O iginal Resea ch
DIGITAL
HEALTH
La ge-scale loyal y ca d da a in heal h esea ch
Jaakko Ne alainen
1
, Maijaliisa E kkola
2
, Hannu Saa ija
¨ i
3
,
Tu kka Na
¨ppila
¨
1
and Mikael Fogelholm
2
Abs ac
Objec i e: To s udy he cha ac e is ics o la ge-scale loyal y ca d da a ob ained in Finland, and o e alua e hei po en ial
and challenges in heal h esea ch.
Me hods: We con ac ed he holde s o a ce ain loyal y ca d li ing in a speci ic egion in Finland ia email, and eques ed
hei elec onic in o med consen o ob ain hei basic backg ound cha ac e is ics and g oce y expendi u e da a om 2016
o heal h esea ch pu poses. Non-pa icipa ion and he cha ac e is ics and expendi u e o he pa icipan s we e mainly
analysed using summa y s a is ics and igu es.
Resul s: The da a on expendi u e came om 14,595 (5.6% o hose con ac ed) consen ing loyal y ca d holde s. A o al o
68.5% o he pa icipan s we e women, wi h an a e age age o 46 yea s. Women and esiden s o Helsinki we e mo e likely
o pa icipa e. Bo h young and old pa icipan s we e unde ep esen ed in he sample. We obse ed ha annual expendi u e
ep esen ed oughly wo- hi ds o he na ionally es ima ed annual a e ages. Cus ome s and pe sonnel di e ed in hei
cha ac e is ics and expendi u e, bu no so much in hei mos equen ly bough i ems.
Conclusions: Loyal y ca d da a om a majo e aile enabled us o each a la ge, he e ogeneous sample wi h ewe
esou ces han con en ional su eys o he same magni ude. The po en ial o he da a was g ea because o hei size,
co e age, objec i i y, and long pe iods o dynamic da a collec ion, which enables imely in es iga ions. The challenges
included bias due o non-pa icipa ion, pu chases in o he s o es, he le el o de ail in p oduc g ouping, and he knowl-
edge gaps in wha is being consumed and by whom. Loyal y ca d da a a e an unde u ilised esou ce in esea ch, and could
be used no only in e aile s’ ac i i ies, bu also o socie al bene i .
Keywo ds
Heal h beha iou , loyal y ca d da a, ood expendi u e, pa icipa ion, pu chases, sales da a, SWOT
Submission da e: 23 May 2018; Accep ance da e: 6 No embe 2018
In oduc ion
E idence o he ela ionship be ween heal h beha iou
and ch onic diseases is well es ablished.
1
The isk ac-
o s o diseases – unheal hy die , smoking and alcohol
consump ion – a e an inequali y conce n, as hey a e
clus e ed in he popula ion.
2
The e o e, i is essen ial o
each hose a g ea es isk o de eloping ch onic con-
di ions such as obesi y, ype 2 diabe es and ca dio as-
cula diseases. Popula ion heal h beha iou moni o ing
and e alua ion play a c i ical ole in unde s anding and
add essing hese challenges. I is impe a i e o ha e
ele an , eliable and imely in o ma ion on he isk
ac o s de i ed om alid, ye cos -e ec i e, sou ces.
O all heal h beha iou s, die migh be ega ded as
he mos challenging o assess, due o i s subs an ial
a iabili y in quan i y and quali y. Die a y da a
1
Heal h Sciences/Facul y o Social Sciences, Uni e si y o Tampe e,
Tampe e, Finland
2
Depa men o Food and Nu i ion, Uni e si y o Helsinki, Helsinki, Finland
3
Business S udies/Facul y o Managemen , Uni e si y o Tampe e,
Tampe e, Finland
Co esponding au ho :
P o esso Jaakko Ne alainen, Heal h Sciences/Facul y o Social Sciences,
A o Ylp€
on ka u 34, FI - 33014 Uni e si y o Tampe e, Finland.
Email: jaakko.ne [email p o ec ed]i
Digi al Heal h
Volume 4: 1–10
!The Au ho (s) 2018
A icle euse guidelines:
sagepub.com/jou nals-
pe missions
DOI: 10.1177/2055207618816898
jou nals.sagepub.com/home/dhj
C ea i e Commons Non Comme cial CC BY-NC: This a icle is dis ibu ed unde he e ms o he C ea i e Commons A ibu ion-NonComme cial
4.0 License (h p://www.c ea i ecommons.o g/licenses/by-nc/4.0/) which pe mi s non-comme cial use, ep oduc ion and dis ibu ion o he wo k
wi hou u he pe mission p o ided he o iginal wo k is a ibu ed as speci ied on he SAGE and Open Access pages (h ps://us.sagepub.com/en-us/nam/
open-access-a -sage).
collec ion ins umen s, such as ood equency ques-
ionnai es and ood dia ies, a e associa ed wi h la ge
andom a ia ion and sys ema ic biases because o
imp ecision in ecall and epo ing, as well as wi h en-
dencies o epo heal hie die a y habi s han hose
ac ually p ac ised.
3,4
I is also known ha popula ion
su eys ypically unde es ima e alcohol consump ion.
5
Many die a y assessmen me hods (e.g. ood eco ds
and die in e iews) a e limi ed o sho pe iods o mea-
su emen , which makes hei alidi y o assessing long-
e m heal h beha iou poo . The s eng hs and he
weaknesses o he me hods a e comp ehensi ely p e-
sen ed elsewhe e.
6,7
Heal h- ela ed su eys ha e gene ally indica ed
declining ends in esponse a es.
8
Mo e speci ically,
he Finnish D inking Habi s Su ey epo ed a
esponse a e o 60% in 2016 (97% in 1968), and US
da a om se en Na ional Alcohol Su eys showed a
esponse a e o 52% in 2010 (77% in 1995).
9,10
The
Na ional Findie Su ey on ood consump ion also ol-
lowed he same long- e m end o dec easing pa ici-
pa ion a es: 60–63% in 1982 and 2007, and 57% in
2012.
