En i onmen In e na ional 197 (2025) 109335
A ailable online 15 Feb ua y 2025
0160-4120/© 2025 The Au ho s. Published by Else ie L d. This is an open access a icle unde he CC BY-NC license (h p://c ea i ecommons.o g/licenses/by-
nc/4.0/).
Full leng h a icle
Iden i ica ion o exposome clus e s based on socie al, social, buil and
na u al en i onmen – esul s o he ABCD coho s udy
Klaus Telkmann
a,b,*
, Helene Gudi-Minde mann
a,b
, Rik Boge s
c
, Jenny Ah ens
a,b
,
Jus us T¨
onnies
a,b
, I ene an Kamp
c
, Tanja V ijko e
d
, Gab iele Bol e
a,b
a
Uni e si y o B emen, Ins i u e o Public Heal h and Nu sing Resea ch, Depa men o Social Epidemiology, 28359 B emen, Ge many
b
Uni e si y o B emen, Heal h Sciences B emen 28359 B emen, Ge many
c
Na ional Ins i u e o Public Heal h and he En i onmen , Cen e o Sus ainabili y, En i onmen and Heal h, 3720 BA Bil ho en, he Ne he lands
d
Uni e si y o Ams e dam, Ams e dam UMC, Ams e dam Public Heal h Resea ch Ins i u e, Depa men o Public and Occupa ional Heal h, 1081 BT Ams e dam, he
Ne he lands
ARTICLE INFO
Handling Edi o : D . Xa ie Que ol
Keywo ds:
Exposome
Clus e
Men al heal h
Social inequali ies
En i onmen
ABSTRACT
Exposome esea ch has seen a ecen inc ease. The concep ual amewo k o he Social Exposome ex ends ini ial
concep s by conside ing he en i e y o socie al, social, buil and na u al en i onmen al exposu es which a e
assumed o holis ically impac de elopmen and heal h ac oss he li ecou se. The aim o his s udy is he
iden i ica ion and cha ac e isa ion o exposome clus e s. Addi ionally, hei ele ance o men al heal h is
in es iga ed. To his end 2,850 pa icipan s aged 11–12 o he Ams e dam Bo n Child en and hei De elopmen
(ABCD) Coho S udy we e analysed. The exposome was cha ac e ized by 60 a iables ep esen ing he socie al,
social, buil and na u al en i onmen . Uni o m mani old app oxima ion and p ojec ion (UMAP) was applied o
dimensionali y educ ion, and subsequen ly clus e ing was pe o med on he e ie ed low-dimensional
embedding. Men al heal h symp oms and beha iou ela ed ou comes we e assessed by he S eng h and Di i-
cul ies Ques ionnai e (SDQ) as well as he Subs ance Use Risk P o ile Scale (SURPS). The esul s sugges ha
exposome clus e s a e mainly d i en by con ex ual socioeconomic and physical cha ac e is ics such as neigh-
bo hood income and dep i a ion a he han social cha ac e is ics a he indi idual le el. Mo eo e , p e alence
o child en’s men al heal h p oblems was mo e p ominen wi hin exposome clus e s cha ac e ized a he
con ex ual le el by mo e dep i ed neighbo hoods and a he indi idual le el by highe p e alence o ma e nal
men al heal h p oblems. This explo a o y exposome clus e iden i ica ion emphasized he ele ance o socio-
economic neighbo hood cha ac e is ics, hus s uc u al inequali ies.
1. In oduc ion
The e is a s eady inc ease in in e es in he exposome pe spec i e on
he li ing en i onmen o indi iduals and popula ions, s essing he
simul aneous in e play o a ious exposu es, including biological,
chemical, physical, and social ac o s and hei holis ic e ec on heal h
(Wild 2012; V ijheid 2014; Ve meulen e al. 2020; Vineis e al. 2020;
Mai e e al. 2021; Moo e e al. 2022). In con as o epidemiological
s udies analysing only one o a ew exposu es a a ime, his app oach
acknowledges he complexi y and co-occu ence o a mul i ude o ex-
posu es and hei collec i e ele ance o e iology o heal h and disease.
Recen s udies again emphasized he impo ance o assess he complex
in e play be ween a di e si y o dynamic en i onmen al exposu es
(Pea son e al. 2022; Tu e al. 2023; Xu e al. 2023). Fo di e en con-
di ions and heal h p oblems, speci ic exposomes ha e been p oposed
(A ias e al. 2022; Beulens e al. 2022; Chen e al. 2023; E zin e al.
2023). These exempla y app oaches di e ega ding he conside a ion
o condi ion speci ic aspec s o mul iple exposu es, such as ood en i-
onmen (diabe es), childhood auma (schizoph enia) o alcohol/ o-
bacco consump ion (colo ec al cance ). All hese app oaches ha e in
common ha physical en i onmen al exposu es o en ecei e g ea e
p ominence compa ed o social en i onmen al exposu es. This ocus on
physical, biological, and chemical exposu es has been c i icized wi h
well- ounded a gumen s o a sys ema ic conside a ion o socie al, social
and psychosocial exposu es in exposome esea ch (Jua ez e al. 2014;
Senie e al. 2017; Neu cou e al. 2022; Vineis and Ba ouki 2022).
* Co esponding au ho a : Uni e si y o B emen, Ins i u e o Public Heal h and Nu sing Resea ch, Depa men o Social Epidemiology, 28359 B emen, Ge many.
E-mail add ess: [email p o ec ed] (K. Telkmann).
Con en s lis s a ailable a ScienceDi ec
En i onmen In e na ional
jou nal homepage: www.else ie .com/loca e/en in
h ps://doi.o g/10.1016/j.en in .2025.109335
Recei ed 28 Oc obe 2024; Recei ed in e ised o m 17 Janua y 2025; Accep ed 14 Feb ua y 2025
En i onmen In e na ional 197 (2025) 109335
2
Cu en ly, nine comp ehensi e collabo a i e esea ch p ojec s on he
human exposome a e unded by Ho izon 2020 o he Eu opean Union
(Benjdi e al. 2021; Me ino Ma inez e al. 2021; Vlaande en e al. 2021;
V ijheid e al. 2021; Laiho e al. 2022; P onk e al. 2022; Ronkainen e al.
2022; Ronsmans e al. 2022; an Kamp e al. 2022), building he Eu-
opean Human Exposome Ne wo k EHEN (EHEN 2024). One o hese
esea ch p ojec s, Equal-Li e, s ands ou o add essing he public heal h
p io i y o ackling heal h inequi ies in child en ( an Kamp e al. 2022;
Faye e al. 2024). Wi hin Equal-Li e, we ha e in oduced a no el con-
cep ual amewo k o he Social Exposome as a pa adigm o in eg a ing
socie al, social and physical exposu es a indi idual and con ex ual
le els and o add ess he p e ious unde ep esen a ion o socie al and
social dimensions in exposome esea ch (Gudi-Minde mann e al. 2023).
