 12: p. 210.

131. The World Bank, 2014. available online: http://data.worldbank.org/country/ghana.

(accessed 13th January 2015)

132. Osam, E.K., An Introduction to the verbal and multi-verbal system of Akan, Proceedings of

the workshop on Multi-Verb Constructions Trondheim Summer School 2003; available

online: http://www.ling.hf.ntnu.no/tross/osam.pdf. (accessed 13th January 2015)

133. Akyeampong, K., The Quality Imperative. Whole school development in Ghana. Paper

commissioned for the EFA Global Monitoring Report 2005; 2004. available online:

http://unesdoc.unesco.org/images/0014/001466/146616e.pdf. (accessed 13th January

2015)

134. National Health Insurance Scheme, available online:http://www.nhis.gov.gh/. (accessed

13th January 2015)

135. WHO, Definition, diagnosis and classification of diabetes mellitus and its complication:

Report of a WHO consultation. Part 1: diagnosis and classification of diabetes mellitus.

Geneva 1999. available online:

http://apps.who.int/iris/bitstream/10665/66040/1/WHO_NCD_NCS_99.2.pdf (accessed 13th

January 2015)

136. Ghana Statistical Service, Population and Housing Census Summary Results of Final

Report, Ghana. 2010. available online: http://www.statsghana.gov.gh/ (accessed 13th

January 2015)

137. Food Research Institute (C.S.I.R), Composition of Foods Commonly Used in Ghana Rome,

Italy: Food and Agricultural Organization of the United Nations. 1975.

138. Food Research Institute (C.S.I.R), Measurements of foods commonly used in Ghana

Rome, Italy: Food and Agricultural Organization of the United Nations. 1975.

139. Stadlmayr, B., et al., West African Food Composition Table; The Food and Agriculture

Organization of the United Nations (FAO); Rome; 2012. available online:

http://www.fao.org/docrep/015/i2698b/i2698b00.pdf. (accessed 13th January 2015)

140. Jannasch, F., Post-hoc Konstruktion eines Messinstruments zur Abbildung des sozio-

ökonomischen Status in einer urbanen ghanaischen Population. Master thesis, 2013.

141. Frank, L.K., et al., A Dietary Pattern Derived by Reduced Rank Regression is Associated

with Type 2 Diabetes in An Urban Ghanaian Population. Nutrients, 2015. 7(7): p. 5497-

514.

142. Ainsworth, B.E., et al., Compendium of physical activities: classification of energy costs of

human physical activities. Med Sci Sports Exerc, 1993. 25(1): p. 71-80.

143. Snounou, G., et al., Identification of the four human malaria parasite species in field

samples by the polymerase chain reaction and detection of a high prevalence of mixed

infections. Mol Biochem Parasitol, 1993. 58(2): p. 283-92.

144. Friedewald, W.T., R.I. Levy, and D.S. Fredrickson, Estimation of the concentration of low-

density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin

Chem, 1972. 18(6): p. 499-502.

145. Matthews, D.R., et al., Homeostasis model assessment: insulin resistance and beta-cell

function from fasting plasma glucose and insulin concentrations in man. Diabetologia,

1985. 28(7): p. 412-9.

146. Esposito, K., et al., Effect of weight loss and lifestyle changes on vascular inflammatory

markers in obese women: a randomized trial. JAMA, 2003. 289(14): p. 1799-804.

147. Silva, F.M., J.C. de Almeida, and A.M. Feoli, Effect of diet on adiponectin levels in blood.

Nutr Rev, 2011. 69(10): p. 599-612.

148. Sierksma, A., et al., Effect of moderate alcohol consumption on adiponectin, tumor

necrosis factor-alpha, and insulin sensitivity. Diabetes Care, 2004. 27(1): p. 184-9.

REFERENCES 102

149. Neuhouser, M.L., et al., A low-glycemic load diet reduces serum C-reactive protein and

modestly increases adiponectin in overweight and obese adults. J Nutr, 2012. 142(2): p.

369-74.

150. Li, S., et al., Adiponectin levels and risk of type 2 diabetes: a systematic review and meta-

analysis. JAMA, 2009. 302(2): p. 179-88.

151. Rimm, E.B., et al., Moderate alcohol intake and lower risk of coronary heart disease: meta-

analysis of effects on lipids and haemostatic factors. BMJ, 1999. 319(7224): p. 1523-8.

152. Volek, J.S., et al., An isoenergetic very low carbohydrate diet improves serum HDL

cholesterol and triacylglycerol concentrations, the total cholesterol to HDL cholesterol ratio

and postprandial pipemic responses compared with a low fat diet in normal weight,

normolipidemic women. J Nutr, 2003. 133(9): p. 2756-61.

153. Ros, E., Dietary cis-monounsaturated fatty acids and metabolic control in type 2 diabetes.

Am J Clin Nutr, 2003. 78(3 Suppl): p. 617S-625S.

154. Ascherio, A., et al., Trans fatty acids and coronary heart disease. N Engl J Med, 1999.

340(25): p. 1994-8.

155. Drew, B.G., et al., The emerging role of HDL in glucose metabolism. Nat Rev Endocrinol,

2012. 8(4): p. 237-45.

156. Haffner, S.M., Lipoprotein disorders associated with type 2 diabetes mellitus and insulin

resistance. Am J Cardiol, 2002. 90(8A): p. 55i-61i.