11–13
In e es in o e coming hese challenges by al e na-
i e means o da a collec ion has g own.
14
The ield o
digi al epidemiology s udies how big da a – social
media da a, wea able de ices, Twi e da a and GPS
acking, o example – can be used o e ec i ely
add ess public heal h p oblems.
15–17
Howe e , inno a-
i e echnologies do no necessa ily o e come he
inhe en indi idual bias ela ed o sel - epo ed die a y
and alcohol in ake. Many measu emen s (e.g. using
wea able de ices) p o ide high- esolu ion da a on a
speci ic indi idual, bu in o ma ion on he popula ion
is limi ed, since his kind o da a collec ion is esou ce
in ensi e. On he o he hand, ins umen s ha p o ide
da a on many indi iduals end o p o ide li le in o -
ma ion on he indi idual le el and/o lack ele an
s uc u e (e.g. social media da a). Ideally, al e na i e
da a sou ces should be su icien ly ‘big’ in e ms o
bo h he numbe o indi iduals and he amoun o
measu emen s, and impo an ly, p o ide unbiased
de ails on hese indi iduals.
Loyal y ca d da a po en ially possess hese ea u es.
By loyal y ca ds we mean elec onic cus ome ca ds
used in g oce y e ailing, which au oma ically egis e
g oce y expendi u e pe pu chased i em e e y ime he
cus ome swipes hei ca d a he shop. The incen i e is
a inancial ewa d in exchange o da a wi h he e ail-
e . Cus ome loyal y p og ammes a e widely used
among e aile s, wi h he aim o inc easing cus ome
loyal y. Fo example, in he UK, nea ly 90% o e ail
cus ome s belong o a leas one cus ome loyal y p o-
g amme.
18
By analysing he da a, e aile s can build
cus omised ma ke ing communica ions, iden i y he
mos p o i able cus ome s, es ablish ele an segmen a-
ion and consume p o iles, and p omo e complemen-
a y o mo e expensi e p oduc s o cus ome s on he
basis o hei p e ious pu chases.
19
P io s udies ha e
indeed obse ed an associa ion be ween ca d owne -
ship and loyal y.
20,21
An ob ious p oblem wi h loyal y ca d da a om a
esea ch pe spec i e is he dis ibu ion o he pu chases
o e se e al e aile s. This p oblem may, howe e , no
always be as es ic i e as i i s seems. The No dic
ma ke s a e highly cen alised: he h ee la ges
ma ke chains claim an a e age o 80–90% o he
ma ke sha e.
22
In Finland, he la ges comme cial
ope a o (S G oup) had a ma ke sha e as high as
47.2% in 2017, which enabled in es iga ion o la ge
popula ion g oups based on da a om a single e ail-
e .
22
This, combined wi h he ac ha in Finland he
loyal y ca d up ake among consume s is he highes o
all he Eu opean coun ies,
23
p o ides a ela i ely eli-
able means wi h which o e alua e na ional-le el ood
in ake ia loyal y ca d da a. In addi ion, au oma ed
egis a ion o expendi u e and he long du a ion o
da a collec ion make he assessmen o die , smoking
and mild alcohol consump ion less subjec i e and ee
om ecollec ion e o . Tin e al.
24
e iewed 18 s udies
using supe ma ke sales da a o a ious popula ion
ood and nu i ion moni o ing pu poses. Thei indings
suppo he easibili y o using supe ma ke sales da a
o moni o a popula ion’s ood pu chasing pa e ns.
Finally, i is possible ha unplanned and unheal hy
pu chases – including alcohol and ciga e es – a e
made in smalle s o es. In Finland, since he opening
hou egula ion (2016), smalle e aile concep s ha e
become mo e compe i i e han kiosks. Pu chases made
in hese smalle s o es a e also eco ded o loyal y ca d
da abases, which educes possible bias due o di e en-
ces in he expendi u e p o iles o la ge and small s o es.
The aim o his pape is o in oduce a esea ch
p ojec add essing he po en ial and he challenges o
loyal y ca d da a o heal h esea ch. Based on la ge-
scale loyal y ca d da a ob ained om Finland, we p e-
sen ou da a collec ion p ocess, as well as he i s
empi ical esul s ega ding pa icipan and pu chasing
p o iles. In he discussion, we also p o ide a SWOT
(s eng hs, weaknesses, oppo uni ies, h ea s) analysis
o he alue o loyal y ca d da a.
Me hods
Se ing and ec ui men
The loyal y ca d da a we e p o ided by he S G oup
(S- yhm€
a), a majo Finnish e aile co-ope a i e ope a ing
in Finland, Russia and he Bal ic coun ies (h ps://www.
s-kana a. i/web/s/en/s- yhma-lyhyes i). We con ac ed he
2DIGITAL HEALTH
ca d holde s (owne s o an S G oup loyal y ca d) ia
email and asked o elec onic in o med consen o
ob ain selec ed backg ound cha ac e is ics (age, gende
and esiden ial pos al code) and g oce y expendi u e
da a conce ning hem om 1s o Janua y o 31s o
Decembe , 2016, o esea ch pu poses, wi hou pe sonal
iden i ie s.
An in i a ion o he s udy, wi h an elec onic con-
sen o m, was emailed o 245,877 cus ome s (‘cus om-
e s’ sample) (i) wi h a known email add ess, (ii) who
had g an ed pe mission o be app oached wi h esea ch
que ies, (iii) who we e a leas 18 yea s o age, and (i )
who held a loyal y ca d o he HOK-Elan o e ail co-
ope a i e (S G oup) ope a ing in he Sou he n Finland
egion, eaching om he capi al ci y o Helsinki abou
50 km no h. The o al numbe o cus ome s in he
HOK-Elan o da abase was 915,797, and geog aphical-
ly hey we e mainly loca ed in Helsinki and nine nea by
municipali ies. The ‘pe sonnel’ sample was based on
email con ac wi h 13,763 S G oup employees h ough-
ou Finland. A he ime o he que ies, he S G oup
had 40,482 employees. The emails we e sen in Ap il
2017 and he collec ion o elec onic in o med consen s
ended in May 2017.