The key ocus o he concep ual amewo k o he Social Exposome is on
unde s anding he unde lying mechanisms ha ansla e socie al and
social exposu es in o heal h ou comes. In pa icula , insigh s om
esea ch on heal h equi y and en i onmen al jus ice ha e been inco -
po a ed o unco e how social inequi ies in heal h eme ge, a e main-
ained, and sys ema ically d i e heal h ou comes. To b ing he no el
amewo k o he Social Exposome o p ac ice, in his s udy, we used
da a om one o he coho s comp ised in Equal-Li e, namely he
Ams e dam Child en and hei De elopmen (ABCD) coho s udy ( an
Eijsden e al. 2011). Ou p ima y aim was o iden i y exposome clus e s
wi h an inno a i e explo a o y s a is ical app oach o be able o
comp ehensi ely cha ac e ize child en’s li ing en i onmen by in e-
g a ing socie al, social, en i onmen al pollu ion, buil and na u al
en i onmen al indica o s a indi idual and con ex ual le els wi h a
heo e ical ounda ion. Impo an ly, exposome clus e s we e iden i ied
solely based on hese ac o s and independen o heal h- ela ed ou -
comes. As an illus a ion o how he exposome migh a ec heal h
ou comes, we linked he impac o he holis ically assessed li ing en i-
onmen and social de e minan s o men al heal h. Thus, in a second
analysis s ep, associa ions be ween exposome clus e s and indi idual
men al heal h ou comes we e in es iga ed.
2. Me hods
2.1. S udy popula ion
The ABCD coho s udy is a long- e m s udy o some 8,000 child en
obse ed a se en phases om p egnancy o adul hood which s a ed in
2003 (see Fig. S1 in he supplemen o a imeline o he di e en pha-
ses). I s aim is he analysis o ea ly li e ac o s po en ially impac ing la e
heal h p oblems ( an Eijsden e al. 2011).
The analysis in his pape ocused on da a on men al heal h p oblems
collec ed in phase 4 o he s udy, i.e. in 2015/2016 when he child en
we e 11–12 yea s old. Du ing he p e ious phases a ound 5,000 child en
we e los o ollow-up, leading o a sample size o 3,253 a phase 4. Da a
we e collec ed by ques ionnai es which we e illed ou by he child en
hemsel es as well as bo h pa en s, i applicable. The esponse a es o
child, mo he and a he we e 93.1 %, 93.2 % and 69.8 % espec i ely.
Since he pa e nal esponse a e was compa ably low, we es ic ed he
analysis o he mo he and child ques ionnai es. Consequen ly, 44 pa -
icipan s o whom only pa e nal in o ma ion was a ailable we e
excluded. Addi ionally, pa icipan s wi h unknown esiden ial add ess
we e emo ed om he da a, educing he sample size o 2,865. Based
on he geocodes o he esiden ial add ess, he da ase was en iched by
ac o s e lec ing he buil and na u al en i onmen (e. g. landuse, ai
pollu ion, noise, g eenspaces, a ailabili y o supe ma ke s, hospi als,
e c) as well as socioeconomic neighbo hood cha ac e is ics (e. g. mean
income, popula ion composi ion, age s uc u e, a ea dep i a ion)
(Lake eld e al. 2020). Fo many o he en i onmen al a iables o
in e es , da a we e only a ailable o phase 3 o da a collec ion (2008/
09) when he child en we e 5–6 yea s old. The e o e, he indica o s we e
linked o esiden ial add esses a phase 3. Some neighbo hood cha ac-
e is ics collec ed a baseline (2004) could also be linked o esiden ial
add esses om 2011 assuming ha hese indica o s did no unde go any
adical changes o e he nex ollowing yea s. These include in o ma ion
on popula ion densi y, income, pe cen age o men and women, pe -
cen age o households wi h and wi hou child en, unemploymen and
social bene i s. 15 pa icipan s wi h in o ma ion a ailable o less han
50 % o he i ems we e emo ed om he da a, leading o a inal sample
size o 2,850 obse a ions.
Neighbo hood cha ac e is ics we e a ailable on he spa ial esolu-
ion o 4-digi pos code a eas. The Ne he lands can be di ided in o 4,068
unique 4-digi pos code a eas, 85 o which a e loca ed in he munici-
pali y o Ams e dam. Pa icipan s’ esiden ial add esses belonged o 471
dis inc neighbou hoods de ined by 4-digi pos codes ac oss he
Ne he lands o which 86 % con ained only en o less obse a ions. 61 o
hese 471 neighbou hoods we e loca ed in he municipali y o Ams e -
dam. Ou o he 2,850 pa icipan s 1,654 we e s ill esiding in Ams e -
dam whe eas he es was loca ed ac oss all pa s o he Ne he lands. The
maximum numbe o pa icipan s belonging o he same a ea was 118.
2.2. Va iable selec ion
2.2.1. Exposome a iables
Clus e ing a iables we e selec ed based on he concep ual ame-
wo k o he Social Exposome acco ding o Gudi-Minde mann e al.
(2023) consis ing o se e al laye s. Fo his analysis, each a iable was
linked o one o he ollowing exposome dimensions. The inne di-
mensions “ac o s”, “ ela ional dynamics” and “places” comp ise he in-
di idual’s social in e ac ions and ne wo k. Fo ins ance, he dimension
“ac o s” ep esen s he pa ne s o social in e ac ion, including amily
membe s, pee s, o icial igu es and digi al con ac s. The dimension
“ ela ional dynamics” cap u es in e ac ions wi h o he indi iduals, eg.
social clima e, social capi al and medium o in e ac ion. “Places” con-
ains in o ma ion on whe e hese in e ac ions occu such as he home
en i onmen , places o leisu e ime and lea ning as well he digi al
wo ld. This dimension also en ails a di ec link o social and physical
neighbo hood cha ac e is ics, hus o con ex ual social, buil and na u al
exposu es. These h ee majo dimensions in u n a e embedded in he
b oade socie al con ex ep esen ed by he dimensions “socioeconomic
ci cums ances and sociodemog aphic cha ac e is ics” (SEC/SDC) and
“sys ems, ins i u ions and p io i ies o he poli ical economy”. Howe e ,
o he la e no in o ma ion was p esen in he da a. Finally, “cul u al
alues and social no ms” ha e been concep ualized by Gudi-Mind-
e mann e al. (2023) as pa o he b oade socie al con ex , which can
a ec all pa s o he concep , he e o e in luencing indi idual belie s,
a i udes, and beha io as well as social in e ac ions, and sys ems, in-
s i u ions, and p io i ies o he poli ical economy.
Va iables we e excluded om he analysis i hey con ained mo e
han 20 % missing alues. Addi ionally, a ca ego ical a iable o which
mo e han 97.5 % o he obse a ions ell in o a single ca ego y, was
conside ed oo spa se and uled ou as well. P io o clus e ing, co e-
la ions and associa ions be ween all a iables we e in es iga ed.
Namely, Pea son co ela ion o con inuous, Spea man co ela ion o
o dinal and o dinal wi h con inuous, C ame ’s V o ca ego ical (bo h
o de ed and uno de ed) and in aclass co ela ions o con inuous wi h
uno de ed ca ego ical ea u es we e calcula ed. I a iables showed an
absolu e co ela ion g ea e han 0.8, only one ep esen a i e was
selec ed. This was done since wo highly co ela ed a iables con ain
simila in o ma ion and, hus, his in o ma ion would be assigned a
la ge weigh by he dimension educ ion algo i hm. High co ela ion
coe icien s we e mainly obse ed o neighbo hood cha ac e is ics
included in he dimensions SEC/SDC and Places. Median alues o ab-
solu e co ela ions wi hin and be ween each o he exposome di-
mensions a e depic ed in Fig. 1.
