Integrative development of a short screening questionnaire of highly processed food consumption (sQ-HPF)

Martinez‑Perezetal. Int J Behav Nutr Phys Act (2022) 19:6 https://doi.org/10.1186/s12966‑021‑01240‑6 RESEARCH Integrative development ofashort screening questionnaire ofhighly processed food consumption (sQ‑HPF) Celia Martinez‑Perez1, Lidia Daimiel1* , Cristina Climent‑Mainar1, Miguel Ángel Martínez‑González2,3,4, Jordi Salas‑Salvadó2,5,6, Dolores Corella2,7, Helmut Schröder8, Jose Alfredo Martinez2,9,10, Ángel M. Alonso‑Gómez2,11, Julia Wärnberg2,12, Jesús Vioque13,14, Dora Romaguera2,15, José López‑Miranda2,16, Ramón Estruch2,17, Francisco J. Tinahones2,18, José Lapetra2,19, Lluis Serra‑Majem2,20, Aurora Bueno‑Cavanillas13,21, Josep A. Tur2,22, Vicente Martín Sánchez13,23, Xavier Pintó2,24, Miguel Delgado‑Rodríguez13,25, Pilar Matía‑Martín26, Josep Vidal27,28, Clotilde Vázquez2,29, Emilio Ros2,28, Javier Basterra3,30, Nancy Babio2,5,6, Patricia Guillem‑Saiz2,7, María Dolores Zomeño8,31, Itziar Abete2,10, Jessica Vaquero‑Luna2,11, Francisco Javier Barón‑López32, Sandra Gonzalez‑Palacios13,14, Jadwiga Konieczna2,15, Antonio Garcia‑Rios2,16, María Rosa Bernal‑López33, José Manuel Santos‑Lozano2,19, Maira Bes‑Rastrollo2,3, Nadine Khoury2,5,6, Carmen Saiz2,7, Karla Alejandra Pérez‑Vega8, María Angeles Zulet2,10, Lucas Tojal‑Sierra2,11, Zenaida Vázquez Ruiz2,3, Maria Angeles Martinez2,5,6, Mireia Malcampo8, José M. Ordovás1,34 and Rodrigo San‑Cristobal9 Abstract Background: Recent lifestyle changes include increased consumption of highly processed foods (HPF), which has been associated with an increased risk of non‑communicable diseases (NCDs). However, nutritional information relies on the estimation of HPF consumption from food‑frequency questionnaires (FFQ) that are not explicitly developed for this purpose. We aimed to develop a short screening questionnaire of HPF consumption (sQ‑HPF) that integrates criteria from the existing food classification systems. Methods: Data from 4400 participants (48.1% female and 51.9% male, 64.9 ± 4.9 years) of the Spanish PREDIMED‑Plus (“PREvention with MEDiterranean DIet”) trial were used for this analysis. Items from the FFQ were classified according to four main food processing‑based classification systems (NOVA, IARC, IFIC and UNC). Participants were classified into tertiles of HPF consumption according to each system. Using binomial logistic regression, food groups associated with agreement in the highest tertile for at least two classification systems were chosen as items for the questionnaire. ROC analysis was used to determine cut‑off points for the frequency of consumption of each item, from which a score was calculated. Internal consistency of the questionnaire was assessed through exploratory factor analysis (EFA) and Cronbach’s analysis, and agreement with the four classifications was assessed with weighted kappa coefficients. © The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. The Creative Commons Public Domain Dedication waiver (http:// creat iveco mmons. org/ publi cdoma in/ zero/1. 0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Open Access *Correspondence: [email protected] 1 Nutritional Genomics and Epigenomics Group, Precision Nutrition and Obesity Program, IMDEA Food, CEI UAM + CSIC, Ctra. Cantoblanco, 8, 28049 Madrid, Spain Full list of author information is available at the end of the article Page 2 of 16 Martinez‑Perezetal. Int J Behav Nutr Phys Act (2022) 19:6 Background Changes in eating patterns are occurring worldwide [1]. A common feature of such changes is the transition from minimally processed food to moderately processed to highly processed (HPF) or ultra-processed foods (UPF) [2–10]. A widely used, although controversial [11], definition of UPF is that these are “industrial formulations made mostly or entirely from substances derived from foods and additives, with little if any intact food” [12]. According to the NOVA classification system, these foods are highly palatable and habit-forming, microbiologically safe, affordable, strongly marketed and advertised, and sold in convenient and attractive packaging, promoting their overconsumption [12, 13]. This, together with the evidence showing their negative impact on health [9, 10, 14–18], has turned UPF consumption into a potential public health concern [19] in need of more solid research. While the term UPF is mainly attributed to the NOVA system, other food processing-based classification systems have described foods and drinks of similar characteristics under their categories of HPF [20, 21], so we will use the term HPF to refer to this type of foods. Although numerous studies have demonstrated a link between the risk of developing non-communicable diseases and HPF consumption [12, 17, 22–27], epidemiological research faces some hurdles in this field. First, the existence of multiple food processing-based classification systems based on different criteria [21, 28] results in heterogeneous conclusions regarding health outcomes depending on the system used, as shown recently for cardiometabolic health markers [20]. Second, the lack of an effective tool to assess HPF consumption in clinical studies severely hinders the downstream analysis of its relationship to disease risk. For