E hical issues
E hical app o al was ob ained om he Uni e si y o
Helsinki Re iew Boa d in humani ies and social and
beha iou al sciences. We commi ed o ollowing he
‘Responsible conduc o esea ch and p ocedu es o
handling allega ions o misconduc in Finland’ (The
Finnish Ad iso y Boa d on Resea ch In eg i y). Bo h
he esea ch g oup and he S G oup signed a con ac
on da a ans e , ensu ing he independence o he
esea ch and scien i ic publishing om busi-
ness in e es s.
S a is ical analysis
The backg ound cha ac e is ics and expendi u e we e
mainly analysed desc ip i ely using summa y s a is ics
and igu es. To analyse he de e minan s o o al expen-
di u e om S G oup s o es using he indi idual-le el
da a a ailable, we applied a linea mixed model. The
model was based on he cus ome sample and included
age g oup (ca ego ised as <25, 25–29, 30–34, ...,
80–84, 85– yea s), gende and hei in e ac ion as
ixed e ec s, and pos al code as a andom e ec o
e lec spa ial he e ogenei y.
To assess he ex en o he simila i y o he cus ome
sample o he gene al popula ion in he same egion, we
ob ained he age–gende dis ibu ion o he 10 munic-
ipali ies in he HOK-Elan o egion om S a is ics
Finland. We could no iden i y any ele an e e ence
popula ion o he pe sonnel sample. We hen modelled
he p obabili y o being a HOK-Elan o loyal y ca d
holde and pa icipa ion among all esiden s o he
egion applying a logis ic eg ession model, which
enabled he in es iga ion o he dis ibu ion o he
dicho omous ou come a iable (pa icipan , yes/no)
on se e al explana o y a iables. We included gende ,
age, gende by age g oup in e ac ion, and municipali y
in he model as ixed e ec s. In his analysis, o ma ch
he a ailable s a is ics, he age g oups we e de ined as
16–19, 20–24, 25–29, ..., 80–84, 85–. F om he model-
ling esul , we used in e se p obabili y weigh ing o
adjus he sample o he egional popula ion and o
educe he non-pa icipa ion bias in expendi u e.
Resul s
Sample cha ac e is ics
The da a on expendi u e consis ed o 14,595 consen ing
loyal y ca d holde s (5.6% o hose con ac ed). We
ob ained he backg ound da a o 14,522 loyal y ca d
holde s al oge he : 13,274 cus ome s (5.4%) and 1248
membe s o pe sonnel (9.1%). Supplemen al igu es 1
and 2 illus a e he geog aphical dis ibu ion o he
pa icipan s. The majo i y o he cus ome pa icipan s
li ed in Sou he n Finland as expec ed, bu he pe son-
nel sample was sp ead h oughou he coun y.
Employees we e included in he sample o o e wide
pe spec i e in o na ionwide ood pu chases and as hey
o m a loyal and in e es ing socio-economic subg oup,
which is on he on -line acing changes in he labou
ma ke due o digi alisa ion.
O e all, mo e han wo- hi ds o he pa icipan s
we e women (Table 1). The gende dis ibu ion was
pa icula ly skewed in he pe sonnel sample, o which
80.5% we e women.
The a e age age o he pa icipan s was 45.7 yea s
(Table 1). No su p isingly, he pe sonnel sample pa -
icipan s ended o be younge han hose in he cus-
ome sample; he a e age age di e ence being
app oxima ely six yea s. The cus ome sample con-
ained a subs an ial p opo ion o olde people: 25%
o he pa icipan s we e a leas 58 yea s old, and 10%
we e a leas 67 yea s old. The same pe cen iles we e 49
and 56 yea s in he pe sonnel sample, espec i ely.
The popula ion age–gende dis ibu ion in he 10
municipali ies showed ha women we e mo e likely
o pa icipa e in he s udy han men. Municipali y s a-
is ics showed ha 52.1% o he HOK-Elan o egion
esiden s, who we e a leas 15 yea s old, we e women.
Al hough he mean age o he esiden s was simila
(46.2 yea s) o ha in he sample, we obse ed ha
young and old pa icipan s we e unde ep esen ed
(Supplemen a y Table 1). Residen s o Helsinki we e
Ne alainen e al. 3
he mos likely o pa icipa e. All ac o s o he model
we e highly signi ican (p<0.0001; da a no shown).
Expendi u e da a and de e minan s
Expendi u e da a con ained he o al g oce y expendi-
u e o >13 million pu chase e en s, which we e p e-
classi ied in o 184 p oduc g oups and accompanied by
a imes amp: da e o pu chase and ime o day. Each
pu chase was eco ded in eu os.
The o e all means o o al annual expendi u e we e
2443 EUR and 4419 EUR in he cus ome and pe son-
nel samples, espec i ely. When co ec ed o he age
and gende dis ibu ion a each municipali y, he
weigh ed mean o he cus ome s was lowe : 2322
EUR. In he cus ome sample, we obse ed ha bo h
gende and age we e signi ican de e minan s o o al
expendi u e (p<0.0001). Mean expendi u e peaked a
middle age and declined owa ds bo h ends o he age
ange (Figu e 1). Men’s expendi u e ended o be g ea -
e han women’s, bu he gende di e ence diminished
in he 35–50-yea age ange (Figu e 1; p¼0.016 o he
gende by age in e ac ion e m). We also no ed some
a ia ion due o pos al code a eas (p<0.0001).
Howe e , his a ia ion was no la ge. The high p o-
po ion o women and less widely sp ead age dis ibu-
ion in he pe sonnel sample did no acili a e he same
analysis o de e minan s as ha o he cus ome s. The
igh panel in Figu e 1, howe e , sugges s simila pa -
e ns o hose in he cus ome sample. Mos no ably,
he mean annual expendi u e was subs an ially la ge
in his sample, which is p obably due o he mo e
a ac i e pe sonnel ewa d sys em o an addi ional
3% discoun .