A e checking he a o emen ioned exclusion c i e ia, a se o 60 ou
o 96 ini ially conside ed a iables emained. Fo example, a a iable
con aining in o ma ion o he ela ion be ween mo he and child (bio-
logical mo he , adop i e child, e c.) was emo ed since 99 % we e
K. Telkmann e al.
En i onmen In e na ional 197 (2025) 109335
3
biological mo he s. The as majo i y o excluded a iables we e
s ongly co ela ed geocoded a iables. The la ges block o co ela ed
o hose a iables was in o ma ion on a ailabili y o gene al p ac i-
ione s, hospi als, cinemas, swimming pools, schools, e c. wi hin a
ce ain dis ance. The inal selec ion o a iables i he dimensions o he
concep ual amewo k o he Social Exposome as ollows:
Rela ional dynamics: 16 a iables con aining in o ma ion on social
clima e such as pa en ing s yles, psychological dis ess, sa e y, ha mony
/ disha mony and media use. Pa en ing s yles o he mo he we e
assessed by he pa en ing s yles and dimensions ques ionnai e (PSDQ,
Robinson e al. 1995). PSDQ sum sco es a e used o dis inguish be ween
au ho i a i e, au ho i a ian and pe missi e pa en ing s yles based on
use o esponsi eness and demandingness. The ques ionnai e consis s o
32 i ems on a 4-poin Like scale. Highe alues on he au ho i a i e
scale (sco e om 15 o 60) a e conside ed a posi i e ou come whe eas
highe alues on he au ho i a ian (sco e om 12 o 48) and pe missi e
scale (sco e om 5 o 20) a e conside ed nega i e.
As a measu e o pa en al s ess expe ienced by he mo he he Nij-
meegse Oude lijke S ess Index (NOSI- K) was used, which is he Du ch
e sion o he Pa en al S ess Index (De B ock e al. 1990). This sum
sco e anges om 11 o 66 and is based on 11 i ems on a 6-poin Like
scale wi h highe alues ep esen ing highe le els o s ess. In ol e-
men o he mo he was assessed by he child ca e gi ing and in ol e-
men scale mean sco e (CCIS). I is based on eigh s a emen s a ed on a
5-poin Like scale wi h adjus men s o make he s a emen s mo e
applicable o 11–12 yea olds. Highe alues co espond o mo e
in ol emen (La sen e al. 2023).
Ma e nal men al heal h, assumed o ha e a s ong e ec on social
clima e wi hin he amily, has been assessed by he Dep ession, Anxie y
and S ess Scale based on 21 ques ions (DASS-21, Lo ibond and Lo i-
bond 1995). Each i em was answe ed wi h “(almos ) ne e ”, “some-
imes”, “o en” o “ e y o en” which ansla es o 0–3 poin s pe
s a emen . The h ee subscales dep ession, anxie y, s ess consis each o
7 o he 21 i ems and a sco e could only be de e mined i a leas 4 o he
7 s a emen s we e answe ed. All subsco es we e mul iplied by 2 o
achie e compa abili y o o he li e a u e elying on he ex ended
e sion o he DASS consis ing o 42 i ems. To gene a e ano he a iable
conce ning ma e nal men al heal h, mo he s we e asked o a e hei li e
on a scale om 1 o 10.
Sa e y and social cohesion sco es we e assessed as neighbo hood
le el indica o s by he “Lee baa ome e ” p o ided by he Du ch Minis y
o he In e io and Kingdom Rela ions. These sco es indica e he de i-
a ion om he a e age ac oss all neighbo hoods in he Ne he lands.
Mo eo e , a a iable coun ing he numbe o s ess ul e en s (such as
di o ce, dea h o a g andpa en , e c.) expe ienced by he child has been
included.
Ac o s: One a iable desc ibing wi h whom he child li es (bo h
pa en s, only mo he , only a he , e c).
Places: 19 a iables e lec ing he na u al and buil en i onmen
including social in as uc u e. These a iables included en i onmen al
exposu es, such as g eenspaces (measu ed by dis ance and NDVI alues),
ai pollu ion (in e ms o pa icula e ma e wi h a diame e o 10 µm o
less (PM10) and 2.5 µm o less (PM2.5) and ni ogen oxide (NOx)) o
a ic noise. Mo eo e , he deg ee o u baniza ion has been assessed
wi hin i e ca ego ies om e y u ban o u al. Simila ly, he buil
en i onmen has been e alua ed in e ms o densi y o buidings, acil-
i ies and in as uc u e wi hin 300 m bu e s a ound he esiden ial
add ess. Housing s ock, public spaces and esou ces in he neighbo hood
we e assessed by Lee baa ome e sco es.
Socioeconomic ci cums ances and sociodemog aphic cha ac-
e is ics: 22 a iables e lec ing hese dimensions based on indi idual
socioeconomic posi ion such as sel - a ed amily inancial si ua ion,
e hnici y o he pa en s, language p ima ily spoken a home and edu-
ca ion o he mo he as well as neighbo hood le el in o ma ion such as
popula ion densi y, a ea dep i a ion, mean neighbo hood income, age
composi ion o gene al social bene i s pe 1,000 households.
Cul u al alues / social no ms: Two a iables assessing whe he
ules conce ning media use and smoking and alcohol consump ion we e
imposed in he amilies.
An o e iew o all exposome a iables wi h in o ma ion on he
assigned exposome dimension, da a sou ce, linkage o esiden ial ad-
d esses and whe he i was used in dimension educ ion can be ound in
Table S1 in he Supplemen .
2.2.2. Men al heal h- ela ed ou comes
Child en’s men al heal h symp oms we e assessed wi h he S eng h
Fig. 1. Hea map o median alues o absolu e co ela ions wi hin and be ween he dimensions o he Social Exposome amewo k. Dimension “ac o s” no included,
since only one a iable was a ailable o his analysis.
K. Telkmann e al.
En i onmen In e na ional 197 (2025) 109335
4
and Di icul y Ques ionnai e (SDQ, Goodman 1997) as well as he
Subs ance Use Risk P o ile Scale (SURPS, Woicik e al. 2009).
The SDQ con ains i e subscales: emo ional symp oms, hype ac i i y
/ ina en ion, conduc p oblems, pee ela ionship p oblems and p o-
social beha io . Fo each subscale a sco e anging om 0 o 10 was
calcula ed based on 25 s a emen s which could be answe ed “no ue”,
“somewha ue” o “ ue”. An SDQ sum sco e be ween 0–40 was
calcula ed by adding all subscale sco es excep o he posi i ely con-
no a ed p osocial beha io . We in es iga ed p e alence o child en’s
men al heal h p oblems based on wo SDQ sco es: one based on he
mo he ques ionnai e and he o he sel -assessed by he child. All sco es
we e analysed in i s aw e sion and also in a ca ego ical e sion con-
sis ing o bands ep esen ing “no mal”, “bo de line” and “c i ical”
anges acco ding o Goodman (1997).
The SURPS measu es he ou pe sonali y dimensions hopelessness,
anxie y, sensi i i y/impulsi i y and sensa ion seeking o he child based
on 23 ques ions on a 4-poin Like scale, sel -assessed by he child. Each
dimension was assigned he mean alue o i s subi ems.
