The u ban physical en i onmen and leisu e- ime physical ac i i y in ea ly
midli e: a FinnTwin12 s udy
Zhiyang Wang
a
, Sa i Aal onen
a
, Roos Teeuwen
b
, Vasileios Milias
b
, Ca men Peu e s
c,d,e
,
B uno Raimbaul
c,d,e
, Teemu Pal iainen
a
, E in Lumpe
, Danielle Dick
g
, Jessica E. Sal a o e
g
,
Ma ia Fo as e
h
, Payam Dad and
c,d,e
, Jo di Júl ez
c,i
, Achilleas Psyllidis
b
, I ene an Kamp
j
,
Jaakko Kap io
a,*
a
Ins i u e o Molecula Medicine Finland, Helsinki Ins i u e o Li e Science, Uni e si y o Helsinki, Helsinki, Finland
b
Depa men o Sus ainable Design Enginee ing, Del Uni e si y o Technology, Del , he Ne he lands
c
ISGlobal, Ba celona Biomedical Resea ch Pa k (PRBB), Ba celona, Spain
d
Uni e si a Pompeu Fab a (UPF), Ba celona, Spain
e
Spanish Conso ium o Resea ch on Epidemiology and Public Heal h (CIBERESP), Ins i u o de Salud Ca los III, Mad id, Spain
Depa men o Psychology, Ru ge s Uni e si y, New B unswick, NJ, USA
g
Depa men o Psychia y, Ru ge s Uni e si y, Pisca away, NJ, USA
h
PHAGEX Resea ch G oup, Blanque na School o Heal h Science, Uni e si a Ramon Llull (URL), Ba celona, Spain
i
Clinical and Epidemiological Neu oscience (Neu o`
Epia), Ins i u d’In es igaci´
o Sani `
a ia Pe e Vi gili (IISPV), Reus, Spain
j
Na ional Ins i u e o Public Heal h and he En i onmen , he Ne he lands
ARTICLE INFO
Keywo ds:
Exe cise
Machine lea ning
U baniza ion
Beha io
ABSTRACT
Unde he exposome amewo k, his s udy examined he ela ionship be ween he u ban physical en i onmen
and leisu e- ime physical ac i i y du ing ea ly midli e based on 394 pa icipan s (mean age: 37, ange 34–40)
om he FinnTwin12 coho , esiding in i e majo Finnish ci ies in 2020. We cu a ed 145 u ban physical ex-
posu es based on esiden ial add esses and measu ed h ee ou comes: o al leisu e- ime physical ac i i y ( o al
LTPA) and wo sub-domains: leisu e- ime physical ac i i y wi hou commu ing ac i i y (LTPA) and commu ing
ac i i y. K-p o o ypes clus e ing iden i ied h ee u ban clus e s: “o iginal ci y cen e ,” “new ci y cen e ,” and
“subu ban,” each wi h dis inc en i onmen al pa e ns. Reg ession models showed ha pa icipan s in he
“subu ban” clus e had lowe le els o o al LTPA and LTPA compa ed o hose in he “o iginal ci y cen e ”
clus e , while we ound null indings o commu ing ac i i y. Then, epea ed eg ession models wi h a p- alue
h eshold o 0.01 we e used o ini ially selec candida es. eX eme G adien Boos ing models iden i ied g een-
spaces and oad cha ac e is ics as he op impo an ac o s in luencing o al LTPA, while pocke pa k and
g eenness we e anked as he op impo an ac o s in luencing LTPA. The ela ionships we e non-linea . The e
we e h esholds o he coun and size o pocke pa ks wi hin 800 m walking dis ance and he modi ied soil
adjus ed ege a ion index, de e mining whe he hey posi i ely o nega i ely p edic LTPA. Ou indings sug-
ges ed ha he u ban en i onmen in Finnish ci ies was associa ed wi h leisu e- ime physical ac i i y, which
e ealed new esiden ial pa e n and iden i ied key exposu es o oad, pocke pa k, and g eenness wi h non-
linea e ec , ha can guide u u e policies.
1. In oduc ion
Regula physical ac i i y has been widely demons a ed o p e en
mul iple non-communicable diseases and educe he isk o p ema u e
dea h (Boo h e al., 2012). The economic and heal h bu den a ising om
physical inac i i y is subs an ial and con inually ising, cos ing public
heal h ca e sys ems an es ima ed USD 47.6 billion globally e e y yea
(San os e al., 2023). Since p e ious s udies show a s ong con ibu ion
o en i onmen al ac o s o physical ac i i y (Boomsma e al., 2011;
Ca lin e al., 2017; Duncan e al., 2008), in e en ions a ge ing he
en i onmen may be a good en y poin o p omo e physical ac i i y.
Ea ly midli e is a c ucial bu ela i ely o e looked s age o in es iga ing
* Co esponding au ho . Ins i u e o Molecula Medicine, Uni e si y o Helsinki, PL 20 (Tukholmanka u 8), FI-00014, Helsinki, Finland.
E-mail add ess: [email p o ec ed] (J. Kap io).