instance, in the PREDIMED-Plus (from the Spanish “PREvention with MEDiterranean DIet”) trial, dietary intake was assessed through a lengthy food frequency questionnaire (FFQ). Therefore, estimation of HPF consumption requires the classification of FFQ items according to the selected classification system, as previously done [20, 24, 25, 29]. This process is time-consuming and subject to bias on the part of the researcher classifying the items since FFQs are not specifically designed to include HPF. This is because an FFQ is designed to estimate consumption of general commonly consumed foods, of which some may fall into the HPF category, but this depends on the classification system used [20]. In addition, calculations of daily consumption of each HPF item and percentage over total intake (in grams or kcal per day) are commonly used to infer their association with health outcomes [9, 17, 30]. These calculations are not direct and time-consuming when derived from current tools such as FFQs or 24 h recalls. There is, therefore, the need for an easy-touse and comprehensive measure that assesses HPF consumption in the general population. This paper describes the development of a new screening questionnaire that allows an easy and quick determination of a subject’s HPF consumption, the sQ-HPF. We view this as an integrative tool since it incorporates criteria from four foodprocessing-based classification systems. We developed this questionnaire based on available data from an FFQ to create a tool that could replace the estimation of HPF consumption from FFQ in future studies. We hypothesize that the sQ-HPF is comparable in terms of evaluating HPF consumption to other dietary assessment tools that must be used in combination with food processing-based classification systems and can effectively capture longitudinal changes in HPF consumption. This questionnaire will be of potential interest to the scientific community, especially in the context of clinical nutrition and public health. Its use should minimize the difficulty in comparing results from studies that use different classification systems, allowing a straightforward evaluation of the HPF dietary pattern in large Results: Regression analysis identified 14 food groups (items) associated with high HPF consumption for at least two classification systems. EFA showed that items were representative contributors of a single underlying factor, the “HPF dietary pattern” (factor loadings around 0.2). We constructed a questionnaire asking about the frequency of consump‑ tion of those items. The threshold frequency of consumption was selected using ROC analysis. Comparison of the four classification systems and the sQ‑HPF showed a fair to high agreement. Significant changes in lifestyle characteristics were detected across tertiles of the sQ‑HPF score. Longitudinal changes in HPF consumption were also detected by the sQ‑HPF, concordantly with existing classification systems. Conclusions: We developed a practical tool to measure HPF consumption, the sQ‑HPF. This may be a valuable instru‑ ment to study its relationship with NCDs. Trial registration: Retrospectively registered at the International Standard Randomized Controlled Trial Registry (ISRCT N8989 8870) on July 24, 2014. Keywords: Ultra‑processed food, Highly processed food, Questionnaire, PREDIMED‑Plus, NOVA, Food processing‑ based classification Page 3 of 16 Martinez‑Perezetal. Int J Behav Nutr Phys Act (2022) 19:6 epidemiological studies efficiently and comparably. Altogether, it will enable the development of tailored nutritional interventions and food policies to limit HPF consumption and prevent diet-related diseases. Methods Study population Data from the PREDIMED-Plus trial (from the Spanish “PREvention with MEDiterranean DIet”) was used. This is an ongoing 6-year, multicenter, randomized, parallelgroup clinical trial launched in Spain in 2013. The main aim of the study is to evaluate the effect on primary cardiovascular disease prevention of an intensive weight loss and its long-term maintenance through a lifestyle intervention based on three pillars: energy-restricted Mediterranean diet (er-MedDiet), increased physical activity (PA) and behavioral support. The study protocol, including study design and data collection, can be found at the PREDIMED-Plus website (https:// www. predi medpl us. com/ en/) and was approved according to the ethical standards of the Declaration of Helsinki by the Institutional Review Boards (IRBs) of all participating centers. All participants provided written consent for their participation in the study. The trial is conducted in 23 Spanish centers and involves 6874 participants (48.5% female, 54.5% male) between 55 and 75 years old (mean age and SD 65.0 ± 4.9) who presented with overweight or obesity (BMI ≥ 27 and < 40 kg/m2) and met at least three criteria for metabolic syndrome (MetS) as previously described [31]. Details about the cohort have been described elsewhere [32]. The trial was retrospectively registered at the International Standard Randomized Controlled Trial Registry with number 89898870 on 24th July 2014. The present analysis used baseline and longitudinal (years 1 and 2) data from the PREDIMED-Plus study data set dated 26th June 2020 was used. Participants with