2.3. Clus e ing
To iden i y clus e s, dimensionali y educ ion o he co a ia e space
by uni o m mani old app oxima ion and p ojec ion (UMAP, McInnes
e al. 2018) was applied as a p ep ocessing s ep. Agglome a i e nes ing
hen de ec ed clus e s on he wo-dimensional embedding o he da a
(Kau man and Rousseeuw 2009). The in ol ed s eps a e desc ibed in
mo e de ail below.
UMAP is a no el machine lea ning echnique which embeds he high-
dimensional co a ia e space in o low-dimensional Euclidean space. This
concep bo ows and combines ideas om s a is ics and algeb aic o-
pology such as uzzy se s. Fo de ails we e e o McInnes e al. (2018).
The ad an ages o his algo i hm lie in i s compu a ional e iciency,
unde lying ma hema ical jus i ica ions and easy o in e p e isual
ou pu . Ano he eason o choose UMAP is ha s anda d echniques such
as p incipal componen analysis and la en class analysis a e no appli-
cable since da a is o mixed ype (con inuous and ca ego ical). UMAP is
o en compa ed o -dis ibu ed s ochas ic neighbo embedding ( -SNE,
Hin on e al., 2003). While UMAP pe o ms simila ly well in e ms o
p ese ing global s uc u e i ou pe o ms -SNE in e ms o compu a-
ional e iciency (Bech e al. 2019). Mo eo e , Allaoui e al. (2020) ha e
demons a ed ha applying UMAP p io o clus e ing can subs an ially
imp o e accu acy o s anda d clus e ing algo i hms as k-means, densi y-
based o agglome a i e clus e ing.
UMAP equi es as an inpu he k-nea es neighbo s o each obse -
a ion in he high-dimensional co a ia e space and ea anges hese
obse a ions in a wo-dimensional space such ha nea obse a ions s ill
appea close in he new embedding. Two pa ame e s a e c ucial o his
algo i hm: k is he numbe o nea es neighbo s used wi h a de aul
se ing o 15. In p inciple, smalle alues end o se ocus on he local
s uc u e o he da a while la ge alues emphasize he global s uc u e.
The pa ame e min_dis is a h eshold on he minimum dis ance and
de e mines how close poin s appea in he inal layou . Since he
objec i e is clus e ing we se his pa ame e close o ze o. A e checking
se e al combina ions o pa ame e s wi h k in he ange o 10–30 and
min_dis om 0.001 o 0.1, we se k =15 and min_dis =0.001. No
se e e isual di e ences we e de ec ed ac oss all se ups (up o o a ion
and symme y).
Since UMAP ope a es on nea es neighbo s, a me ic is needed o
de ine dis ances be ween obse a ions. As da a is o mixed ype (ca e-
go ical, o dinal and nume ic), Euclidean dis ance is no an app op ia e
me ic. A be e sui ed measu e is Gowe ’s dis ance (Gowe 1971) which
akes alues in he in e al [0,1]. Fo any wo m-dimensional ec o s o
obse a ions xi= (xi1,⋯,xim)and xj= (xj1,⋯,xjm) his me ic is
de ined by
d(xi,xj)=∑m
k=1
ω
k(xik,xjk)⋅dk(xik,xjk)
∑m
k=1
ω
k(xik,xjk)
whe e dk indi idually calcula es he dis ance o each single a iable.
Fo ins ance, he dis ance be ween obse a ions i and j on a nume ical
a iable is measu ed as he absolu e di e ence di ided by he ange. The
dis ance be ween ca ego ical ea u es is ei he 0 i hey ake he same
alue and 1 else. O dinal a iables a e so ed by ank and hen ea ed as
nume ical as p oposed by Kau man and Rousseeuw (2009). Mo eo e ,
he weigh s
ω
k a e ei he 1 i nei he xik no xjk a e missing and 0 else.
This ensu es ha o each wo obse a ions only he a e age o he
dis ances be ween pai wise comple e a iables is calcula ed. Ano he
ad an age o his me hod is ha i a oids a comple e case analysis o
needs o ely on impu a ion me hods.
Clus e ing was pe o med on he wo-dimensional ep esen a ion
e ie ed om UMAP by agglome a i e nes ing, a hie a chical clus-
e ing algo i hm. This is an unsupe ised ask since no a ge a iable o
clus e ing is de ined. The goal was o ind a s uc u e in he da a by
g ouping simila obse a ions.
2.4. Analysis o exposome clus e s conce ning child en’s men al heal h
p oblems
In a second analysis s ep, p e alence o child en’s men al heal h
p oblems was calcula ed o each iden i ied exposome clus e . Since his
is an explo a o y app oach, we e ain om assessing s a is ical signi -
icance o di e ences in heal h- ela ed ou comes be ween clus e s by
s a is ical es s such as ANOVA o pos -hoc - es s. We, he eby, ollow
he ecommenda ion by G eenland e al. (2016).
2.5. So wa e
All analyses we e ca ied ou wi h he s a is ical p og amming so -
wa e R e sion 4.3.1 (R Co e Team 2023). C ame ’s V was calcula ed
using c ame s in he con in package (Maye 2023), in aclass co e-
la ions by ICCba e in package ICC (Wolak e al. 2012). Gowe ’s dis ance
was calcula ed wi h gowe .dis in he S a Ma ch package (D’O azio
2022). UMAP elies on he unc ion umap in he package o he same
name (Konopka 2023). Clus e ing was pe o med wi h agnes om he
clus e package (Maechle e al. 2022). Maps o clus e dis ibu ion
ac oss Ams e dam we e gene a ed wi h he s package (Pebesma 2018).
3. Resul s
3.1. Clus e ing
The s udy popula ion comp ised 2,850 child en aged 11–12 yea s
(50.3 % gi ls, 49.8 % boys), li ing in neighbo hoods wi h a mean
dep i a ion sco e o 3.3 on a scale o 1 (leas dep i ed) o 5 (mos
dep i ed). 58 % o he child en li ed in Ams e dam and a ound 80 % o
he mo he s we e o Du ch e hnici y. A he neighbo hood le el, he
a e age numbe o unemploymen assis ance ecipien s pe 1,000 in-
habi an s was 27.67 and he mean annual income pe ecipien was
20,030
€
.
An applica ion o UMAP o he 60-dimensional co a ia e space,
ep esen ing he exposome in o ma ion, led o an a angemen o pa -
icipan s in wo-dimensional Euclidean space. Clus e s e ie ed by
agglome a i e nes ing based on his embedding a e shown in Fig. 2.
Recall ha he coo dina es o his embedding do no esemble any
geog aphical in o ma ion. Each poin ep esen s one pa icipan . Pa -
icipan s wi h he same o simila alues o he majo i y o exposu es
(and he e o e ha ing a small Gowe ’s dis ance) appea close in his
embedding han hose wi h dispa a e exposomic in o ma ion.