Con en s lis s a ailable a ScienceDi ec
Heal h and Place
jou nal homepage: www.else ie .com/loca e/heal hplace
h ps://doi.o g/10.1016/j.heal hplace.2025.103495
Recei ed 26 July 2024; Recei ed in e ised o m 22 Ap il 2025; Accep ed 23 May 2025
Heal h & Place 94 (2025) 103495
A ailable online 4 June 2025
1353-8292/© 2025 The Au ho s. Published by Else ie L d. This is an open access a icle unde he CC BY license (
h p://c ea i ecommons.o g/licenses/by/4.0/ ).
physical ac i i y, as i e lec s es ablished long- e m heal h beha io
pa e ns and ma ks he pe iod be ween key ansi ions in young adul -
hood and la e li e. Unheal hy beha io s du ing his s age can inc ease
he isk o non-communicable diseases (Dagenais e al., 2021), and a
ecen s udy showed ha ea ly midli e physical ac i i y s ongly p edic s
u u e beha io (Be n zen e al., 2024).
U baniza ion s ands as a ans o ma i e end, wi h mo e han hal
he wo ld’s popula ion cu en ly esiding in u ban a eas (Uni ed Na ions
Human Se lemen s P og amme, 2022). Many e iews ha e summa ized
he salien link be ween he u ban en i onmen and physical ac i i y
(Ding e al., 2011; Du and e al., 2011; K¨
a meniemi e al., 2018). The
exposome o e s a heo e ical amewo k wi h an umb ella pe spec i e
o depic he o ali y o he en i onmen ha people expe ience (Wild,
2012) and examine heal h e ec s om he eal-wo ld u ban en i on-
men , o which he u ban physical componen plays an impo an ole.
The exposome s udies ha e he po en ial o un eil mo e comp ehensi e
non-gene ic p edic o s h ough la ge-scale cha ac e iza ion o he
en i onmen . Go man e al. (2021) ha e ou lined he bidi ec ional e ec
be ween he exposome and physical ac i i y bu poin ed ou he un-
ce ain y in mechanisms and in e ac ions. The u ban physical exposome
is ubiqui ous and mul i ace ed, which makes i a complex en i y o
s udy.
E e y en i onmen al ac o con ibu es o his complex o ali y o
exposu es, and no ac o is isola ed. U ban egene a ion p ojec s a e a
good example, usually designed o imp o e public heal h by imple-
men ing s uc u al and isk-minimizing solu ions. They o en yield
colla e al e ec s on o he aspec s, such as b inging economic, social, and
cul u al bene i s, wi hin he ci y’s complex sys em (Chen e al., 2023;
Sonnenschein e al., 2022; Wang e al., 2021). Fo example, an u ban
i e side pa k egene a ion p ojec in Ba celona, Spain was es ima ed o
a ac o e i e housand adul use s daily o pe o m di e en ypes o
physical ac i i y (Ve e al., 2019). Beyond he p ojec ’s basic objec-
i es, an open-ai museum will be buil he e, ans o ming social and
buil en i onmen s. Nowadays, egene a ion p ojec s a ound he wo ld
a e o en mul i-componen and in e sec o al. In ano he Ba celona
egene a ion p og am, aiming o imp o e li ing condi ions in he mos
disad an aged neighbo hoods (in ol ing, o example, social se ices,
g eenspaces, and household suppo ), esea che s ound ha he
neighbo hood wi h a bigge p ojec budge was associa ed wi h a highe
equency o physical ac i i y among esiden s (Ba oll-Roca e al.,
2024). P e ious s udies elying on single exposu es o limi ed se s we e
ela i ely inadequa e o depic he b oade u ban en i onmen and i s
heal h e ec s.
Se e al app oaches ha e been de eloped o add ess he challenges o
high-dimensional exposome da a. Fea u e ex ac ion in ol es ans-
o ming nume ous exposu es in o in o ma i e and compac indices. One
such me hod, clus e ing analysis, p o ides insigh s in o eal-wo ld sce-
na ios and o e s high scalabili y o unco e hidden pa e ns ins ead p e-
speci ied classi ica ion. Ano he app oach in ol es machine lea ning
models, widely used in exposome esea ch, which su pass con en ional
analyses by cap u ing non-linea ela ionships, disen angling in-
e ac ions, and p o iding obus compu a ion o mul i-in e ence
(Ohanyan e al., 2022). This app oach may deepen he unde s anding
o he dis inc and complex cha ac e is ics o he u ban physical expo-
some. Rela ed o he u ban physical exposome and physical ac i i y,
p e ious s udies ha e applied such an app oach, howe e , he numbe o
exposu es included in hose s udies (Lee e al., 2023; Ping e al., 2023;
Zang e al., 2022) we e small (6–28), es ic ing hei abili y o cap u e
he complexi y o he u ban exposome. In addi ion, esea ch linking
de ailed exposu e p o iling and leisu e- ime physical is also limi ed in
Finland and No dic coun ies.