implausible energy intakes (< 500 or > 3500 kcal for females and < 800 or > 4000 kcal for males) were excluded. In addition, participants with missing values for dietary, lifestyle and socioeconomic variables were not included in the analysis. After the definition of the tertile agreement variable (see Statistical analyses section), participants with (1) no coincidence in extreme tertiles of HPF consumption (either in tertile 1 or tertile 3) according to at least two classification systems and (2) participants that were classified in tertile 1 for two classification systems and tertile 3 for the other two classification systems were excluded. This was done to select extreme HPF consumers to further construct the binomial regression model. The final number of participants included in the analysis for the development of the questionnaire was 4400 (48.1% female and 51.9% male, 64.9 ± 4.9 years) (Fig.1). This study adhered to the STROBE-nut reporting guidelines [33]. Blood measurements Trained nurses collected blood samples after overnight fasting at the recruiting centers or primary health care centers. Plasma glucose, triglycerides and total cholesterol levels were measured following standard enzymatic methods. Anthropometric measurements Weight and waist circumference measurements were taken from participants in light clothing with no shoes or accessories, using an electronic calibrated scale and an anthropometric tape, respectively. Waist circumference was measured midway between the lowest rib and the iliac crest. Height measurements were taken using a wallmounted stadiometer. Body Mass Index (BMI) was calculated as the weight in kilograms divided by the square of height in meters. Lifestyle measurements Socioeconomic and PA data were collected through the general PREDIMED-Plus questionnaire. Based on the validated questionnaires, the REGICOR [34] and the Rapid Assessment of Physical Activity (RAPA) [35], participants were asked about the frequency and intensity of physical activities, and three levels of PA were defined as follows: low (frequent sitting and little walking and/or frequent sitting and moderate sustained efforts), medium (frequent walking with no vigorous efforts), high (frequent walking and vigorous efforts and/or frequent vigorous efforts). Sedentary behaviors were assessed through a validated questionnaire, the Spanish version of the Nurses’ Health Study (NHS) questionnaire [36]. Data about eating habits (binge eating, snacking) were collected within the multidimensional scale of weight locus control questionnaire [37]. Trained interviewers administered questionnaires in individual face-to-face sessions. Dietary measurements To assess the dietary intake of participants over the last year, a validated semi-quantitative 143-item FFQ [38– 40] that considers variations in dietary patterns among seasons, weekdays and weekends was used. Participants were asked the average frequency of consumption of a commonly used portion size (e.g., glass, cup, slice) for each food or beverage item. Nine options for frequency of consumption are given, ranging from “never or hardly ever” to “more than six times a day.” To estimate the daily consumption for each item, the portion size was multiplied by the frequency of consumption and then expressed as grams per day. This calculation Page 4 of 16 Martinez‑Perezetal. Int J Behav Nutr Phys Act (2022) 19:6 was not possible for the fried foods item as the portion size is not specified in the FFQ. To assess adherence to an er-MedDiet, a 17-item questionnaire specially developed and validated for the PREDIMED-Plus trial [41, 42] was used. Through this questionnaire, the frequency of consumption of traditional Mediterranean food items is evaluated. One point is scored when the answer meets specific criteria defining er-MedDiet, so the higher the score, the better adherence to this diet. HPF consumption andfood processing‑based classification systems Four food-processing based classification systems – the NOVA [12, 43, 44], the International Agency for Research on Cancer (IARC) [7, 45], the International Food Information Council Foundation (IFIC) [46, 47], and the University of North Carolina (UNC) [48] systems – were used to classify FFQ items into processing categories as previously described [20]. In the present Fig. 1 Flowchart of the PREDIMED‑Plus participants. Number of subjects shown in bold Page 5 of 16 Martinez‑Perezetal. Int J Behav Nutr Phys Act (2022) 19:6 study, HPF refers to the following groups: Group 4 for NOVA, Group 3 for IARC, Groups 4 and 5 for IFIC, and Groups 4.1 and 4.2 for UNC. For each participant, HPF consumption was estimated according to each classification system as the sum of grams per day consumed from foods in the HPF group, divided by the total grams of food consumed per day and multiplied by 100 [20]. The frequency of consumption was directly obtained from FFQ answers in times per day. Statistical analyses Data analysis was conducted using R programming language [49] in RStudio [50] and with the following statistical packages: “DescTools” [51], “psych” [52], “tableone” [53], “cutpointr” [54], “corrplot” [55], “vcd” [56] and rstatisx [57]. Participants were classified according to tertiles of HPF consumption for each classification system (T1 – low HPF consumption, T3 high HPF consumption). Tertiles were chosen to capture and show the variability in HPF consumption while allowing a straightforward comparison between classification systems. Then, subjects were classified according to tertile agreement for, at least, two classification systems. Those classified in T3 in at least two classification systems scored “1” for tertile agreement, while those classified in T1 for at least two classification systems scored “0”. FFQ items were classified into food groups according to their similar nature, nutritional characteristics, and/or form of consumption (Supplementary Table1). The association between tertile agreement and frequency of consumption for each food group was analyzed through binomial logistic regression, adjusted for age, sex, recruitment center, energy intake, physical activity level, medication for blood pressure and diabetes, working status, educational level, and civil status, using the “glm” function from R base package “stats.” These covariates were selected due to their potential effect on HPF consumption. To calculate optimal cut-off points of frequency of consumption for the selected items, receiver operating characteristic (ROC) analysis was performed using the R package “cutpointr” and the “cutpointr” function [54]. The method to estimate the cut-off points was based on the maximization of the Youden-Index [58]. Estimated cut-off points were used to establish the criteria for scoring 1 point in the sQ-HPF, which indicated high HPF consumption, or 0, indicating low HPF consumption. Cut-off points were adapted to the nine possible answers of the FFQ, so the criteria for scoring were based on a threshold frequency of consumption for each item (i.e., food group). In parallel, exploratory factor analysis (EFA) was performed to identify underlying relationship patterns between items included in the sQ-HPF using the “fa” function from the “psych” package [52]. To test for data suitability for the EFA, the Kaiser-Meyer-Olkin Criterion [59] and Bartlett’s test of sphericity [60] were applied. The EFA was performed without rotation and with a principal factor solution as a factoring method. Factor retention was based on the scree plot and Kaiser’s criterion [61]. Cronbach’s alpha [62] was calculated as a measure of internal consistency of the questionnaire. Using the criteria for scoring, the sQ-HPF score was calculated for each subject in the PREDIMED-Plus database, and the questionnaire estimated HPF consumption was calculated through linear regression analysis using the sQ-HPF score as the dependent variable. For descriptive analyses, participants were classified into tertiles of the sQ-HPF score. Data is shown in tables as “mean (standard deviation, SD)” for continuous variables and as “number of subjects (%)” for categorical variables. Statistically significant differences (p< 0.05) in dietetic and lifestyle variables among tertiles were compared using a one-way ANOVA test for continuous variables and a Chi-squared test for categorical variables. P-values were adjusted for age, sex, recruitment center, energy intake, physical activity level, medication for blood pressure and diabetes, working status, educational level, and civil status. To assess the concordance between tertiles of HPF consumption calculated by the four classification systems and by the sQ-HPF, weighted Cohen’s kappa (κ) coefficients were calculated with the function “Kappa” from the R package “vcd”. For longitudinal analysis of HPF consumption, a linear mixed model was performed using the R package “lme4” and “emmeans” with the same covariates as previous analyses. The number of subjects used for this analysis was 3284 due to longitudinal data loss. Results General characteristics ofthePREDIMED‑Plus cohort according totertile agreement Subjects scored 1 in the tertile agreement variable if they were classified in the highest tertile (T3) of HPF consumption for at least two classifications systems, while they scored 0 if they were classified in the lowest tertile (T1) of HPF consumption for at least two classifications systems. General characteristics of PREDIMED-Plus participants at baseline according to the scores of the tertile agreement variable are shown in Table1. Subjects who scored 1 (high HPF consumption by at least two classification systems) were mainly men (73.1%), had higher energy intake (2559.84 kcal/day) and lower MedDiet adherence (7.65 points) than subjects who scored 0 (low HPF consumption by at least two classifications systems). Among the high HPF subjects, 78.8% were married and 41.8% had a primary education level. Around half of the subjects who scored 1 showed a low level of PA (55.3%) and were taking medication for cholesterol (49.8%). Page 6 of 16 Martinez‑Perezetal. Int J Behav Nutr Phys Act (2022) 19:6 In addition, 75.7% of them were taking medication for blood pressure and 25.9% were on diabetes medication. According to these results, the variables age, sex, recruitment center, energy intake, physical activity level, medication for blood pressure and diabetes, working status, educational level, and civil status were selected as covariates for further analysis due to their potential effect on HPF consumption. MedDiet adherence was not