When anked by eu os spen annually, we obse ed
ha he 10 mos popula p oduc g oups we e he same
in he wo samples, and hei expendi u e pa e ns we e
simila (Figu e 2). In e es ingly, bee and ciga e e
p oduc g oups we e among hese, and he mean ciga-
e e expendi u e o he pe sonnel sample was app ox-
ima ely wice ha o he cus ome sample.
Cyclic pu chasing beha iou du ing he week was
e iden (Figu e 3); F idays and Sa u days we e he
mos ac i e pu chasing days, and Sundays and
Tuesdays he leas ac i e. Na ional holidays appea ed
o be p eceded by a peak in expendi u e. ‘Skiing holi-
day week’ was in week 8 in 2016, and he associa ed
decline in expendi u e is obse able in Figu e 3.
Discussion
Ou aim was o in oduce a esea ch p ojec ha would
add ess he po en ial o loyal y ca d da a o heal h
esea ch. The p ojec was based on la ge-scale loyal y
ca d da a ob ained in Finland: al oge he 14,595 loyal-
y ca d holde s consen ed o eleasing hei e ospec-
i e expendi u e da a.
Pa icipa ion in he s udy
In ecen decades, pa icipa ion a es in heal h- ela ed
su eys in Finland, as in mos o he coun ies, ha e
declined.
25
In ou s udy, he pa icipa ion a e ell
well below hose o comp ehensi e su eys, and was
highes among pe sonnel, women, middle-aged
people, and cus ome s li ing in Helsinki. Se e al ea-
sons could explain his. Fi s , i is possible ha emails
Table 1. Obse ed backg ound cha ac e is ics o cus ome s and pe sonnel. No backg ound da a we e a ailable o n¼73 pa icipan s.
Gende Age (a 1.1.2016)
n(%)
Numbe o dis inc
pos al codes Female Male Mean (SD) Range
Cus ome s 13,274 (91.4%) 655 8937 (67.3%) 4336 (32.7%) 46.2 (14.7) 16–90
Pe sonnel 1248 (8.6%) 669 1004 (80.5%) 244 (19.6%) 40.1 (11.6) 18–74
O e all 14,522 (100%) 962 9941 (68.5%) 4580 (31.5%) 45.7 (14.5) 16–90
6000
4000
2000
0
20 40 60 80 20 40 60 80
Cus ome s Pe sonnel
Age (yea s)
Mean annual expendi u e (in EUR)
Gende : Male Female
Figu e 1. Mean annual expendi u e by age and gende in he wo
samples. Obse ed age–gende speci ic means a e deno ed by
ma ke s, and solid lines indica e weigh ed loess cu e i s.
4DIGITAL HEALTH
do no each e e yone who is in i ed, because
add esses may be ou da ed, o he in i a ion could be
di ec ed o he junk olde . Second, an in i a ion by
email is easy o dele e, o ge o igno e, pa icula ly i
i does no aise he immedia e in e es o he ecipien .
Thi d, he ecipien s could indeed be less willing o
sha e de ailed and e ospec i e expendi u e da a on
hemsel es han o p o ide answe s o su eys. The
de e minan s o he pa icipa ion a es a e pa ly in
con as wi h he esul s o he FINRISK su eys con-
duc ed in 1997–2012, in which women li ing in he
capi al egion we e mo e likely o be non-pa icipan s
in su eys han women li ing in u al a eas.
25
In su -
eys, he non-pa icipan s a e mo e likely o be young,
male, unma ied, ha e a lowe educa ion le el and
income, and o o eign o igin, ac o s ha a e also
associa ed wi h less op imal heal h beha iou .
8,26
As
we did no ob ain he backg ound cha ac e is ics o
e e yone in i ed, we could no di ec ly compa e he
non-pa icipan s and he pa icipan s. Howe e , we
could ob ain summa y measu es among HOK-Elan o
egion loyal y ca d owne s as a whole, including hose
who did no ul il he in i a ion c i e ia: hei mean age
was 46.0 yea s and 59.4% o hem we e women. This
pa ly explains he di e ence in he gende dis ibu ion
be ween he esiden s (o whom 52.1% we e emale)
and he cus ome s (67.3%), bu no comple ely. The
mean age ell e y close o he means o he esiden s
and o he cus ome s. Howe e , he a ailabili y o an
a e age measu e does no shed ligh on he obse ed
unde ep esen a ion o he young and old pa icipan s
in he sample.
We had limi ed access o socio-demog aphic meas-
u es. Howe e , we expec ed ou pe sonnel sample o be
less educa ed han a e age Finnish employees, as 13%
o people wo king in he e ail sec o ha e a low-le el
(less han high school), 62% a mid-le el (high school,
oca ional school), and 25% a high-le el (BSc deg ee
o highe ) educa ion.
27
The espec i e p opo ions
among all Finnish employees we e 10%, 45% and
45%. Despi e he appa en di e ence be ween he edu-
ca ion equency dis ibu ion o he wo samples, we
obse ed a highe pa icipa ion a e in he pe sonnel
sample han in he cus ome sample.
0 100 200 300
200000
150000
100000
50000
0
Day o he yea
Expendi u e (in EUR)
Figu e 3. To al expendi u e o cus ome s du ing calenda yea .
Daily sums (g ey solid line) and cen ed se en-day mo ing a e age
(black solid line). The e ical e e ence lines a e posi ioned a
Good F iday, 1s o May, Midsumme E e and Ch is mas E e.