The ep esen a ion e ealed wo sepa a e la ge agglome a ions o
pa icipan s a he op and bo om le along wi h a smalle isola ed
K. Telkmann e al.
En i onmen In e na ional 197 (2025) 109335
5
clus e on he igh hand side. Agglome a i e nes ing wi h comple e
linkage was applied o e ie e a clus e ing ee. A co ela ion o 0.85
be ween he cophene ic dis ances o he clus e ing ee and he
Euclidean dis ances o obse a ions in he embedding indica ed high
alidi y o he clus e ing ee (Sokal and Rohl 1962). We hen chose o
cu he ee in o eigh clus e s based on isual inspec ion and o ensu e
ha sample sizes wi hin clus e s we e nei he oo small no oo big. This
led o a ine pa i ioning o he wo la ge amassmen s in o smalle
clus e s. The numbe ing o he clus e s was de e mined by hei occu -
ence in he clus e ing ee p oduced by agglome a i e nes ing.
3.2. Desc ip ion o clus e s
3.2.1. Exposome cha ac e isa ion o he clus e s
Close in es iga ion o he dis ibu ion o he Social Exposome a -
iables ac oss he wo-dimensional embedding e ealed ha he mos
disc imina ing ea u es we e neighbo hood cha ac e is ics such as
add ess densi y, deg ee o u baniza ion, a e age income pe esiden ,
pe cen age o immig an s and numbe o gene al social assis ance ben-
e i s pe 1,000 households. The e was also a clea dis inc ion be ween
child en li ing in he municipali y o Ams e dam and hose who mo ed
o di e en municipali ies al hough his in o ma ion was no used in he
UMAP algo i hm. Mo eo e , ma e nal men al heal h as measu ed by he
DASS-21 ques ionnai e showed subs an ial a ia ion ac oss clus e s. The
dis ibu ion o selec ed a iables ac oss he embedding is shown in
Fig. 3.
The la ge agglome a ion o obse a ions consis ing o clus e s 2, 3, 6
and 8 a he bo om o he embedding, mos ly consis ed o pa icipan s
li ing in he municipali y o Ams e dam (a leas 92–99 % pe clus e ).
Howe e , he e we e clea di e ences be ween he indi idual clus e s,
in pa icula a g adual inc ease in socioeconomic posi ion om le o
igh .
S a ing on he igh hand side o he bo om agglome a ion, clus e 3
was mos ly comp ised o child en li ing in densely popula ed, e y
u ban a eas (99 %) wi h lowes a ea dep i a ion (a e age quin ile 1.8 ±
1.0) e lec ed by he highes income pe esiden on he neighbo hood
le el (18,060 ±2,750
€
), lowes numbe o gene al social assis ance
bene i s pe 1,000 households (45.4 ±35.4), highes p opo ions o
wes e n immig an s (24.8 ±2.2 %), good in as uc u e and good sco es
o esou ces, public spaces, housing and popula ion composi ion. On
he o he hand, pa icipan s in his clus e we e also exposed o he
highes le els o ai pollu ion (in e ms o PM10, PM2.5 and NOx) as well
as noise. On he social indi idual le el, his clus e showed low DASS-21
sco es o anxie y (1.1 ±2.0) and s ess (5.7 ±5.1). Mo eo e , he
NOSI-K sum sco e o pa en al s ess (18.2 ±6.4) anked lowes
amongs all clus e s and he DASS-21 dep ession sco e (2.8 ±4.2)
anked second o lowes . Addi ionally, he majo i y o mo he s a ed
hei amily inancial si ua ion as good o e y good (78 %) and he
highes p e alence o mo he s wi h high educa ion (89 % uni e si y)
was obse ed. Fo child en in clus e 3, i was much less common o own
a mobile phone (8 % as compa ed o a popula ion a e age o 18 %).
Clus e 2 was e y simila o clus e 3 in e ms o deg ee o u bani-
za ion, popula ion densi y, en i onmen al exposu es, in as uc u e and
sco es o public spaces and esou ces. Howe e , i anked second
highes wi h espec o a ea dep i a ion in quin iles (4.4 ±0.9), which
was also e lec ed by he highes a e o unemploymen assis ance e-
cipien s pe 1,000 inhabi an s (37.3 ±5.2), high numbe s o gene al
social assis ance bene i s pe 1000 households (90.6 ±34.5) and a low
housing sco e. Mo eo e , DASS-21 le els o ma e nal anxie y (1.7 ±
2.7), dep ession (3.5 ±4.6) and s ess (7.7 ±6.1) and he NOSI-K sum
sco e (20.6 ±7.5) we e subs an ially highe . P e alence o sel - a ed
amily inancial si ua ion (65 % good o e y good) and mo he s’ high
educa ion (85 % uni e si y) we e sligh ly lowe han in clus e 3.
Clus e 6 anked highes in e ms o a e age a ea dep i a ion in
quin iles (4.9 ±0.3). This was also subs an ia ed by he highes numbe
o gene al social assis ance bene i s pe 1000 households (141.6 ±38.4)
and he lowes income pe esiden (10,960 ±1,120
€
). This coincided
wi h he sel - a ed amily inancial si ua ion (47 % o he mo he s a ed
hei si ua ion as good o e y good) and he lowes p opo ion o highly
educa ed mo he s compa ed o o he clus e s (45 % wi h uni e si y
deg ee). Mo eo e , he a e age pe cen age o non-wes e n immig an s
in he neighbo hood anked by a highes among all clus e s (51 ±14
Fig. 2. Clus e s o he UMAP embedding de e mined by agglome a i e nes ing ( he coo dina es do no esemble geog aphical in o ma ion).
K. Telkmann e al.
En i onmen In e na ional 197 (2025) 109335
6
%) which was unde pinned by he highes p opo ions o non-Du ch
e hnici ies amongs he a he s (43 %) and he mo he s (48 %). In
u n, only 64 % el ha hey belong o he Du ch communi y (compa ed
o 88 % o he s udy popula ion). No ewo hy we e also he lowes sco es
in sa e y, housing and popula ion composi ion. Mo eo e , he highes
p opo ion o single mo he s could be ound in his clus e (19 %, which
was almos 11 % abo e he popula ion a e age). Conside ing ma e nal
men al heal h, we obse ed he highes a e age DASS-21 sco es o
anxie y (2.4 ±3.6) and dep ession (3.7 ±4.9) and he second highes
NOSI-K sum sco e (20.7 ±8.3).
Clus e 8 also exhibi ed a highe a ea dep i a ion in quin iles (4.2 ±
1.3) wi h low sco es in esou ces, housing and popula ion composi ion.
This was also e lec ed by he sel - a ed amily inancial si ua ion (63 %
a ed as good o e y good) and compa ably lowe educa ion s a us (63
% o mo he s wi h uni e si y deg ee) in ela ion o he whole s udy
popula ion. Especially no iceable we e he high p opo ion o esiden s
65 yea s and olde (22 %) and he highes numbe o old age assis ance
ecipien s pe 1,000 inhabi an s (129.7 ±41.3). This was also e lec ed
by he highes a e o dea hs pe 1,000 inhabi an s (13.3 ±4.2 as
opposed o 7.1 ±4.4 on a e age). Howe e , his clus e anked second
highes o he social cohesion sco e. S iking we e also he highes
a e age DASS-21 sco e o s ess (7.9 ±6.1), NOSI-K (21.4 ±8.6) and
he second highes p opo ion o child en no li ing wi h bo h pa en s
(27 %). I is also wo h no ing ha child en in clus e 8 we e mo e han
wice as likely o ha e a TV in hei bed oom (33 %) as in he o e all
popula ion (16 %).