157. Hanley, J.A. and B.J. McNeil, The meaning and use of the area under a receiver operating

characteristic (ROC) curve. Radiology, 1982. 143(1): p. 29-36.

158. Steyerberg and E.W., Clinical Prediction Models. A Practical Approach to Development,

Validation, and Updating. Springer Science+Business Media, LLC, New York 2009.

159. DeLong, E.R., D.M. DeLong, and D.L. Clarke-Pearson, Comparing the areas under two or

more correlated receiver operating characteristic curves: a nonparametric approach.

Biometrics, 1988. 44(3): p. 837-45.

160. Youden, W.J., Index for rating diagnostic tests. Cancer, 1950. 3(1): p. 32-5.

161. Adeboye, B., G. Bermano, and C. Rolland, Obesity and its health impact in Africa: a

systematic review. Cardiovasc J Afr, 2012. 23(9): p. 512-21.

162. Addo, J., L. Smeeth, and D.A. Leon, Obesity in urban civil servants in Ghana: association

with pre-adult wealth and adult socio-economic status. Public Health, 2009. 123(5): p. 365-

70.

163. Benkeser, R.M., R. Biritwum, and A.G. Hill, Prevalence of overweight and obesity and

perception of healthy and desirable body size in urban, Ghanaian women. Ghana Med J,

2012. 46(2): p. 66-75.

164. Frank, L.K., et al., Measures of general and central obesity and risk of type 2 diabetes in a

Ghanaian population. Trop Med Int Health, 2013. 18(2): p. 141-51.

165. Park, S.W., et al., Decreased muscle strength and quality in older adults with type 2

diabetes: the health, aging, and body composition study. Diabetes, 2006. 55(6): p. 1813-8.

166. Adeyemo, A., et al., FTO genetic variation and association with obesity in West Africans

and African Americans. Diabetes, 2010. 59(6): p. 1549-54.

167. Bressler, J., et al., Risk of type 2 diabetes and obesity is differentially associated with

variation in FTO in whites and African-Americans in the ARIC study. PLoS One, 2010. 5(5):

p. e10521.

168. Hennig, B.J., et al., FTO gene variation and measures of body mass in an African

population. BMC Med Genet, 2009. 10: p. 21.

169. Despres, J.P., et al., Abdominal obesity and the metabolic syndrome: contribution to global

cardiometabolic risk. Arterioscler Thromb Vasc Biol, 2008. 28(6): p. 1039-49.

170. Asghar, S., et al., Incidence of diabetes in Asian-Indian subjects: a five year follow-up

study from Bangladesh. Prim Care Diabetes, 2011. 5(2): p. 117-24.

171. Snijder, M.B., et al., Independent and opposite associations of waist and hip

circumferences with diabetes, hypertension and dyslipidemia: the AusDiab Study. Int J

Obes Relat Metab Disord, 2004. 28(3): p. 402-9.

REFERENCES 103

172. Arner, P., Differences in lipolysis between human subcutaneous and omental adipose

tissues. Ann Med, 1995. 27(4): p. 435-8.

173. Snijder, M.B., et al., Associations of hip and thigh circumferences independent of waist

circumference with the incidence of type 2 diabetes: the Hoorn Study. Am J Clin Nutr,

2003. 77(5): p. 1192-7.

174. Wai, W.S., et al., Comparison of measures of adiposity in identifying cardiovascular

disease risk among Ethiopian adults. Obesity (Silver Spring), 2012. 20(9): p. 1887-95.

175. Cheng, C.H., et al., Waist-to-hip ratio is a better anthropometric index than body mass

index for predicting the risk of type 2 diabetes in Taiwanese population. Nutr Res, 2010.

30(9): p. 585-93.

176. Kaur, P., et al., A comparison of anthropometric indices for predicting hypertension and

type 2 diabetes in a male industrial population of Chennai, South India. Ethn Dis, 2008.

18(1): p. 31-6.

177. Wang, Z., et al., Anthropometric indices and their relationship with diabetes, hypertension

and dyslipidemia in Australian Aboriginal people and Torres Strait Islanders. Eur J

Cardiovasc Prev Rehabil, 2007. 14(2): p. 172-8.

178. Stevens, J., et al., Sensitivity and specificity of anthropometrics for the prediction of

diabetes in a biracial cohort. Obes Res, 2001. 9(11): p. 696-705.

179. Frank, L.K., et al., Dietary patterns in urban Ghana and risk of type 2 diabetes. Br J Nutr,

2014. 112(1): p. 89-98.

180. Kant, A.K., Dietary patterns and health outcomes. J Am Diet Assoc, 2004. 104(4): p. 615-

35.

181. Villegas, R., et al., Dietary patterns are associated with lower incidence of type 2 diabetes

in middle-aged women: the Shanghai Women's Health Study. Int J Epidemiol, 2010. 39(3):

p. 889-99.

182. Kant, A.K. and B.I. Graubard, A comparison of three dietary pattern indexes for predicting

biomarkers of diet and disease. J Am Coll Nutr, 2005. 24(4): p. 294-303.

183. Naja, F., et al., Dietary patterns and odds of Type 2 diabetes in Beirut, Lebanon: a case-

control study. Nutr Metab (Lond), 2012. 9(1): p. 111.

184. Eilat-Adar, S., et al., Dietary patterns and their association with cardiovascular risk factors

in a population undergoing lifestyle changes: The Strong Heart Study. Nutr Metab

Cardiovasc Dis, 2013. 23(6): p. 528-35.