100000 300000 500000 20000 40000 60000 80000
200
150
100
50
Numbe o imes pu chased
Annual mean expendi u e (EUR)
Cus ome s Pe sonnel
CHEESE
CHEESE
CIGARETTES
CIGARETTES
MILK AND CREAM
MILK AND CREAM
READY-TO-EAT MEALS
READY-TO-EAT MEALS
BEER
BEER
FRESH BREAD
FRESH BREAD
POULTRY
POULTRY
VEGETABLES
VEGETABLES BEEF
BEEF
SOFT DRINKS, WATER
SOFT DRINKS, WATER
Figu e 2. Mos pu chased p oduc g oups ( op 10, anked by expendi u e) among cus ome s and pe sonnel.
Ne alainen e al. 5

Ou s udy shows ha loyal y ca d da a o e he
oppo uni y o each a la ge, po en ially mo e he e o-
geneous sample wi h a less esou ces han o he su -
eys o he same magni ude. Howe e , he
pa icipa ion a es we e low, and he da a did no
seem o ill all known pa icipa ion gaps, o example,
hose ela ed o age, and we canno ule ou he possi-
bili y o he socio-demog aphic g adien in non-
pa icipa ion. The e o e, in s udies such as ou s, wi h
poo pa icipa ion a es bu cos e ec i ely ob ained
la ge sample sizes, i would be e en mo e impo an
han in su eys o collec he a iables ela ed o non-
pa icipa ion in o de o a leas pa ially co ec any
biases ha may esul om his (e.g. by in e se p ob-
abili y o pa icipa ion weigh ing). Un o una ely, his
is no always possible, and such a iables may di e
om hose ha a e p edic i e o non-pa icipa ion in
su eys, which means ha hey would ha e o be iden-
i ied in ad ance. Pa allel in e es in he de elopmen o
me hods o co ec ing o da a missing no a andom
may u n ou o be use ul o la ge-scale da a wi h
selec i e pa icipa ion.
28,29
In he u u e, inc easing popula ion di e si y along-
side declining ends in esponse a es will challenge
la ge na ional die a y su eys, as a ep esen a i e
sample will become e en ha de o ec ui , and he ap-
idly changing supply o ood p oduc s will equi e com-
p ehensi e die a y assessmen me hods and
con inuously upda ed ood composi ion da abases. A
he same ime, digi al de elopmen allows easy access
o expendi u e da a wi h a ailable de ailed in o ma-
ion, enabling he de ec ion o di e ences in he die s
o di e en popula ion g oups. Howe e , his o en
equi es addi ional in o ma ion on cus ome s, a ype
o da a ha is mo e di icul o ob ain di ec ly om
e aile s due o da a con iden iali y. A sepa a e ques-
ionnai e on backg ound a iables o linkage wi h
o he ele an da ase s could p o ide addi ional
socio-demog aphic measu es, which would inc ease
he alue o he da a.
Expendi u e p o iles and hei de e minan s
We obse ed ha he op p oduc s we e he same in he
wo samples, e en hough he pa icipan s in he pe -
sonnel sample pu chased conside ably mo e om he S
G oup. The o al expendi u e does no seem o be
s ongly associa ed wi h he mos commonly pu chased
p oduc s, which could mean ha he pu chases om
o he e aile s a e simila o hose in ou da a.
Demons a ing a high p opo ion o expendi u e
wi hin he e aile ’s s o es combined wi h simila ela-
i e dis ibu ion o pu chased p oduc s ac oss di e en
le els o e aile -speci ic expendi u e add o he
c edibili y o he da a. Howe e , we canno empi ically
ule ou he possibili y o di e ences be ween he le els.
In o ma ion on household size and membe s was
no a ailable, bu i is likely ha he numbe o chil-
d en and adul s in a household could la gely explain
he di e ences be ween age g oups in mean annual
expendi u e.
The mean o ciga e e expendi u e o he pe sonnel
sample was app oxima ely wice ha o he cus ome
sample. This could be pa ly due o he lowe socio-
economic s a us o he pe sonnel and hence he highe
p e alence o smoke s, o because hey pu chase a
la ge sha e o hei g oce ies and ciga e es in he
s o e whe e hey wo k. The a ia ion due o pos al
code a eas could be indica i e o socio-economic di e -
ences o due o he geog aphical loca ion o he s o es.
S eng hs
One o he clea undamen al s eng hs o loyal y ca d
da a is he ac ha hey a e no based on pe cep ion:
wi hou isk o imp eciseness o o e - o unde es ima-
ion, loyal y ca d da a cap u e he kind o i ems a pa -
icula pe son has bough om a speci ic e aile . They
use e aile s’ exis ing da a in as uc u e, cash egis e s,
and IT in as uc u e o collec as amoun s o da a
ha accumula e con inuously o e ime. Loyal y ca d
da a can be ega ded as a by-p oduc : hey esul om
e aile s’ exis ing p ocesses and can be sha ed o
heal h esea ch pu poses wi h ma ginal addi ion-
al cos s.
The shee size o loyal y ca d da a is a clea asse . I
no only allows in es iga ion o he whole sample, bu
also s udies smalle subpopula ions such as he elde ly,
on whom comp ehensi e da a a e o en ha d o ob ain.
As cus ome loyal y p og ammes accumula e da a
on a household le el, loyal y ca d da a au oma ically
p o ide longi udinal insigh in o how ood expendi u e
e ol es o e ime, gi en ha he sha e and he compo-
si ion o ood expendi u e o a pa icula e aile does
no change. T adi ional me hods a e o en limi ed o
c oss-sec ional da a o ocus on sho e ime pe iods
(e.g. one mon h). In con as o loyal y ca d da a, con-
sume panel da a (e.g. Nielsen Homescan) co e all
pu chases o a ce ain ime pe iod bu may su e
om selec ion bias o mo e a luen popula ions and
mis epo ing, simila o adi ional die a y assess-
men me hods.
Weaknesses
Cus ome s o en pu chase ood om di e en sou ces.