The la ge agglome a ion a he op le con aining clus e s 1, 5 and 7
con ained mos ly child en li ing in less densely popula ed a eas,
exposed o less noise and ai pollu ion and gene ally g eene a eas as
indica ed by highe NDVI alues. Mo eo e , he p opo ion o ma ied
couples was conside ably highe in hese clus e s whe eas he numbe o
gene al social assis ance bene i s pe 1,000 households was much lowe .
This was also cap u ed by low o medium a e ages o a ea dep i a ion in
quin iles. Mo eo e , he as majo i y o child en no li ing in he mu-
nicipali y o Ams e dam anymo e was ound in hese clus e s. I is also
no able, ha he DASS-21 sco es o anxie y we e qui e low compa ed o
Fig. 3. Dis ibu ion o selec ed a iables ac oss he embedding. Missing alues a e shown in g ey.
K. Telkmann e al.
En i onmen In e na ional 197 (2025) 109335
7
o he clus e s. Howe e , he e we e s ill subs an ial di e ences be ween
hese clus e s.
Clus e 1 con ained pa icipan s o m mixed deg ees o u baniza ion
wi h 50 % li ing in u ban o e y u ban a eas. Mo eo e , 60 % did no
li e in Ams e dam anymo e. The neighbo hoods ep esen ed in his
clus e we e cha ac e ized by low a e age income pe esiden (12,270
±1,100
€
), highes p opo ion o households wi h child en (44 ±9 %),
low sco es in a ing o esou ces and a high p opo ion o nonwes e n
immig an s (25.9 ±11.9 %). The la e was also e lec ed by he high
p opo ion o pa en s o non-Du ch e hnici y (25 % mo he s, 25 % a-
he s). Pa en al s ess as measu ed by NOSI-K sum sco e was qui e low
(19.4 ±7.6) and he DASS-21 sco e o dep ession was lowes amongs
all clus e s (2.8 ±4.2).
The la ges p opo ion o child en no li ing in Ams e dam was seen
in clus e 5 a 96 %. The deg ee o u baniza ion was simila ly dis ibu ed
as in clus e 1 wi h only 42 % o obse a ions li ing in u ban o e y
u ban a eas. The neighbo hood cha ac e is ics s ood ou h ough he
highes a e age p opo ion o ma ied couples (44.2 ±5 %), highes
sco es o housing, sa e y and social cohesion and he second highes o
popula ion composi ion, lowes noise pollu ion (55.2 ±4.2) and mos
g eenspaces (NDVI mean o 0.5 ±0.1), he lowes numbe o gene al
social assis ance bene i s pe 1,000 households (22 ±22.9) and high
mean income pe esiden (15,050 ±2,670
€
). The la e was backed by
he sel - a ed amily inancial si ua ion wi h 79 % good o e y good
a ings and he high p e alence o highly educa ed mo he s (86 % o he
mo he s wi h uni e si y deg ee).
Clus e 7 ep esen ed he mos u ban communi y wi hin he uppe
la ge agglome a ion wi h 93 % li ing in ei he u ban o e y u ban
a eas. S ill, 41 % o pa icipan s did no li e in Ams e dam. The neigh-
bo hoods we e on a e age mo e dep i ed as in clus e s 1 and 5 (mean
quin ile o 3.4 ±1.2). This was indica ed by a low housing sco e and a
high numbe o old age assis ance ecipien s pe 1,000 inhabi an s (90.8
±23.2). The a e age income pe esiden in he neighbo hood (13,590
±1,600
€
) and he sel - a ed inancial si ua ion we e close o he s udy
popula ion a e age.
The mos no able ea u e o he isola ed clus e 4 was ha 96 ou o
99 pa icipan s li ed in he neighbo hood “Oos elijk Ha engebied”
(li e al ansla ion: “Eas e n Po A ea”). As he name sugges s, his e y
u ban neighbo hood wi h good in as uc u e is close o he wa e and
p o i s om low le els o ai pollu ion and he highes sco e o public
spaces. Mo eo e , he a ea dep i a ion was second lowes (a e age
quin ile 2 ±0.3) amongs all clus e s. This a ea also con ained he
younges popula ion (p opo ion o esiden s 65 yea s and olde was he
lowes a 4.5 ±2.3 %). Al hough he a e age income pe esiden in he
neighbo hood was he second highes ac oss all clus e s (15,600 ±
560
€
), he sel - a ed amily inancial si ua ion did no e lec his wi h
only 61 % good o e y good a ings. While he NOSI- K sco e o pa en al
s ess was qui e low (18.9 ±6.8), ma e nal DASS-21 sco es o anxie y
(1.7 ±3), dep ession (3.6 ±4.9) and s ess (7.7 ±6.4) we e all well
abo e he s udy popula ion a e age.
Fo a mo e de ailed exposome cha ac e iza ion o he clus e s wi h
emphasis on he dimensions o he concep ual amewo k o he Social
Exposome please see Table S2 in he supplemen a y ma e ial.
3.2.2. Spa ial cha ac e iza ion o he clus e s
Since neighbo hood cha ac e is ics a ied subs an ially be ween
clus e s, we analysed he spa ial dis ibu ion o clus e s. We es ic ed
his analysis o he 1,654 pa icipan s li ing in he municipali y o
Ams e dam. Fig. 4a) shows he numbe o s udy pa icipan s pe
neighbo hood. 24 o hese 85 neighbo hoods did no con ain any
pa icipan . Fu he mo e, he e was no neighbo hood in which e e y
pa icipan was assigned o he same clus e . Thus, only he domina ing
clus e (highes pe cen age o clus e membe ship) pe neighbo hood is
shown in Fig. 4b).
The leas dep i ed clus e 3 is dominan in he cen e o Ams e dam
whe eas he mos dep i ed clus e s 6 and 8 a e mo e p ominen in he
ou ski s o he ci y. In e es ing is also clus e 4, which almos
comple ely con ained s udy pa icipan s om he neighbo hood “Oos-
elijk Ha engebied” nea he ci y cen e . Wi h 101 pa icipan s, his
neighbo hood is also he mos sampled in he s udy popula ion.
3.3. Dis ibu ion o heal h- ela ed ou comes ac oss he clus e s
Heal h- ela ed ou comes we e no included in he dimensionali y
educ ion and clus e ing s ep. Thus, hese measu es we e analysed only
a e exposome clus e s ha e been o med.
3.3.1. Men al heal h p oblems assessed wi h S eng hs and Di icul ies
ques ionnai e
The popula ion mean o SDQ sum sco es assessed by he child was a
7.6 (±4.6) and a 6.3 (±4.9) when assessed by he mo he . O e all, 94 %
ell in o he ca ego y “no mal” as p oposed by Goodman (1997) o he
o me and 92 % o he la e . The highes sco es we e obse ed in
clus e 6 (child: 8.4 ±5.3, mo he 7.7 ±5.7), ollowed by clus e s 7
(child: 8.1 ±4.4, mo he : 6.4 ±4.8) and 8 (child: 8 ±4.1, mo he : 6.9
±4.5). Consequen ly, hese clus e s also ea u ed he lowes p opo ion
o child en in he no mal bound. In pa icula , o clus e 6 we obse e
91 % (child) o 86 % (mo he ) in no mal ange. Howe e , o clus e 7
(child: 94 %, mo he : 92 %) and clus e 8 (child: 96 %, mo he : 90 %)
hese numbe s we e s ill close o he popula ion a e age. Lowes SDQ
sum sco e means we e obse ed in clus e s 3 (child: 7 ±4.1 and 97 % in
no mal ange, mo he : 5.2 ±4.1 and 96 % in no mal ange) and 4 (child:
7 ±4.3 and 95 % in no mal ange, mo he : 5.1 ±4.2 and 95 % in no mal
ange). An o e iew on how SDQ mean sco es ac oss clus e s compa ed
o he s udy popula ion mean is gi en in Fig. 5.