185. Jeppesen, C., P. Bjerregaard, and M.E. Jorgensen, Dietary patterns in Greenland and their

relationship with type 2 diabetes mellitus and glucose intolerance. Public Health Nutr,

2014. 17(2): p. 462-70.

186. Noel, S.E., et al., A traditional rice and beans pattern is associated with metabolic

syndrome in Puerto Rican older adults. J Nutr, 2009. 139(7): p. 1360-7.

187. Nilsson, L.M., et al., A traditional Sami diet score as a determinant of mortality in a general

northern Swedish population. Int J Circumpolar Health, 2012. 71(0): p. 1-12.

188. Azadbakht, L. and A. Esmaillzadeh, Dietary diversity score is related to obesity and

abdominal adiposity among Iranian female youth. Public Health Nutr, 2011. 14(1): p. 62-9.

189. Azadbakht, L., et al., Dietary diversity score and cardiovascular risk factors in Tehranian

adults. Public Health Nutr, 2006. 9(6): p. 728-36.

190. Azadbakht, L., P. Mirmiran, and F. Azizi, Dietary diversity score is favorably associated

with the metabolic syndrome in Tehranian adults. Int J Obes (Lond), 2005. 29(11): p. 1361-

191. Pretorius, S., The impact of dietary habits and nutritional deficiencies in urban African

patients living with heart failure in Soweto, South Africa--a review. Endocr Metab Immune

Disord Drug Targets, 2013. 13(1): p. 118-24.

192. Nnanyelugo, D.O., E.C. Okeke, and V. Ibeanu, Knowledge, attitude and usage patterns of

fermented and germinated complementary foods in Nigeria. Plant Foods Hum Nutr, 2003.

58(1): p. 41-51.

REFERENCES 104

193. Schulze, M.B., et al., An approach to construct simplified measures of dietary patterns from

exploratory factor analysis. Br J Nutr, 2003. 89(3): p. 409-19.

194. Kroger, J., et al., Specific food group combinations explaining the variation in intakes of

nutrients and other important food components in the European Prospective Investigation

into Cancer and Nutrition: an application of the reduced rank regression method. Eur J Clin

Nutr, 2009. 63 Suppl 4: p. S263-74.

195. Schulze, M.B., et al., Risk of hypertension among women in the EPIC-Potsdam Study:

comparison of relative risk estimates for exploratory and hypothesis-oriented dietary

patterns. Am J Epidemiol, 2003. 158(4): p. 365-73.

196. Weikert, C., et al., A homocysteine metabolism-related dietary pattern and the risk of

coronary heart disease in two independent German study populations. J Nutr, 2005.

135(8): p. 1981-8.

197. Schulz, M., et al., Identification of a food pattern characterized by high-fiber and low-fat

food choices associated with low prospective weight change in the EPIC-Potsdam cohort.

J Nutr, 2005. 135(5): p. 1183-9.

198. Hoffmann, K., et al., A dietary pattern derived to explain biomarker variation is strongly

associated with the risk of coronary artery disease. Am J Clin Nutr, 2004. 80(3): p. 633-40.

199. Liese, A.D., et al., Food intake patterns associated with carotid artery atherosclerosis in the

Insulin Resistance Atherosclerosis Study. Br J Nutr, 2010. 103(10): p. 1471-9.

200. Micha, R., S.K. Wallace, and D. Mozaffarian, Red and processed meat consumption and

risk of incident coronary heart disease, stroke, and diabetes mellitus: a systematic review

and meta-analysis. Circulation, 2010. 121(21): p. 2271-83.

201. Feskens, E.J.M., D. Sluik, and G.J. van Woudenbergh, Meat Consumption, Diabetes, and

Its Complications. Current Diabetes Reports, 2013. 13(2): p. 298-306.

202. Pan, A., et al., Red meat consumption and risk of type 2 diabetes: 3 cohorts of US adults

and an updated meta-analysis. American Journal of Clinical Nutrition, 2011. 94(4): p. 1088-

1096.

203. Marriott, J., Robinson, M., & Karikari, S. K. , Starch and sugar transformation during the

ripening of plantains and bananas. Journal of the Science of Food and Agriculture, 1981.

32(10): p. 1021-1026.

204. Bhupathiraju, S.N., et al., Glycemic index, glycemic load, and risk of type 2 diabetes:

results from 3 large US cohorts and an updated meta-analysis. American Journal of

Clinical Nutrition, 2014. 100(1): p. 218-232.

205. Liu, S., et al., Dietary glycemic load assessed by food-frequency questionnaire in relation

to plasma high-density-lipoprotein cholesterol and fasting plasma triacylglycerols in

postmenopausal women. Am J Clin Nutr, 2001. 73(3): p. 560-6.

206. Jeppesen, J., et al., Effects of low-fat, high-carbohydrate diets on risk factors for ischemic

heart disease in postmenopausal women. Am J Clin Nutr, 1997. 65(4): p. 1027-33.

207. Mensink, R.P. and M.B. Katan, Effect of dietary fatty acids on serum lipids and lipoproteins.

A meta-analysis of 27 trials. Arterioscler Thromb, 1992. 12(8): p. 911-9.

208. Rosa, M.L.G., et al., Brazil's staple food and incident diabetes. Nutrition, 2014. 30(3): p.

365-368.