selected as a covariate due to the presence of collinearity with HPF consumption. Development ofthesQ‑HPF Food groups were defined based on the similarity in nature, nutritional profile and/or form of consumption of the PREDIMED-Plus FFQ food and beverage items (Supplementary Table1). Food groups chosen to be included in the sQ-HPF showed a significant positive association (p-value adjusted by Bonferroni < 0.05) between their frequency of consumption and a value of 1 of tertile agreement, i.e., the subject is in tertile 3 of HPF consumption for at least two food processing-based classification systems (Table2). A final solution of 14 food groups was selected and included the following: fatty dairy products, sugary dairy products, cured meat, fats, fermented alcohol, distilled alcohol, sugary and artificially sweetened drinks, sweets, snacks, ready to eat products, refined cereals, sauces, additives, and fried foods. EFA revealed that most of the items selected for the sQ-HPF had factor loadings higher than 0.2, indicating that they were representative contributors to the factor (Table3). The measure of sample adequacy (MSA) was 0.78, considered as “good” for the verification of the proportion of variance in variables that can be caused by factors, according to the Kaiser-Meyer-Olkin Criterion. Bartlett’s test of sphericity was highly significant (p< 0.001), indicating that variables were correlated in the population. This, together with the MSA value, indicated the adequacy of the data to proceed with EFA. Factor retention applying the Kaiser criterion revealed a single underlying factor being identified by the questionnaire items, namely the HPF dietary pattern. The internal consistency of the questionnaire items was evaluated with Cronbach’s alpha, which had a moderate value of 0.67. Table 1 General characteristics of PREDIMED‑Plus participants at baseline according to tertile agreement Data shown as “mean (standard deviation, SD)” for continuous variables and as “number of subjects (%)” for categorical variables. One‑way ANOVA test used for continuous variables and Chi‑squared test used for categorical variables. Significant p‑values (< 0.05) shown in bold MedDiet Mediterranean diet, PA Physical activity a Tertile agreement variable: score 0 – “low HPF consumer” if a subject is classified in T1 of HPF consumption by at least two classification systems; score 1 – “high HPF consumer” if a subject is classified in T3 by at least two classification systems. HPF: highly processed food. N= 4400 Tertile agreementa Low HPF High HPF p n2186 2214 Age (years) 66.14 (4.50) 63.68 (5.07) < 0.001 Female sex (%) 1522 (69.6) 595 (26.9) < 0.001 Civil status (%) 0.001 Single 115 (5.3) 109 (4.9) Married 1622 (74.2) 1745 (78.8) Widowed/divorced 449 (20.5) 360 (16.3) Education level (%) Primary 1252 (57.3) 926 (41.8) < 0.001 Secondary 538 (24.6) 704 (31.8) College 396 (18.1) 584 (26.4) Active working status (%) 307 (14.0) 642 (29.0) < 0.001 Energy intake (kcal/day) 2158.46 (480.16) 2559.84 (571.11) < 0.001 MedDiet adherence score 9.46 (2.54) 7.65 (2.61) < 0.001 PA level (%) 0.001 Low 1188 (54.3) 1225 (55.3) Medium 904 (41.4) 841 (38.0) High 94 (4.3) 148 (6.7) Blood pressure medication (%) 1717 (78.5) 1676 (75.7) 0.027 Cholesterol medication (%) 1123 (51.4) 1102 (49.8) 0.303 Diabetes medication (%) 631 (28.9) 573 (25.9) 0.029 Page 7 of 16 Martinez‑Perezetal. Int J Behav Nutr Phys Act (2022) 19:6 The sQ-HPF was based on the 14 food groups selected previously. Each item asked about the frequency of consumption of a particular food group (Table4). Examples of representative food and beverage items included in each food group were provided for each item. Estimated cut-off points were used to determine the threshold frequency of consumption to consider the respondent as an HPF consumer for the item, as shown in the column “Criteria for 1 point” in Table4. For the self-reported version of the questionnaire, this column should be removed from the questionnaire form since it is intended for the use of the person assessing the score only. Using baseline data from the PREDIMED-Plus FFQ, the percentage of HPF over total grams per day according to the questionnaire items was calculated for each patient. This was used as the dependent variable in a linear regression with the sQ-HPF score obtained for each patient to establish the following equation for the regression line: “HPF consumption (% g/ day) = (3.7 x sQ-HPF score) + 7.6”. This equation allows the estimation of the HPF consumption from the sQ-HPF score, as shown at the bottom of Table4. Weighted κ coefficients were calculated between HPF consumption tertiles according to the sQ-HPF and the four existing classification systems (Supplementary Table 2). The highest agreement was for the comparison with UNC tertiles (κ = 0.88), followed by IFIC tertiles (κ = 0.65) and IARC tertiles (κ = 0.50). The comparison of tertiles according to the sQ-HPF with NOVA tertiles of HPF showed a fair agreement (κ = 0.36). These comparisons were in accordance with the corresponding agreement plots (Fig.2). Dietary, lifestyle andcardiometabolic characteristics ofPREDIMED‑Plus participants according tothesQ‑HPF score We aimed to investigate whether changes in HPF consumption as estimated from the FFQ and the