Fo he SDQ subscale conduc p oblems, popula ion means o 1.3 ±
1.3 and 94 % in no mal ange o 0.8 ±1.2 and 91 % in no mal ange
we e obse ed o he child and mo he , espec i ely. The highes
a e age was ound in clus e 6 (child: 1.6 ±1.4 and 91 % in no mal
ange, mo he : 1.1 ±1.4 and 83 % in no mal ange). As o he sum
Fig. 4. A) s udy pa icipan s pe neighbo hood in ams e dam b) domina ing
clus e pe neighbo hood as de ined by he highes pe cen age o clus e
membe ship. he 24 g ey a eas (labelled na) did no con ain any pa icipan .
K. Telkmann e al.
En i onmen In e na ional 197 (2025) 109335
8
sco es, clus e 3 pe o med e y well wi h below a e age sco es (child:
1.2 ±1.2 and 96 % in no mal ange, mo he : 0.6 ±1 and 95 % in no mal
ange).
Emo ional symp oms showed a disc epancy be ween he sel -
comple ed and mo he -comple ed e sions. The sco es a e aged ac oss
he popula ion a 1.9 (±1.8) wi h 95 % in no mal ange and 1.7 (±1.9)
wi h 85 % in no mal ange o he child and mo he , espec i ely.
Assessed by he child, well abo e a e age sco es we e seen in clus e s 6
(2.1 ±2), 7 (2.1 ±1.8), and 8 (2 ±1.8). The p opo ion o child en in
no mal ange was s ill close o a e age wi hin hese h ee clus e s wi h
clus e 6 ha ing he smalles sha e a 92 %. The o he clus e s sco ed
a e age o sligh ly below a e age. Rega ding he mo he ’s assessmen ,
clus e 6 s ill sco ed highes a 2.1 (±2.1) wi h only 78 % in no mal
ange, whe eas clus e s 7 and 8 we e a e age o sligh ly abo e a e age.
Sco es well below a e age we e obse ed in clus e s 3 (1.3 ±1.6, 90 %
in no mal ange) and 4 (1.4 ±1.6, 92 % in no mal ange).
The hype ac i i y / ina en ion subscale sco es a e aged a 3.4 (±
2.4) and 82 % in no mal ange (child) and 2.7 (±2.5) and 85 % in
no mal ange (mo he ). Highes clus e means we e obse ed in clus e s
7 (child: 3.7 ±2.4, 79 % no mal, mo he : 2.9 ±2.5, 85 % no mal) and 8
(child: 3.8 ±2.4, 78 % no mal, mo he : 3.1 ±2.6, 80 % no mal). Clus e
6 ma ched he a e age when assessed by he child, bu was second o
wo s by he mo he ’s a ing (3 ±2.5, 82 % no mal). Bes sco es we e
achie ed in clus e 3 assessed by he child (3 ±2.2 and 86 % no mal)
and clus e 4 assessed by he mo he (2.1 ±2.1 and 93 % no mal).
SDQ mean sco es o pee ela ionship p oblems a e aged a 1 ±1.3
and 94 % no mal (child) and 1 ±1.46 and 86 % no mal (mo he ).
Clus e 6 anked wo s o bo h assessmen s (child: 1.4 ±1.4, mo he :
1.5 ±1.7) and also showed conside ably lowe p opo ions o child en
in no mal ange (child: 91 %, mo he : 76 %). On he o he hand, clus e s
3 and 4 showed highe p opo ions o child en in no mal ange a mo e
han 95 % o he sel -comple ed e sion and abo e 91 % o he mo he -
comple ed e sion.
No se e e di e ences could be seen o he SDQ subscale p osocial
beha io a e aging a 8.5 (±1.4) and 96 % no mal as assessed by he
child and 8.6 (±1.6) and 95 % no mal as assessed by he mo he . Nei he
o child’s assessmen no o mo he ’s assessmen alues wi h s ong
de ia ions om he s udy popula ion mean s ood ou .
Di e ences be ween boys and gi ls we e mos p onounced in SDQ
sum sco es assessed by he mo he : boys on a e age sco e 1.4 poin s
highe han gi ls. This gende di e ence was much less se e e by he
child en’s a ings a 0.3. Fo he subscales conduc p oblems (child: 0.3,
mo he : 0.2) and hype ac i i y / ina en ion (child: 0.5, mo he : 1.1)
boys sco ed highe on a e age compa ed o gi ls whe e he opposi e is
ue o emo ional symp oms (child: −0.5, mo he : −0.1) and p osocial
beha io (child: −0.5, mo he : −0.5). No no able gende di e ences
we e obse ed o he subscale pee ela ionship p oblems (child: 0,
mo he : 0.1).
Clus e 4 showed lowe SDQ sum sco es o boys compa ed o gi ls
(gende di e ence assessed by child: −1.8, mo he : −0.6) as opposed o
almos all o he clus e s (excep ions a e clus e 5 and 8 in he child
ques ionnai e a 0.4 and 0.1, espec i ely). On he o he hand, sum
sco es o boys in clus e s 1, 3, and 6 we e subs an ially highe han
hose o gi ls (child: 0.8 o 0.9, mo he : 2 o 2.6). In pa icula , he
pe cen age o gi ls alling in o no mal ange in clus e 6 was 11 % highe
han o boys. In e es ingly, he ac ha boys in clus e 4 sco ed lowe
han gi ls applies o all subscales excep hype ac i i y / ina en ion
assessed by he mo he (he e, boys sco e on a e age 0.3 highe han
gi ls).
3.3.2. Men al heal h p oblems assessed wi h Subs ance use isk P o ile scale
SURPS sco es we e in es iga ed a ou subscales. O e all, he e we e
no subs an ial di e ences wi h clus e means only sligh ly de ia ing
om he popula ion means. The popula ion a e age o he subscale
anxie y sensi i i y was 1.9 (±0.6) and he s onges de ia ion om his
alue was seen in clus e 6 a 2.1 (±0.7). All o he clus e means luc-
ua ed only sligh ly ac oss he sample mean. In case o hopelessness,
clus e means ba ely de ia ed om he popula ion mean o 1.3 (±0.4).
The highes alue was again ound in clus e 6 a 1.3 (±0.4). Simila ly,
o impulsi i y clus e 6 de ia ed a 2 (±0.7) he mos om he popu-
la ion mean o 1.9 (±0.6). No ewo hy was also clus e 3 wi h he lowes
a e age o 1.8 (±0.6). Fo he dimension sensa ion seeking, we obse ed
a popula ion mean o 2.7 (±0.7). The la ges de ia ions could be seen in
clus e 7 a 2.8 (±0.6) and in he o he di ec ion in clus e 4 a 2.7
(±0.7). A hea map depic ing clus e means in compa ison o popula ion
means is shown in Fig. 6.