Some may p e e buying ege ables om ma ke places,
mea s om p ima y e aile s, and b ead om a local
bake y. Simila ly, as cus ome s may di ide hei
6DIGITAL HEALTH
g oce y shopping be ween se e al e aile s, analysing
da a om only one e aile could gi e an incomple e
pic u e o household-le el expendi u e. This is one lim-
i a ion o loyal y ca d da a in ma ke s wi h many ood
e aile s, such as he UK. Howe e , in he No dic
coun ies in pa icula , whe e he le el o ma ke con-
cen a ion in g oce y e ailing is high, loyal y ca d da a
ob ained om e en a single e aile could p o ide a
good pic u e o household expendi u e. S a is ics
Finland epo s (p elimina y da a om 2016) ha he
a e age annual consump ion o g oce ies and non-
alcoholic d inks is 2916 EUR pe household, and o
alcohol and ciga e es 578 EUR pe household.
30
Ou
es ima es all close o hese na ional benchma ks.
Al hough indi ec , hey a e indica i e o a e y high
p opo ion o expendi u e among loyal y ca d holde s
in S G oup s o es. In e es ingly, he obse ed means o
pe sonnel exceeded he na ional es ima es. This may
indica e ha pe sonnel hea ily concen a e all hei
shopping in S G oup s o es due o pe sonnel discoun s.
Al hough loyal y ca d da a cap u e wha is bough ,
hey canno di ec ly e eal exac ly wha is e en ually
consumed and by whom. Consume s li ing in he same
household may ea (o d ink) ou , sha e dinne s wi h
iends and ela i es, o buy ood o pe s o o he s no
li ing in he same household. Loyal y ca d da a he e-
o e only pa ially e lec s consump ion by household
membe s. Good compa ibili y be ween esponden -
collec ed household-le el ood pu chase da a (co e ing
all pu chases) and indi idual-le el die a y da a has
been demons a ed,
31–33
and hus, household-le el
ood pu chase da a can easonably eliably be used o
model indi idual-le el die a y pa e ns i all pu chases
a e eco ded. In addi ion, consume s wi h speci ic
socio-economic backg ounds may be p one o speci ic
ypes o e aile s. Fo example, he cus ome s o a p e-
mium ood e aile o a ha d discoun e may o e
hea ily biased consump ion pa e ns in ela ion o
bo h ood consump ion and heal hiness. The e o e, o
add ess his weakness, loyal y ca d da a should be
compa ed wi h popula ion da a whene e possible.
We a e cu en ly planning a alida ion s udy ha will
add ess hese limi a ions mo e closely.
The alue o loyal y ca d da a is highly dependen
on da a accu acy. Fo example, loyal y ca d da a can
be collec ed on a o al sum le el (i.e. expendi u e o
43.50 EUR), p oduc ca ego y le el (24.20 EUR
wo h o ege ables), o p oduc le el (4.20 EUR
wo h o ca o s). Howe e , ood expendi u e da a
may no be g ouped in o clea ly dema ca ed unc ional
ca ego ies, which would be essen ial o linking wi h
heal h da a. In nu i ional science, commonly used
ood g oupings a e mainly based on ea lie indings
ega ding he associa ions be ween die a y componen s
and heal h. The use o di e en ca ego ies weakens
compa abili y be ween s udies. Too aguely o inco -
ec ly ( om he esea ch poin o iew) ca ego ised
da a – such as plan -based p o ein p oduc s being clas-
si ied in o he same ca ego y as whole mea p oduc s –
e en in la ge olumes, ha e li le alue. In his espec ,
access o loyal y ca d da a does no necessa ily mean
high heal h esea ch po en ial.
Oppo uni ies
Loyal y ca d da a may o e a unique, unde u ilised
da a sou ce o heal h esea ch ha signi ican ly com-
plemen s exis ing da a sou ces. Second, u ilising loyal y
ca d da a o he bene i o heal h esea ch could
unco e he socie al po en ial o cus ome loyal y
ca d da a o collec i e bene i . In addi ion o hei
business alue, he da a can be used as a esou ce o
c ea ing socie al alue. This could also o e companies
ano he way in which o ul il hei co po a e social
esponsibili y: Th ough hei own ac ions, hey can
le e age exis ing da a asse s o socie al bene i .
Thi d, combining loyal y ca d da a wi h o he da a
sou ces could u he ampli y hei po en ial o any
esea ch. Fo example, linking loyal y ca d da a wi h
heal h ou come da a, ei he on he indi idual o egion-
al le el, could o e new pe spec i es o he associa ion
be ween speci ic diseases and die / ood (un)heal hiness.
This could be done using da a om na ional heal h
su eys o medicine s a is ics conduc ed in he same
egion, which could be linked by, o example, pos al
code a ea. This s a egy would enable ecological anal-
ysis o he associa ion be ween egional expendi u e
and heal h ou comes. Ano he possibili y would be o
app oach he loyal y ca d owne s di ec ly h ough su -
eys o in i a ions o heal h examina ions. Howe e ,
his could p o e o be di icul in p ac ice. Fou h, loy-
al y ca d da a may p o ide a lens h ough which
esea che s can e alua e he impac o a ious heal h
p omo ion campaigns and policy in e en ions (e.g.
axes) on ac ual ood consump ion. Fo example, in
close collabo a ion wi h ood e aile s, esea che s
could design a se o in e en ions o guide consume s
owa ds heal hie die s, and h ough loyal y ca d da a
e alua e hei impac on di e en cus ome g oups.
This would help us unde s and wha ypes o in e en-
ions, mechanisms o incen i es in luence consume s’
ood consump ion.
In addi ion o hese oppo uni ies, Tin e al.