four different systems were also detected when the sQ-HPF was Table 2 Associations between candidate sQ‑HPF items and tertile agreement variable by binomial logistic regression Binomial logistic regression adjusted for age, sex, recruitment center, energy intake, physical activity level, medication for blood pressure and diabetes, working status, educational level, and civil status. Tertile agreement is the outcome variable (score 0 – “low HPF consumer” if a subject is classified in T1 of HPF consumption by at least two classification systems; score 1 – “high HPF consumer” if a subject is classified in T3 by at least two classification systems). Food groups expressed in frequency of consumption (times/day). HPF: highly processed food. Items selected for the sQ‑HPF are shown in bold. Bonf: Bonferroni adjustment. N= 4400 Predictor variable β SE p (Bonf) Fatty dairy products 0.33 0.10 0.024 Sugary dairy products 1.37 0.22 < 0.001 Cured meat 0.59 0.07 < 0.001 Fats 0.63 0.14 0.001 Fermented alcohol 1.59 0.07 < 0.001 Distilled alcohol 3.90 0.45 < 0.001 Sugary drinks 3.90 0.17 < 0.001 Sweets 0.46 0.05 < 0.001 Snacks 3.93 0.43 < 0.001 Ready To Eat (RTE) 2.91 0.33 < 0.001 Refined cereals 0.30 0.04 < 0.001 Sauces 0.97 0.18 < 0.001 Additives 0.10 0.02 < 0.001 Fried food 0.84 0.15 < 0.001 Whole dairy products −0.01 0.06 1 Semi‑skimmed dairy products − 0.15 0.03 < 0.001 Eggs −0.31 0.19 1 Red meat 0.05 0.15 1 White meat −1.41 0.17 < 0.001 White fish −1.40 0.16 < 0.001 Blue fish −1.29 0.15 < 0.001 Vegetables −0.55 0.03 < 0.001 Fruit −1.01 0.04 < 0.001 Potatoes −0.88 0.18 < 0.001 Nuts −0.87 0.07 < 0.001 Legumes −2.46 0.23 < 0.001 Oils −0.21 0.03 < 0.001 Non‑sugary drinks −0.09 0.03 0.302 Wholegrain cereals −0.53 0.04 < 0.001 Vitamin/supplements −0.10 0.12 1 Binge eating 0.17 0.14 1 N° binges/week 0.05 0.05 1 Snacking 0.06 0.08 1 Table 3 Exploratory factor analysis for sQ‑HPF items Bartlett’s test of sphericity =p< 0.001 Measure of Sample Adequacy (MSA) = 0.78 Standardized Cronbach’s α = 0.67 N= 4400 Factor 1: HPF diet Factor loadings Ready To Eat (RTE) 0.45 Fried food 0.44 Fermented alcohol 0.42 Snacks 0.42 Fatty dairy products 0.38 Sugary drinks 0.37 Distilled alcohol 0.36 Sauces 0.35 Refined cereals 0.35 Cured meat 0.34 Sweets 0.3 Sugary dairy products 0.29 Additives 0.26 Fats 0.23 Page 8 of 16 Martinez‑Perezetal. Int J Behav Nutr Phys Act (2022) 19:6 used. In addition, we wanted to analyze the dietary profile across tertiles of the sQ-HPF score, since this is a tool to evaluate a particular dietary pattern. Dietary characteristics of PREDIMED-Plus participants at baseline are shown in Table5, grouped by HPF consumption tertiles according to the score obtained in the sQ-HPF (T1 – lowest score, T3 – highest score). Consumption of food groups included in the questionnaire was the highest for subjects in T3 of the sQ-HPF score, which was not the case for vegetables, fruits, and legumes. Dairy products and fish consumption were not different among tertiles. With all the classification systems, subjects in T3 showed the highest HPF consumption (percentage over total grams per day: 12.45 ± 8.51 for NOVA, 54.85 ± 11.01 for IARC, 29.26 ± 12.02 for IFIC and 30.34 ± 12.40 for UNC, 35.05 ± 12.24 for questionnaire estimated HPF consumption). Associations between lifestyle and cardiometabolic parameters and HPF consumption have been previously reported, so we next aimed to analyze if we could detect changes in these variables across tertiles of the sQ-HPF score. Lifestyle and cardiometabolic characteristics of PREDIMED-Plus participants at baseline grouped by HPF consumption tertiles showed differences according to the sQ-HPF score (Table6). Participants ranked in the highest tertile had higher levels of triglycerides (161.05 ± 90.55 mg/dL), higher weight (90.68 ± 13.10 kg) and waist circumference (109.83 ± 9.45 cm) compared to those in the lowest tertile. No significant changes among tertiles were detected in fasting glucose and total cholesterol levels. Subjects in T3 of the sQ-HPF score spent more time watching TV than T1 subjects (4.05 ± 2.02 h/ day), while sleeping hours were similar across tertiles. Around half of the subjects classified in T3 were classified as sedentary (51.5%). Concerning eating habits, 30.5% of the subjects with the highest sQ-HPF score reported snacking. Assessing longitudinal changes inHPF consumption withthesQ‑HPF Longitudinal analysis of HPF consumption estimated from the FFQ by each classification system revealed significant differences across the first 3 years of the PREDIMED-Plus study. In all cases, mean HPF consumption showed a trend towards a decrease that was lower in year 2 than baseline (Table7). This was also the case when Table 4 Screening questionnaire of highly processed food consumption (sQ‑HPF) What was your average frequency of consumption over the past year of… Criteria for 1 point Score (0 or 1) Q1. Fatty dairy products (cream, cured or semi‑cured cheese, processed soft cheese wedges)? > = 2 t/week Q2. Sugary dairy products (condensed milk, industrially produced milkshakes, flavored Petit Suisse yogurt, custard, crème caramel flan, pudding)? > 3 t/month Q3. Cured meats (serrano ham, sandwich (deli) ham, cured cold meats, pâté, bacon, marbling)? > = 1 t/day Q4. Fats (margarine, butter, lard (animal fat))? > 3 t/month Q5. Fermented alcohols (rosé wine, muscatel wine, young red wine, aged red wine, white wine, Spanish sparkling wine (cava), beer)? > 1 t/day Q6. Distilled alcohols (liquors, anisette, whisky, gin, vodka, cognac)? > 3 t/month Q7. Sugary and artificially sweetened drinks (soft drinks, artificially sweetened drinks, bottled juice, grape must)? > = 2 t/week Q8. Sweets (ice‑cream and sorbets, canned fruit in juice or syrup, biscuits, whole meal biscuits, chocolate biscuits, honey, homemade baking products, industrially produced confectionery, donuts, muffins, cupcakes, industrially produced cakes, churros, chocolates, soluble cocoa powder, nougat, marzipan, shortbread biscuits, jam)? > 1 t/day Q9. Snacks (packaged potato crisps/chips, packaged snacks)? > 3 t/month Q10. Ready To Eat products (pizza, croquettes, instant soup)? > 3 t/month Q11. Refined cereals (white and sliced bread, breakfast cereals, Spaghetti, macaroni, Spanish noodles, white rice)? > = 2 t/week Q12. Sauces (mustard, mayonnaise, tomato sauce, Ketchup)? > 1 t/week Q13. Additives (sugar, table salt)? > 3 t/day Q14. Fried foods (eat out and homemade)? > = 2 t/week TOTAL SCORE: Equivalency between sQ‑HPF score and the estimated percentage of HPF consumption over the total intake in grams per day: Score 1 2 3 4 5 6 7 8 9 10 11 12 13 14 % HPF 11.3 15 18.7 22.4 26.1 29.8 33.5 37.2 40.9 44.6 48.3 52 55.7 59.4 Page 9 of 16 Martinez‑Perezetal. Int J Behav Nutr Phys Act (2022) 19:6 HPF consumption was estimated through the sQ-HPF (22.9 ± 0.22% of g/day at baseline and 17.3 ± 0.22% of g/ day in year 2). Discussion This article presents the sQ-HPF, a short, integrative, and easy-to-use questionnaire to estimate HPF consumption (Table4). The development process involved a carefully designed statistical analysis and practical considerations for its use in a clinical/epidemiological context. This new tool includes 14 food and beverage items of which the frequency of consumption is recorded, based on a previously validated FFQ from the PREDIMED-Plus Trial [38–40]. Each item is scored as 1 if the frequency of consumption corresponds to the HPF dietary pattern, according to the calculated thresholds, and as 0 otherwise. Therefore, the higher the score, the higher the consumption of HPF. Moreover, this score can be used to estimate the percentage of HPF consumption over total intake, avoiding the need to administer a lengthy FFQ (Table4) [63]. The present work demonstrates that statistical approaches such as EFA and Cronbach’s analysis are valuable tools for developing integrative tools related to dietary patterns and eating habits. Indeed, analysis of questionnaire items through EFA demonstrated that they identified one core construct, the HPF dietary pattern (Table3) [64]. This was expected considering that items were selected based on a positive association with the variable tertile agreement through binomial logistic regression (Table 1). This resulted in 14 questions based on the frequency of consumption of food groups associated with a higher HPF consumption. One of the reasons for this was that, in this way, the questionnaire could identify people with an HPF dietary pattern with a focus on the frequency of consumption and not on the specific HPF items they consumed, so that the questionnaire could detect different HPF dietary patterns. Fig. 2 Agreement plots of tertiles of HPF consumption. Visual representation of contingency tables between the sQ‑HPF tertiles and A NOVA HPF tertiles, B IARC HPF tertiles, C IFIC HPF tertiles and D UNC HPF tertiles. Marginal totals of the contingency table are located on the top and right axis. Shading represents the level of agreement, black indicates “perfect agreement”, and grey indicates “partial agreement”. The extent to which rectangles deviate from the diagonal line of no bias indicates the extent of disagreement, and the position (above/below) indicates direction of the disagreement. HPF: highly processed food Page 16 of 16 Martinez‑Perezetal. Int J Behav Nutr Phys Act (2022) 19:6 • fast, convenient online submission • thorough peer review by experienced researchers in your field • rapid publication on acceptance • support for research data, including large and complex data types • gold Open Access which fosters wider collaboration and increased citations maximum visibility for your research: over 100M website views per year • At BMC, research is always in progress. Learn more biomedcentral.com/submissions Ready to submit your research Ready to submit your research ? Choose BMC and benefit from: ? Choose BMC and benefit from: 62. Cronbach LJ. Coefficient alpha and the internal structure of tests. Psycho‑ metrika. 1951;16(3):297–334 [cited 2021 Feb 22]. Available from: https:// link. sprin ger. com/ artic le/ 10. 1007/ BF023 10555. 63. Fangupo LJ, Haszard JJ, Leong C, Heath ALM, Fleming EA, Taylor RW. Relative validity and reproducibility of a food frequency questionnaire to assess energy intake from minimally processed and ultra‑processed foods in young children. Nutrients. 2019;11(6):1–13. 64. Koslowsky M, Scheinberg Z, Bleich A, Mark M, Apter A, Danon Y, et al. The factor structure and criterion validity of the short form of the eating attitudes test. J Pers Assess. 