209. Rothmann, J., Greenland, S. & Lash, T., Modern Epidemiology third edition. Lippincott

Williams & Wilkins, 2008.

210. Gail, M. H., Interviewer Bias Wiley StatsRef: Statistics Reference Online., 2014.

211. Seidell, J.C., et al., Report from a Centers for Disease Control and Prevention Workshop

on use of adult anthropometry for public health and primary health care. Am J Clin Nutr,

2001. 73(1): p. 123-6.

212. Bouchard, C., BMI, fat mass, abdominal adiposity and visceral fat: where is the 'beef'? Int J

Obes (Lond), 2007. 31(10): p. 1552-3.

213. Posner, B.M., et al., Comparison of techniques for estimating nutrient intake: the

Framingham Study. Epidemiology, 1992. 3(2): p. 171-7.

REFERENCES 105

214. Thompson, F.E. and Subar., A.F., Dietary Assessment Methodology, in Nutrition in the

Prevention and Treatment of Disease. 2008, Academic Press.

215. Harrison, G.G., et al., Underreporting of food intake by dietary recall is not universal: a

comparison of data from egyptian and american women. J Nutr, 2000. 130(8): p. 2049-54.

216. WHO, National Policy for the Prevention and Control of Chronic Non-communicable

Diseases in Ghana. 2012. available online: http://www.mindbank.info/item/1932 (accessed

13th January 2015)

217. Lopriore, C. and E. Muehlhoff, Food Security and Nutrition Trends in West Africa -

Challenges and the Way Forward; Nutrition Programmes Service, Food and Agriculture

Organization Rome, Italy; 2003. available online:

ftp://ftp.fao.org/es/esn/nutrition/ouagafinal.pdf. (accessed 13th January 2015)

218. Subar, A.F., et al., Using intake biomarkers to evaluate the extent of dietary misreporting in

a large sample of adults: the OPEN study. Am J Epidemiol, 2003. 158(1): p. 1-13.

219. Fortson, J.G., The gradient in sub-Saharan Africa: socioeconomic status and HIV/AIDS.

Demography, 2008. 45(2): p. 303-22.

220. Maruf, F.A. and N.V. Udoji, Prevalence and Socio-Demographic Determinants of

Overweight and Obesity in a Nigerian Population. J Epidemiol, 2015. 25(7): p. 475-81.

221. Delisle, H., et al., Urbanisation, nutrition transition and cardiometabolic risk: the Benin

study. Br J Nutr, 2012. 107(10): p. 1534-44.

222. Agardh, E., et al., Type 2 diabetes incidence and socio-economic position: a systematic

review and meta-analysis. Int J Epidemiol, 2011. 40(3): p. 804-18.

223. Greiner, M., D. Pfeiffer, and R.D. Smith, Principles and practical application of the receiver-

operating characteristic analysis for diagnostic tests. Prev Vet Med, 2000. 45(1-2): p. 23-

41.

224. van Dam, R.M., New approaches to the study of dietary patterns. Br J Nutr, 2005. 93(5): p.

573-4.

225. Martinez, M.E., J.R. Marshall, and L. Sechrest, Invited commentary: Factor analysis and

the search for objectivity. Am J Epidemiol, 1998. 148(1): p. 17-9.

226. Hu, F.B., et al., Reproducibility and validity of dietary patterns assessed with a food-

frequency questionnaire. Am J Clin Nutr, 1999. 69(2): p. 243-9.

227. Hoffmann, K., et al., Comparison of two statistical approaches to predict all-cause mortality

by dietary patterns in German elderly subjects. Br J Nutr, 2005. 93(5): p. 709-16.

228. de-Graft Aikins, A., P. Boynton, and L.L. Atanga, Developing effective chronic disease

interventions in Africa: insights from Ghana and Cameroon. Global Health, 2010. 6: p. 6.

229. Ley, S.H., et al., Prevention and management of type 2 diabetes: dietary components and

nutritional strategies. Lancet, 2014. 383(9933): p. 1999-2007.

230. Nathan, D.M., Diabetes: Advances in Diagnosis and Treatment. JAMA, 2015. 314(10): p.

1052-62.

231. World Diabetes Foundation, National Diabetes Programme WDF08-403 Ghana; available

online: http://www.worlddiabetesfoundation.org/projects/ghana-wdf08-403 (accessed 13th

January 2015)

APPENDIX 106

APPENDIX

Figure S1: Food frequency questionnaire used in the KDH study

APPENDIX 107

Figure S1 continued

APPENDIX 108

Figure S1 continued

APPENDIX 109

Figure S1 continued

APPENDIX 110

Figure S2: Example of the assessment of one 24 hour dietary recall used in the KDH study

APPENDIX 111

Table S1: Input variables of the FFQ for factor analysis and reduced rank regression

Original food group

Original food item

input variable for factor

analysis

predictor variable for

RRR

Scientific rationale

Starchy roots and

tubers

Cassava

Plantain

Cocoyam

Yam

Sweet potato

Excluded, because 86% of the participants never

consumed this item

Cereal and cereal

products

Maize (Banku)

Millet

Oats (porridge)

Rice

Bread

Animal products

Fish

Red meat

Poultry

Eggs

Milk

Crab

Legumes, nuts and

beans

Beans

Groundnut

Agushie (pumpkin seeds)

Fruits

Orange

Mango

Papaya

Pineapple

Fruits

Single fruit items were combined into one food group

“Fruits” to avoid overrepresentation of fruit intake

Banana

Pae (avocado)