Fig. 5. Hea map o s anda dized mean di e ences o he s udy popula ion o SDQ clus e mean sco es. No e ha o he subscale “p osocial beha io ” highe alues
a e ega ded as a posi i e ou come. Consequen ly, his subscale is no included in he sumsco e.
K. Telkmann e al.
En i onmen In e na ional 197 (2025) 109335
9
Fo he subscales anxie y sensi i i y and hopelessness gi ls sco ed on
a e age 0.2 and 0.1 poin s highe han boys, espec i ely. Whe eas o
impulsi i y and sensa ion seeking boys sco ed 0.1 and 0.3, espec i ely,
highe han gi ls. No se e e de ia ions om hese di e ences could be
obse ed in he clus e s wi h h ee excep ions: gi ls in clus e 4 sco ed on
a e age 0.5 and 0.2 poin s highe o anxie y sensi i i y and impulsi i y
han boys, espec i ely. Addi ionally, gi ls in clus e 8 sco ed 0.1 poin s
highe on a e age han boys.
Desc ip i e s a is ics o all heal h- ela ed ou comes o each clus e
can be ound in Table S3 in he supplemen a y ma e ial.
4. Discussion
Ou p ima y aim was he applica ion o no el s a is ical me hods o
iden i y exposome clus e s based on a a ie y o social and socie al
cha ac e is ics as well as ac o s o he buil and na u al en i onmen . In
summa y, he applica ion o UMAP o dimensionali y educ ion and
clus e ing esul ed in clea ly dis inc exposome clus e s. Clus e o ma-
ion was mainly d i en by con ex ual socioeconomic and physical
neighbo hood cha ac e is ics, ela ed o he dimensions “socioeconomic
ci cums ances and sociodemog aphic cha ac e is ics” and “places” o
he concep ual amewo k o he Social Exposome (Gudi-Minde mann
e al. 2023). Fo ins ance, clus e s disc imina ed clea ly be ween a eas
exhibi ing di e en deg ees o u baniza ion and, in pa icula , be ween
hose li ing in Ams e dam and hose who mo ed away. A ine
dis inc ion could be made in conjunc ion wi h socioeconomic neigh-
bo hood cha ac e is ics. Thus, s iking di e ences in a ea dep i a ion
we e p esen ac oss clus e s exp essed h ough a e age income pe
esiden , numbe o gene al social assis ance bene i s o a ings o e-
sou ces, housing, public spaces o social cohesion. Mo eo e , clus e
di e ences we e also p esen in e ms o ma e nal men al heal h, which
is conside ed as one c ucial aspec o he dimension “ ela ional dy-
namics” o he domain o social in e ac ions wi hin he Social Exposome
concep .
A subsequen analysis o he exposome clus e s showed conside able
di e ences in child en’s men al heal h. In gene al, he mo e dep i ed
he neighbo hoods and he highe he p e alence o ma e nal men al
heal h p oblems, he highe he p e alence o child en wi h men al
heal h issues, indica ed by alling in o he bo de line and c i ical ange
o he SDQ. This is in line wi h p e ious s udies on isk ac o s o
child en’s men al heal h p oblems (Ra ens-Siebe e e al. 2007; Becke
e al. 2015). The SDQ is a widely applied sc eening ins umen in
epidemiological s udies and has been shown o be alid o de ec ion o
emo ional and beha io al p oblems in child en/adolescen s, especially
when aking sel - epo and pa en al epo in o accoun (Wa nick e al.
2008; Theunissen e al. 2019). Howe e , no ema kable p e alence
di e ences in pe sonali y ai s assessed wi h he SURPS we e obse ed
be ween he exposome clus e s. This migh be due o he ela i ely
young age o he s udy pa icipan s. SURPS pe sonali y p o iles ha e
been shown o be associa ed wi h subs ance use beha io in a s udy o
Du ch adolescen s aged 11–15 (Malmbe g e al. 2010).
4.1. Rela ion o o he s a is ical app oaches in exposome esea ch
Wi h ou analysis s a egy we iden i ied exposome clus e s based on
simila i y o exposu es and hen desc ip i ely analysed dis ibu ions o
men al heal h- ela ed ou comes ac oss hose clus e s. In con as o
me hods di ec ly linking he exposome o a heal h- ela ed ou come and
hen isola ing a smalle se o isk ac o s while adjus ing o con-
ounde s, his me hod akes in o accoun he en i e y o exposu es. To
ou knowledge, his was he i s ime in epidemiological esea ch ha
his inno a i e app oach was used o iden i y clus e s based on expo-
su es o he socie al, social, buil and na u al en i onmen . In pa icula ,
he applica ion o UMAP as a dimension educ ion echnique is a no el y
o his kind o esea ch. So a , UMAP has been mos ly applied o la ge
scale da a such as single-cell da a (Bech e al. 2019) o gene and p o ein
da a (Do i y e al. 2020; Lunjani e al. 2024), o o de ec u ban
con igu a ion ypes (Iungman e al. 2024). Up o now, he e is, o ou
knowledge, only one epidemiological applica ion by Bej e al. (2022)
u ilizing UMAP o de ec clus e s based on sociodemog aphic and so-
cioeconomic cha ac e is ics and li es yle habi s o desc ibe he e oge-
neous Type-2 diabe es subpopula ions. Howe e , in ou s udy we
applied a uni ied me ic o mixed da a, while Bej e al. (2022) employed
UMAP on h ee se s o a iables depending on hei ype (nume ic,
ca ego ical, o dinal) and only hen combined he esul s.
A e iew o s a is ical me hods by San os e al. (2020) p o ided an
o e iew o issues o add ess and easible algo i hms wi hin he speci ic
con ex o bi h coho s udies. Simila ly, S a oggia e al. (2017) p e-
sen ed a se ies o applicable me hods o analyse mul iple en i onmen al
exposu es. Ou app oach is in line wi h he ecommended s eps o co -
ela ion analysis, dimension educ ion, and clus e ing. Fo ins ance,
Robinson e al. (2015) and Tamayo-U ia e al. (2019) pu sued an
app oach based on in es iga ion o co ela ions be ween se e al amilies
o exposu es and hen applied p incipal componen analysis. Since ou
da a con ained a iables o mixed ype we op ed o a di e en solu ion
o dimension educ ion.
Since i s in oduc ion, exposome esea ch has seen a a ie y o
di e en app oaches o unco e associa ions be ween exposu es and
heal h ou comes. Many au ho s conside me hods linking a la ge se o
exposu es di ec ly o a speci ic p e-de ined heal h ou come. In analogy
o genome-wide associa ion s udies, exposome-wide associa ion s udies
(ExWAS) ha e been applied o link he mos ele an pa s o he
exposome o se e al heal h ou comes such as BMI in adul s (Haddad
e al. 2022), bi h weigh (Nieuwenhuijsen e al. 2019), hype ensi e
diso de s (Hu e al. 2020) and COVID-19 mo ali y (Hu e al. 2021),
among o he s. This me hod has also been inco po a ed in o clus e ap-
p oaches. Guillien e al. (2021, 2022) applied ExWAS o educe he se o
exposu es and subsequen ly de ine exposome p o iles and in es iga e
Fig. 6. Hea map o s anda dized mean di e ences o he s udy popula ion o SURPS clus e mean sco es.
K. Telkmann e al.