24
iden-
i ied o he a eas o nu i ion moni o ing ha could use
supe ma ke sales da a: he assessmen o ood pu -
chase pa e ns wi hin and be ween popula ion g oups,
longi udinal compa ison o popula ion ood pu chase
pa e ns wi h ega d o policy o economic changes, he
de i a ion o nu ien a ailabili y by linking ood pu -
chase da a o ood composi ion da a, he alida ion o
Ne alainen e al. 7
sel - epo ed die a y da a by compa ing ac ual ood
pu chases wi h epo ed pu chases, and he assessmen
o key de e minan s o he heal hie ood choices o a
popula ion. We add o his he possibili y o ailo ing
pe sonal eedback by using cus ome s’ usual ood
expendi u e and hence in luencing hei ood pu -
chases. Ideally, an online eedback sys em could
enable a cus ome o check he die a y quali y o
ca bon oo p in o hei own expendi u e. Such in e -
ac ion be ween he cus ome , esea che s and he e ail-
e would subs an ially add o he public heal h alue o
he da a. In New Zealand, indi idualised elec onic
supe ma ke sales da a we e used o ailo cul u ally
a ge ed nu i ion esou ces o e hnically di e se shop-
pe s in a la ge supe ma ke ial.
34
The po en ial o pu chase da a would be g ea e i
hey we e eliably linked wi h o he ele an da ase s,
and wi h co ec ion ac o s o p o ide alid es ima es
o consumed ood om bough ood.
24
Keeping up
wi h he mode n ood landscape equi es sys ema ic,
meaning ul linkages ac oss da a on ood pu chase/
sales, ood in ake, ood composi ion, and nu i ion
ac panels.
35
Ng and Popkin
35
concluded ha he abil-
i y o esea che s and nu i ion p o essionals o p op-
e ly in eg a e di e en da a sou ces and o ully
capi alise on he a ising oppo uni ies emains some-
wha unde eloped. Howe e , he numbe o s udies
based on pu chase da a supplemen ed wi h o he da a
sou ces is inc easing. A ecen Danish s udy
36
com-
bined comp ehensi e household ood pu chase da a
wi h nu i ion in o ma ion and indi idual egis e
da a when assessing he e ec o e ime o unemploy-
men on ood pu chase beha iou and die composi-
ion. The s udy p o ided aluable me hodological
examples o be u ilised wi h simila da ase s. The
Die -Rela ed GHG Index has also been modelled on
he basis o he same Danish household ood pu chase
da a and a ood equency ques ionnai e.
33
Howe e ,
digi al epidemiology is s ill in i s ea ly s age o de el-
opmen . Me hodological obus ness is an e hical as
well as a scien i ic equi emen , in ol ing, o example,
he alida ion o algo i hms, a be e unde s anding o
con ounding, il e ing sys ems o noisy da a, manage-
men o biases, and he selec ion o app op ia e
da a s eams.
37
Th ea s
In he pas decade, da a p i acy has become an impo -
an issue o bo h companies and consume s; ecen
discussion on he issue may ha e made consume s mo e
awa e o possible misuses o da a. Fo example, in
e ms o social media usage, se e al epo s and news
i ems ha e highligh ed how consume s should be mo e
conce ned abou wha da a a e being collec ed on
hem.
22,38,39
In gene al, mos consume s a e conce ned
abou he da a collec ed on hem on he in e ne . In
pa icula , ecen da a p i acy issues wi h Facebook,
which linked consume s’ social media usage o a ge ed
poli ical ad e isemen s may ha e spillo e e ec s on
o he o ms o cus ome da a usages, including loyal y
p og ammes. Howe e , wi h w i en consen app o ed
by an e hical boa d, pu chase da a could be u ilised o
he common good in a anspa en manne , while
espec ing indi idual igh s and libe ies, which is he
c ux o he deba e on he e hics o big da a.
37
F om a company’s poin o iew, cus ome da a
ha e become a key componen o almos any business.
Consequen ly, da a p i acy and p o ec ion ha e
become a ocal compe ence a ea. The Gene al Da a
P o ec ion Regula ion p o ides new guidelines o
o ganisa ions in he Eu opean Union o imp o ing
he managemen o isks and p ac ices ela ed o cus-
ome da a. The egula ion may also limi e aile s’
willingness o sha e da a o hi d pa ies, including
esea che s, and companies may become a aid o he
nega i e publici y ha could esul om allowing
esea che s o in es iga e hei cus ome da a.
Howe e , sha ing aluable da a o e s a company he
oppo uni y o ca y socie al esponsibili y.
Conclusions
Mo e a en ion needs o be paid o he po en ial o
loyal y ca d da a o cus ome s, esea ch o socie y a
la ge, a he han o e aile pu poses only.
40
In his
pape , we ha e in oduced a unique da ase o 14595 S
G oup ( e ail ma ke s) loyal y ca d holde s. These da a
consis o all ood pu chases, spanning o e a pe iod o
one yea , in o al >13 million pu chase e en s. They
accumula e au oma ically, wi h high esolu ion, and
wi hou pa icipan -based epo ing bias. Al hough
he po en ial o hese kinds o digi al big da a is con-
side able, challenges also a ise: 1) he da a collec ed
ep esen household pu chases o mos cus ome s, 2)
only a pa o all ood pu chases is om one single
e ail chain; and 3) possible u u e es ic ions and con-
ce ns ega ding indi idual da a p i acy. In concen a -
ed ma ke s such as hose in he No dic coun ies, hese
esou ces a e ne e heless highly in iguing o
esea che s. In he u u e, e o s o o e come hei
limi a ions, and balancing size wi h su icien le el o
de ail could lead o new iewpoin s ega ding popula-
ion exposu e, beha iou and li es yle.
Acknowledgemen s: We hank he S G oup o hei
collabo a ion.
Con ibu o ship: JN, ME and MF concei ed he s udy. JN
w o e he i s d a o he manusc ip and conduc ed he da a
8DIGITAL HEALTH
analysis. ME and HS esea ched he li e a u e. TN was
esponsible o da a managemen and pa icipa ed in he
da a analysis. All au ho s e iewed and edi ed he manusc ip
and app o ed he inal e sion.
Con lic o in e es s: The au ho s decla e no con lic s
o in e es .
E hical app o al: The e hics commi ee o he Uni e si y o
Helsinki Re iew Boa d in he humani ies and social and
beha iou al sciences app o ed his s udy (REC numbe :
#43/2016).