APPENDIX 112

Table S1 continued

Original food group

Original food item

input variable for

factor analysis

predictor variable

for RRR

Scientific rationale

Vegetables

Tomatoes

Excluded, because 100% of the participants daily consumed this item and

thus did not contribute to variation in the usual diet

Sweet pepper

Excluded, because 100% of the participants daily consumed this item and

thus did not contribute to variation in the usual diet

Garden egg

Okra

Green leafy

vegetables

Green leafy

vegetables

Green leafy vegetables

Carrot

Cucumber

Lettuce

Fats and oils

Palm oil

Vegetable oil

Margarine

Salt and spices

Salt

Excluded, because these items did not contribute to energy and

macronutrient intake

Salt with iodine

Red pepper (dried)

Sweets

Chocolate

Ice cream

Sweets

Single sweets were combined into one food group “Sweets”, because 89%

of the participants consumed these items less than once a week

Toffee

Liquids

Water

Excluded, because 100% of the participants daily consumed this item; this

item did not contribute to energy and macronutrient intake

Juice

Soft drinks

Coffee

Milo (Hot chocolate)

Beer

Wine

Alcoholic drinks

Beer, wine and spirits were combined into one food group “Alcoholic

drinks”, because >90% of the participants never consumed these items

Spirits

APPENDIX 113

Figure S3: Scree plot of eigenvalues ≥ 1.0 among women and men of the KDH study

Figure S4: Scree plot of eigenvalues ≥ 1.0 among controls and total study population

0,0

1,0

2,0

3,0

4,0

5,0

6,0

012345678910 11

Eigenvalue

Factor Number

Men

Women

0,0

0,5

1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

5,0

012345678910 11

Eigenvalue

Factor Number

Controls

Total study population

APPENDIX 114

Table S2: Rotated factor-loadings1 for the two identified dietary patterns among women and

men

Food item

Factor 1

Factor 2

Women

Men

Women

Men

Juice

0.64

0.58

-0.06

-0.24

Sweets

0.62

0.64

-0.07

-0.15

Rice

0.60

0.53

0.05

0.09

Soft drinks

0.57

0.61

-0.03

-0.09

Vegetable oil

0.60

0.55

-0.03

-0.07

Milo (chocolate drink)

0.54

0.55

-0.13

Red meat

0.48

0.55

0.05

-0.03

Chicken eggs

0.46

0.51

-0.11

Margarine

0.42

0.55

0.11

0.03

Fruits

0.42

0.52

0.39

0.38

Carrot

0.37

0.53

0.33

0.34

Lettuce

0.34

0.55

0.37

0.19

Cow milk

0.38

0.46

0.09

0.02

Poultry

0.39

0.38

0.23

0.16

Cucumber

0.31

0.47

0.32

0.29

Plantain

-0.54

-0.26

0.36

0.52

Green leaves

-0.08

0.07

0.50

0.57

Beans

0.07

0.24

0.52

0.48

Gardenegg (egg plant)

-0.21

-0.10

0.49

0.43

Smoked fish

-0.22

-0.18

0.44

0.46

Banku (fermented, boiled maize

product)

0.08

0.13

0.42

0.44

Palm oil

0.05

0.3

0.39

0.42

Okro

0.17

0.21

0.40

0.18

Agushie (pumpkin seeds)

0.21

0.16

0.36

0.47

Crab

-0.01

0.08

0.28

0.43

Bread

0.15

0.29

0.34

0.09

Cassava (maniok)

-0.28

-0.04

0.23

0.29

Millet

-0.08

0.34

0.06

Yam

0.04

0.06

0.18

0.30

Cocoyam

-0.04

0.01

0.20

0.22

Groundnut

0.27

0.38

0.28

0.20

Porridge (from fermented maize)

0.28

0.29

0.18

Coffee

0.20

0.33

-0.02

1 Factor loadings correspond to correlation coefficients between food intake and the dietary pattern score;

factor loadings ≥ 0.35 are marked in bold

APPENDIX 115

Table S3: Rotated factor-loadings1 for the two identified dietary patterns among the total

study population and controls

Food item

Factor 1

Factor 2

Total study

population

Controls

Total study

population

Controls

Juice

0.62

0.59

-0.14

-0.15

Sweets

0.62

0.60

-0.12

-0.14

Rice

0.58

0.54

0.04

0.01

Soft drinks

0.58

0.51

-0.06

0.03

Vegetable oil

0.58

0.49

-0.08

-0.16

Milo (chocolate drink)

0.54

0.50

-0.06

Red meat

0.50

0.47

0.02

-0.08

Chicken eggs

0.48

0.47

-0.03

-0.07

Margarine

0.46

0.50

0.09

0.11

Fruits

0.46

0.49

0.38

0.37

Carrot

0.43

0.50

0.30

0.12

Lettuce

0.42

0.51

0.32

0.11

Cow milk

0.40

0.47

0.06

-0.08

Poultry

0.40

0.19

0.05

Cucumber

0.37

0.45

0.28

0.07

Plantain

-0.45

-0.32

0.43

0.52

Green leaves

-0.02

0.23

0.54

0.46

Beans

0.14

0.22

0.51

0.41

Gardenegg (egg plant)

-0.16

-0.01

0.49

0.47

Smoked fish

-0.19

-0.13

0.46

0.51

Banku (fermented, boiled maize

product)