1992;58(1):27–35 [cited 2021 May 14]. Avail‑ able from: https:// www. tandf online. com/ doi/ abs/ 10. 1207/ s1532 7752j pa5801_3. 65. Gibney MJ, Forde CG, Mullally D, Gibney ER. Ultra‑processed foods in human health: a critical appraisal. Am J Clin Nutr. 2017;106(3):717–24 [cited 2020 Jun 25]. Available from: http:// ajcn. nutri tion. org/ lookup/ doi/ 10. 3945/ ajcn. 117. 160440. 66. Nieuwenhuijsen MJ. Design of exposure questionnaires for epidemiologi‑ cal studies. Occup Environ Med. 2005;62(4):272–80 [cited 2021 May 11]. Available from: http:// oem. bmj. com/. 67. San‑Cristobal R, Navas‑Carretero S, Celis‑Morales C, Brennan L, Walsh M, Lovegrove JA, et al. Analysis of dietary pattern impact on weight status for personalised nutrition through on‑line advice: the food4Me Spanish cohort. Nutrients. 2015;7(11):9523–37. 68. San‑Cristobal R, Navas‑Carretero S, Celis‑Morales C, Livingstone KM, Stew‑ art‑Knox B, Rankin A, et al. Capturing health and eating status through a nutritional perception screening questionnaire (NPSQ9) in a randomised internet‑based personalised nutrition intervention: the Food4Me study. Int J Behav Nutr Phys Act. 2017;14(1):1–12. 69. Trichopoulou A, Costacou T, Bamia C, Trichopoulos D. Adherence to a Mediterranean diet and survival in a Greek population. N Engl J Med. 2003;348(26):2599–608 [cited 2021 May 10]. Available from: https:// pubmed. ncbi. nlm. nih. gov/ 12826 634/. 70. Tertsunen H‑M, Hantunen S, Tuomainen T‑P, Virtanen JK. Adherence to a healthy Nordic diet and risk of type 2 diabetes among men: the Kuopio Ischaemic heart disease risk factor study. Eur J Nutr. 2021; [cited 2021 May 10];Online pub. Available from: https:// link. sprin ger. com/ 10. 1007/ s00394‑ 021‑ 02569‑1. 71. Toh DWK, Xia X, Sutanto CN, Low JHM, Poh KK, Wang J‑W, et al. Enhanc‑ ing the cardiovascular protective effects of a healthy dietary pattern with wolfberry (Lycium barbarum): a randomized controlled trial. Am J Clin Nutr. 2021; [cited 2021 May 10];Online ahe. Available from: https:// acade mic. oup. com/ ajcn/ advan ce‑ artic le/ doi/ 10. 1093/ ajcn/ nqab0 62/ 62726 07. 72. Paterson EN, Neville CE, Wallace SM, Woodside JV, Kee F, Young IS, et al. Dietary patterns associated with renal impairment in the Northern Ireland cohort for the longitudinal study of ageing (NICOLA). Eur J Nutr. 2021; [cited 2021 May 10];Published. Available from: https:// link. sprin ger. com/ 10. 1007/ s00394‑ 021‑ 02579‑z. 73. Guenther PM, Kirkpatrick SI, Reedy J, Krebs‑Smith SM, Buckman DW, Dodd KW, et al. The healthy eating Index‑2010 is a valid and reliable measure of diet quality according to the 2010 dietary guidelines for Americans 1‑3. J Nutr Methodol Math Model J Nutr. 2014;144:399–407 [cited 2021 May 10]. Available from: http:// jn. nutri tion. org. 74. Krebs‑Smith SM, Pannucci TRE, Subar AF, Kirkpatrick SI, Lerman JL, Tooze JA, et al. Update of the healthy eating index: HEI‑2015. J Acad Nutr Diet. 2018;118(9):1591–602 [cited 2021 May 14]. Available from: https:// pub‑ med. ncbi. nlm. nih. gov/ 30146 071/. 75. McCullough ML, Willett WC. Evaluating adherence to recommended diets in adults: the Alternate Healthy Eating Index. Public Health Nutr. 2006;9(1A):152–7 [cited 2021 May 14]. Available from: https:// pubmed. ncbi. nlm. nih. gov/ 16512 963/. 76. Food Standards Scotland. Briefing paper on discretionary foods. 2018. 77. National Health and Medical Research Council. Australian dietary guide‑ lines. Canberra; 2013. [cited 2021 May 10]. Available from: http:// www. nhmrc. gov. au 78. Grieger JA, Wycherley TP, Johnson BJ, Golley RK. Discrete strategies to reduce intake of discretionary food choices: a scoping review. Int J Behav Nutr Phys Act. 2016;13(1):1–22 [cited 2021 May 10]. Available from: https:// ijbnpa. biome dcent ral. com/ artic les/ 10. 1186/ s12966‑ 016‑ 0380‑z. 79. Livingstone KM, Celis‑Morales C, Navas‑Carretero S, San‑Cristobal R, Forster H, Woolhead C, et al. Personalised nutrition advice reduces intake of discretionary foods and beverages: findings from the Food4Me randomised controlled trial. Int J Behav Nutr Phys Act. 2021;18(1):70 [cited 2021 Jun 8]. Available from: https:// ijbnpa. biome dcent ral. com/ artic les/ 10. 1186/ s12966‑ 021‑ 01136‑5. 80. Phulkerd S, Lawrence M, Vandevijvere S, Sacks G, Worsley A, Tangcharoen‑ sathien V. A review of methods and tools to assess the implementation of government policies to create healthy food environments for preventing obesity and diet‑related non‑communicable diseases. Implement Sci. 2016;11(15) [cited 2021 May 11]. Available from: https:// pubmed. ncbi. nlm. nih. gov/ 26846 789/. 81. Sarbagili‑Shabat C, Zelber‑Sagi S, Fliss Isakov N, Ron Y, Hirsch A, Maharshak N. Development and validation of processed foods question‑ naire (PFQ) in adult inflammatory bowel diseases patients. Eur J Clin Nutr. 2020;74(12):1653–60 Available from: https:// doi. org/ 10. 1038/ s41430‑ 020‑ 0632‑5. 82. Motta VW de L, Lima SCVC, Marchioni DML, Lyra C de O. Food frequency questionnaire for adults in the Brazilian northeast region: emphasis on the level of food processing. Rev Saude Publica 2021;55(51). [cited 2021 Aug 24]. Available from: https:// www. revis tas. usp. br/ rsp/ artic le/ view/ 189620. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in pub‑ lished maps and institutional affiliations.