Funding: This esea ch ecei ed no speci ic g an om any
unding agency in he public, comme cial, o no - o p o-
i sec o s.
Gua an o : JN
Pee e iew: This manusc ip was e iewed by Cliona
Mhu chu, The Uni e si y o Auckland, New Zealand, and
one o he indi idual who has chosen o emain anonymous.
ORCID iD
Jaakko Ne alainen h p://o cid.o g/0000-0001-6295-0245
Tu kka N€
appil€
ah p://o cid.o g/0000-0002-0562-7254
Supplemen al Ma e ial: Supplemen al Ma e ial o his
a icle is a ailable online.
Re e ences
1. Wo ld Heal h O ganiza ion. Global ac ion plan o he
p e en ion and con ol o noncommunicable diseases
2013–2020. Gene a: WHO, h p://www.who.in /nmh/
e en s/ncd_ac ion_plan/en/ (2013, accessed 1
Feb ua y 2018).
2. Schui AJ, an Loon AJM, e al. Clus e ing o li es yle
isk ac o s in a gene al adul popula ion. P e Med
2002; 35:219–224.
3. Wille W. Co ec ion o he e ec s o measu emen
e o . In: Nu i ional Epidemiology. Ox o d, UK:
Ox o d Uni e si y P ess, 2013.
4. Benne DA, Land y D, e al. Sys ema ic e iew o s a-
is ical app oaches o quan i y, o co ec o , measu e-
men e o in a con inuous exposu e in nu i ional
epidemiology. BMC Med Res Me hodol 2017; 17:146.
5. Li ings on M and Callinan S. Unde epo ing in alcohol
su eys: whose d inking is unde es ima ed? J S ud
Alcohol D ugs 2015; 76:158–64.
6. Lo eg o e JA, Hodson L, Sha ma S, e al. Nu i ion
esea ch me hodologies. Hoboken, NJ: Wiley
Blackwell, 2015.
7. Wille W. Nu i ional epidemiology. 3 d ed. Ox o d, UK:
Ox o d Uni e si y P ess, 2012.
8. Tolonen H, Helako pi S, Talala K, e al. 25-yea ends
and socio-demog aphic di e ences in esponse a es:
Finnish adul heal h beha iou su ey. Eu J Epidemiol
2006; 21: 409–415
9. H€
a k€
onen J, Sa onen J, Vi ala E, e al. Suomalais en
alkoholink€
ay €
o a a 1968–2016: Juoma apa u kimus en
uloksia [In Finnish, abs ac in English]. Helsinki,
Finland: Yliopis opaino, h ps://www.julka i. i/bi
s eam/handle/10024/134585/URN_ISBN_978-952-302-
873-9.pd ?sequence¼1 (2018, accessed 23 Feb ua y 2018).
10. Rossow I, M€
akel€
a P and Ke W (2014). The collec i i y
o changes in alcohol consump ion e isi ed. Addic ion
2014; 109:1447–1455.
11. Pie inen P, Va iainen E, Sepp€
anen R, e al. Changes in
die in Finland om 1972 o 1992: impac on co ona y
hea disease isk. P e Med 1996; 25: 243–250.
12. Pa u i M, Tapanainen H, Reini uo H, e al. (eds).The
Na ional FINDIET 2007 Su ey [In Finnish, ables, ig-
u es and summa y in English]. Publica ions o he
Na ional Public Heal h Ins i u e B23/2008. Helsinki:
Yliopis opaino, h p://www.julka i. i/bi s eam/handle/
10024/78088/2008b23.pd (2008, accessed 01
Ma ch 2018).
13. Hellda
´n A, Raulio S, Kosola M, e al. The Na ional
FINDIET 2012 Su ey [In Finnish, ables, igu es and
summa y in English]. Publica ions o he Na ional
Ins i u e o Heal h and Wel a e 16/2013. Tampe e,
Finland: Yliopis opaino, h ps://www.julka i. i/bi
s eam/handle/10024/110839/THL_RAP2013_016_%
26slii ee .pd (2013, accessed 23 Feb ua y 2018).
14. Illne AK, F eisling H, Boeing H, e al. Re iew and e al-
ua ion o inno a i e echnologies o measu ing die in
nu i ional epidemiology. In J Epidemiol 2012;
41: 1187–1203.
15. Signal LN, Smi h MB, Ba M, e al. Kids’ Cam: an
objec i e me hodology o s udy he wo ld in which chil-
d en li e. Am J P e Med 2017; 53: e89–e95.
16. Sala he
´M, Beng sson L, Bodna TJ, e al. Digi al epide-
miology. PLoS Compu Biol 2012; 8: e1002616.
17. Lee EC, Ashe JM, Goldlus S, e al. Mind he
scales: Ha nessing spa ial big da a o in ec ious
disease su eillance and in e ence. J In ec Dis 2016;
214: S409–13.
18. Tonini T. Consume ’s pe cep ion o digi al ewa ds in
loyal y p og ams: insigh s om a mul i-coun y esea ch,
2nd Edi ion, Ta ge Resea ch. h p://www.imaeu ope.
com/wp-con en /uploads/2017/05/Ta ge Resea ch.pd
(2017, accessed May 15, 2018).
19. Peppe s D, Roge s M and Do B. Is you company
eady o one- o-one ma ke ing? Ha a d Business
Re iew 1999; 77: 151–160.
20. Meye -Waa den L. The in luence o loyal y p og amme
membe ship on cus ome pu chase beha iou . Eu J
Ma ke ing 2008; 42: 87–114.
21. Tu ne JJ and Wilson K. G oce y loyal y: Tesco
Clubca d and i s impac on loyal y, B i Food J 2006;
108: 958–964.
22. S a is a. S a is a: he po al o s a is ics, h p://www.s a
is a.com (2018, accessed 15 Feb ua y 2018).
23. Nielsen. Sha e o consume s ha ha e loyal y ca ds in
selec ed Eu opean coun ies in 2016, h ps://www.s a
Ne alainen e al. 9