0.11

0.13

0.43

0.44

Palm oil

0.06

0.17

0.41

0.51

Okro

0.20

0.25

0.35

0.31

Agushie (pumpkin seeds)

0.20

0.25

0.33

0.23

Crab

0.02

0.03

0.32

0.34

Bread

0.20

0.29

0.21

Cassava (maniok)

-0.20

-0.23

0.28

0.52

Millet

-0.04

-0.05

0.27

Yam

0.06

0.21

0.20

Cocoyam

-0.02

-0.12

0.20

0.36

Groundnut

0.30

0.33

0.25

0.21

Porridge (from fermented maize)

0.29

0.30

0.15

0.09

Coffee

0.24

0.23

0.02

0.08

1 Factor loadings correspond to correlation coefficients between food intake and the dietary pattern score;

factor loadings ≥ 0.35 are marked in bold

APPENDIX 116

Table S4: Rotated factor-loadings1 for the two identified dietary patterns among the total

study population

Food item

"Purchase" dietary

pattern

"Traditional”

dietary pattern

Juice

0.62

-0.14

Sweets

0.62

-0.12

Rice

0.58

0.04

Soft drinks

0.58

-0.06

Vegetable oil

0.58

-0.08

Milo (chocolate drink)

0.54

-0.06

Red meat

0.50

0.02

Chicken eggs

0.48

-0.03

Margarine

0.46

0.09

Fruits

0.46

0.38

Carrot

0.43

0.30

Lettuce

0.42

0.32

Cow milk

0.40

0.06

Poultry

0.40

0.19

Cucumber

0.37

0.28

Plantain

-0.45

0.43

Green leaves

-0.02

0.54

Beans

0.14

0.51

Gardenegg (egg plant)

-0.16

0.49

Smoked fish

-0.19

0.46

Banku (fermented, boiled maize product)

0.11

0.43

Palm oil

0.06

0.41

Okro

0.20

0.35

Agushie (pumpkin seeds)

0.20

0.33

Crab

0.02

0.32

Bread

0.20

0.29

Cassava (maniok)

-0.20

0.28

Millet

-0.04

0.27

Yam

0.06

0.21

Cocoyam

-0.02

0.20

Groundnut

0.30

0.25

Porridge (from fermented maize)

0.29

0.15

Coffee

0.24

0.02

1 Factor loadings correspond to correlation coefficients between food intake and the dietary pattern score;

factor loadings ≥ 0.35 are marked in bold

APPENDIX 117

Table S5: Rotated Factor loadings1 of a three-factor solution among the total study

population

Factor 1

Factor 2

Factor 3

Soft drinks

0.64

0.06

Juice

0.63

0.11

-0.09

Sweets

0.61

0.14

-0.10

Vegetable oil

0.59

0.10

-0.03

Rice

0.58

0.15

0.08

Red meat

0.54

0.03

0.11

Milo (chocolate drink)

0.53

0.13

-0.04

Chicken eggs

0.50

0.06

0.03

Margarine

0.46

0.12

0.13

Fruits

0.39

0.31

0.34

Groundnut

0.32

0.07

0.32

Coffee

0.26

0.01

0.06

Plantain

-0.51

0.07

0.37

Cucumber

0.04

0.79

-0.09

Carrot

0.11

0.77

-0.06

Lettuce

0.14

0.71

Cow milk

0.27

0.37

-0.08

Porridge (from fermented maize)

0.16

0.35

0.01

Poultry

0.29

0.34

0.09

Agushie (pumpkin seeds)

0.08

0.33

0.22

Bread

0.11

0.27

0.21

Crab

-0.08

0.26

0.22

Millet

-0.12

0.19

0.18

Banku (fermented, boiled maize product)

0.14

0.02

0.52

Palm oil

0.11

-0.03

0.51

Gardenegg (egg plant)

-0.16

0.04

0.51

Beans

0.07

0.24

0.47

Smoked fish

-0.21

0.06

0.45

Green leaves

-0.12

0.28

0.45

Cassava (maniok)

-0.10

-0.23

0.43

Okra

0.23

0.03

0.43

Cocoyam

0.06

-0.13

0.32

Yam

0.05

0.23

1 Factor loadings correspond to correlation coefficients between food intake and the dietary pattern score;

factor loadings ≥ 0.35 are marked in bold

APPENDIX 118

Table S6: Rotated Factor loadings1 of a four-factor solution among the total study

population

Factor 1

Factor 2

Factor 3

Factor 4

Soft drinks

0.68

0.03

0.02

Juice

0.67

0.12

-0.07

-0.09

Sweets

0.67

0.15

-0.05

-0.13

Milo (chocolate drink)

0.58

0.13

-0.01

-0.07

Red meat

0.51

0.02

-0.03

0.21

Vegetable oil

0.49

0.08

-0.27

0.3

Margarine

0.49

0.12

0.1

0.07

Chicken eggs

0.48

0.05

-0.08

0.13

Rice

0.45

0.11

-0.25

0.44

Fruits

0.39

0.29

0.22

0.29

Coffee

0.25

0.09

Cucumber

0.03

0.79

-0.10

0.03

Carrot

0.12

0.77

-0.06

0.03

Lettuce

0.10

0.70

-0.08

0.16

Cow milk

0.35

0.38

0.05

-0.19

Porridge (from fermented maize)

0.23

0.37

0.13

-0.14

Poultry

0.27

0.33

0.16

Agushie (pumpkin seeds)

0.05

0.32

0.10

0.26

Bread

0.15

0.26

0.21

0.08

Crab

-0.05

0.25

0.23

0.07

Gardenegg (egg plant)

-0.04

0.04

0.62

0.02

Plantain

-0.36

0.08

0.60

-0.17

Cassava (maniok)

0.03

-0.23

0.57

-0.05

Smoked fish

-0.15

0.05

0.47

0.14

Green leaves

-0.06

0.27

0.46

0.15

Palm oil

0.11

-0.06

0.36

Cocoyam

0.06

-0.15

0.23

0.21

Yam

0.06

0.04

0.18

0.15

Beans

-0.05

0.19

0.12

0.66

Groundnut

0.22

0.03

0.02

0.50

Banku (fermented, boiled maize product)

0.09

-0.02

0.28

0.49

Millet

-0.23

0.16

-0.05

0.40

Okra

0.22

0.01

0.28

0.33

1 Factor loadings correspond to correlation coefficients between food intake and the dietary pattern score;

factor loadings ≥ 0.35 are marked in bold

APPENDIX 119

Table S7: Rotated Factor loadings1 of a five-factor solution among the total study population

Factor 1

Factor 2

Factor 3

Factor 4

Factor 5

Soft drinks

0.70

0.03

0.01

0.05

-0.06

Juice

0.69

0.14

-0.12

-0.03

Sweets

0.68

0.16

-0.14

-0.02

0.03

Milo (chocolate drink)

0.55

0.06

-0.06

-0.05

0.22

Red meat

0.52

0.01

0.20

-0.05

0.04

Vegetable oil

0.51

0.1

0.27

-0.28

-0.05

Margarine

0.46

0.04

0.09

0.05

0.23

Rice

0.46

0.13

0.42

-0.27

-0.02

Chicken eggs

0.46

0.01

0.13

-0.12

0.14

Fruits

0.39

0.26

0.30

0.19

0.14

Coffee

0.22

-0.06

0.11

-0.06

0.19

Cucumber

0.07

0.84

-0.03

0.01

Carrot

0.15

0.81

0.06

Lettuce

0.14

0.74

0.13

-0.04

0.03

Poultry

0.26

0.28

0.16

-0.04

0.19

Crab

-0.03

0.27

0.08

0.26

Beans

-0.05

0.18

0.67

0.06

0.07

Banku (fermented, boiled maize product)

0.09

-0.04

0.51

0.23

0.05

Groundnut

0.25

0.08

0.49

0.02

-0.12

Millet

-0.25

0.13

0.41

-0.11

0.11

Palm oil

0.09

-0.12

0.40

0.29

0.16

Okra

0.25

0.03

0.33

0.28

-0.05

Agushie (pumpkin seeds)

0.03

0.26

0.27

0.05

0.21

Cassava (maniok)

0.08

-0.13

-0.04

0.64

-0.27

Plantain

-0.36

0.07

-0.12

0.62

0.05

Gardenegg (egg plant)

-0.05

-0.01

0.08

0.60

0.15

Smoked fish

-0.14

0.06

0.17

0.47

0.01

Green leaves

-0.08

0.19

0.20

0.41

0.27

Cocoyam

0.09

-0.10

0.21

0.25

-0.14

Bread

0.04

0.01

0.16

0.03

0.73

Cow milk

0.25

0.15

-0.14

-0.09

0.69

Porridge (from fermented maize)

0.20

0.27

-0.12

0.09

0.34

Yam

0.03

-0.03

0.18

0.12

0.20

1 Factor loadings correspond to correlation coefficients between food intake and the dietary pattern score;

factor loadings ≥ 0.35 are marked in bold

APPENDIX 120

Table S8: Rotated Factor loadings1 of a two-factor solution with single fruit and vegetable

items among the total study population

Factor 1

Factor 2

Juice

0.63

-0.12

Sweets

0.62

-0.12

Soft drinks

0.58

-0.02

Vegetable oil

0.57

0.03

Rice

0.56

-0.09

Milo (chocolate drink)

0.54

-0.05

Red meat

0.49

0.02

Banana

0.47

0.23

Eggs

0.47

-0.04

Margarine

0.45

0.08

Carrot

0.43

0.26

Pineapple

0.42

0.26

Lettuce

0.41

0.27

Milk

0.40

0.02

Poultry

0.39

0.16

Cucumber

0.36

0.23

Orange

0.35

0.25

Groundnut

0.29

0.25

Porridge (from fermented maize)

0.29

0.14

Coffee

0.24

0.01

Plantain

-0.45

0.42

Green leafy vegetable

-0.02

0.52

Beans

0.12

0.48

Garden egg (egg plant)

-0.17

0.46

Papaya

0.23

0.46

Avocado

0.20

0.45

Fish

-0.20

0.45

Banku (fermented, boiled maize product)

0.09

0.44

Palm oil

0.05

0.41

Mango

0.29

0.37

Okra

0.18

0.33

Crab

0.02

0.32

Agushie (pumpkin seeds)

0.18

0.31

Cassava (maniok)

-0.21

0.29

Millet

-0.04

0.26

Bread

0.18

0.24

Cocoyam

-0.03

0.23

Yam

0.04

0.17

1 Factor loadings correspond to correlation coefficients between food intake and the dietary pattern score;

factor loadings ≥ 0.35 are marked in bold

APPENDIX 121

Table S9: Rotated Factor loadings1 of a two-factor solution with various fruit and vegetable

groups among the total study population

Factor 1

Factor 2

Juice

0.65

-0.07

Sweets

0.64

-0.06

Soft drinks

0.61

0.01

Vegetable oil

0.58

-0.02

Rice

0.57

0.10

Milo (chocolate drink)

0.56

Red meat

0.51

0.07

Eggs

0.49

0.01

Margarine

0.47

0.12

Milk

0.39

0.07

Poultry

0.37

0.21

Carrot

0.33

0.31

Citrus fruits (orange)

0.31

0.28

Porridge

0.27

0.17

Coffee

0.25

0.05

Plantain

-0.49

0.36

Leafy vegetables (green leafy vegetables and

lettuce)

0.07

0.54

Beans

0.07

0.53

Exotic fruits (mango, papaya, and pineapple)

0.34

0.50

Banku

0.07

0.46

Vegetables (garden egg, okra, and cucumber)

0.1

0.45

Classical fruits (banana, avocado)

0.36

0.44

Fish

-0.23

0.42

Palm oil

0.05

0.41

Agushie

0.14

0.36

Crab

-0.04

0.34

Groundnut

0.27

0.30

Millet

-0.09

-0.29

Bread

0.18

0.28

Cassava

-0.21

0.26

Cocoyam

-0.03

0.25

Yam

0.04

0.19

1 Factor loadings correspond to correlation coefficients between food intake and the dietary pattern score;

factor loadings ≥ 0.35 are marked in bold

DANKSAGUNG 122

DANKSAGUNG

Die vorliegende Dissertation wurde in der Abteilung Molekulare Epidemiologie am

Deutschen Institut für Ernährungsforschung Potsdam-Rehbrücke (DIfE) angefertigt. An

dieser Stelle möchte ich mich bei allen Personen bedanken, die mich auf diesem Weg

unterstützt haben.

Mein besonderer Dank gilt dabei:

Meiner Betreuerin Frau Dr. Ina Danquah, für die Möglichkeit dieses interessante Thema

bearbeiten zu können, sowie die kreativen und konstruktiven Anregungen.

Herrn Prof. Dr. Schulze für die Möglichkeit in seiner Abteilung promovieren zu können,

den großen Freiraum bei der Ausarbeitung der Dissertation, die konstruktiven

Anregungen und seine Unterstützung während der gesamten Promotionszeit.

Herrn Prof. Dr. Reinhard Busse für die Betreuung und Begutachtung der Arbeit an der

Technischen Universität Berlin.

Meinen Co-Autoren der veröffentlichten Manuskripte (insbesondere Dr. Janine Kröger und

Dr. Alexandros Heraclides) für deren wertvolle Anregungen und die Unterstützung vor

allem zu Beginn meiner Promotionszeit.

Charlotte Jeppesen und Simone Jacobs für das Korrekturlesen meiner Arbeit und deren

freundschaftliche Unterstützung.

Kristin Mühlenbruch und Olga Kuxhaus für ihre statistische Expertise.

Allen Kollegen für die angenehme und freundschaftliche Atmosphäre in der Abteilung.

Meinen Eltern, Großeltern und meinem Bruder, die mich in meinem Werdegang immer

unterstützt haben.

Ein ganz besonders lieber Dank geht an meinen Ehemann Christian und meine Tochter

Amelie, deren Liebe und Unterstützung, insbesondere in den schwierigen Phasen meiner

Promotionszeit, maßgeblich zum Gelingen dieser Arbeit beigetragen haben.

EIDESSTATTLICHE ERKLÄRUNG 123

EIDESSTATTLICHE ERKLÄRUNG

Hiermit erkläre ich des Eides statt, dass ich die im Fachbereich Management im

Gesundheitswesen der Technischen Universität Berlin eingereichte Dissertation mit dem

Titel „Type 2 diabetes in urban Ghana - the role of anthropometry and nutrition“

selbstständig verfasst und ohne unerlaubte Hilfsmittel angefertigt habe. Literatur und

Hilfsmittel, die zur Anfertigung der Arbeit herangezogen wurden, sind im

Literaturverzeichnis aufgelistet. Teile dieser Arbeit sind im Rahmen des

Promotionsvorhabens bereits in ähnlicher Form veröffentlicht worden und als solche

gekennzeichnet *. Weiterhin versichere ich, die Arbeit an keiner anderen Hochschule oder

Fachhochschule eingereicht zu haben.

*Relevante Publikationen:

1. Frank LK, Heraclides A, Danquah I, Bedu-Addo G, Mockenhaupt FP, Schulze MB:

Measures of general and central obesity and risk of type 2 diabetes in a

Ghanaian population. Trop Med Int Health 2013, 18:141-151.

2. Frank LK, Kroger J, Schulze MB, Bedu-Addo G, Mockenhaupt FP, Danquah I:

Dietary patterns in urban Ghana and risk of type 2 diabetes. Br J Nutr 2014,

112:89-98.

3. Frank LK, Jannasch F, Kroger J, Bedu-Addo G, Mockenhaupt FP, Schulze MB,

Danquah I: A dietary pattern derived by reduced rank regression is

associated with type 2 diabetes in an urban Ghanaian population Nutrients

2015, 7, 5497-5514.

Laura Frank Berlin, den 14.01.2016