In his s udy, we adop ed a holis ic pe spec i e o examine he as-
socia ion be ween he u ban physical componen o he exposome and
leisu e- ime physical ac i i y du ing ea ly midli e, mo ing beyond he
con en ional “one exposu e” epidemiological model. The e we e wo
key objec i es. The i s was o de ec people wi h he e ogeneous u ban
li ing en i onmen s using clus e analysis and o compa e leisu e- ime
physical ac i i y le els be ween clus e s. The second objec i e was o
ank u ban physical exposu es by impo ance wi h espec o leisu e-
ime physical ac i i y, o examine non-linea ela ionships, and o
de ec pai wise in e ac ions be ween exposu es by he g adien boos ing
me hod. The aim was o p o ide ac ionable e idence o local policy-
make s and o he s akeholde s abou he complex e ec o he u ban
en i onmen on leisu e- ime physical ac i i y. This would acili a e he
de elopmen o p ecise and cos -e ec i e in e en ions o add ess he
physical inac i i y challenge.
2. Ma e ial and me hods
The low cha o his s udy is p esen ed in Fig. 1.
2.1. Pa icipan s
Pa icipan s we e om he FinnTwin12 coho , a na ionwide p o-
spec i e coho o all Finnish wins bo n be ween 1983 and 1987,
iden i ied and collec ed om he na ion’s Cen al Popula ion Regis y.
B ie ly, a baseline (1994–1999), 5522 12-yea -old wins we e in i ed o
pa icipa e and 87 % o hem ag eed o ake pa . The e we e ou
ollow-ups: age 14, age 17, young adul hood (mean age 24), and ea ly
midli e (mean age 37), wi h e en ion a es o 92 %, 75 %, 66 %, and 41
%, espec i ely. A ecen s udy has de ailed he la es ollow-up o he
coho (Cooke e al., 2025). In his s udy, we included indi idual wins
who li ed in i e la ge ci ies o Finland, namely Helsinki, Tampe e,
Espoo, Oulu, o Jy ¨
askyl¨
a, in 2020, esul ing in o a smalle sample in he
p esen analysis. Acco ding o he S a is ics Finland, 11.9 %, 4.4 %, 5.3
%, 3.7 %, and 2.6 % o he o al popula ion esided in Helsinki, Tampe e,
Espoo, Oulu, and Jy ¨
askyl¨
a in 2020, espec i ely.
2.2. Measu es
2.2.1. Leisu e- ime physical ac i i y
Ou s udy ocuses on ea ly midli e leisu e- ime physical ac i i y,
which is pe o med a he pe son’s disc e ion, along wi h essen ial daily
li ing ac i i ies o wo k- ela ed asks (Caspe sen e al., 1985). This ype
o physical ac i i y is conside ed one o he mos e ec i e ways o in-
c ease o e all physical ac i i y le els (Bo odulin e al., 2008). I was
measu ed h ough s uc u ed and alida ed ques ions on he equency,
mean du a ion, and mean in ensi y o pa icipan s’ leisu e- ime physical
ac i i y sessions, as well as a ques ion on hei commu ing ac i i y
(Leskinen e al., 2009; Walle e al., 2008). Based on hese s uc u ed
ques ions, we quan i ied mean me abolic equi alen o ask (MET) hou s
pe day, which exp essed he ene gy cos o physical ac i i ies in he
o m o he es ing me abolic a e (Je ´
e e al., 1990), as he uni o
physical ac i i y le el. The calcula ion o mula o he mean MET hou s
pe day was he ollowing: physical ac i i y equency (a e age pe day)
×physical ac i i y du a ion (a e age hou s) ×physical ac i i y in-
ensi y (physical ac i i y MET alue) (Kujala e al., 1998). The MET
alues o ac i i y in ensi y we e: 4 o in ensi y co esponding o
walking, 6 o igo ous walking o jogging, 10 o jogging, and 13 o
unning. All ypes o leisu e- ime physical ac i i ies we e conside ed
when MET hou s pe day we e calcula ed. We assumed ha commu ing
ac i i y was done on 5 days pe week and on he in ensi y o walking.
The ques ions a e lis ed in Supplemen al No e 1 and he compa ison o
MET speci ica ion be ween he Compendium o Physical Ac i i y and
his s udy is p esen ed in Supplemen a y Table 1.
The p ima y measu e, o al leisu e- ime physical ac i i y ( o al LTPA),
was he sum o wo sub-domains (seconda y measu es): 1) leisu e- ime
physical ac i i y wi hou commu ing ac i i y (LTPA) and 2) commu ing ac-
i i y. All hese measu es we e exp essed as he mean MET hou s pe day.
The highe le el o MET hou s pe day means he highe he le el o
ac i i y. Pa icipan s wi h o e mean 45 MET hou s/day o o al LTPA
we e iden i ied as ou lie s and emo ed. This h eshold co esponds o,
Z. Wang e al.
Heal h and Place 94 (2025) 103495
2
o example, app oxima ely 3.5 h o as unning daily, which is likely
un ealis ic (Ainswo h e al., 2011; He mann e al., 2024). The dis i-
bu ions o all h ee measu es a e shown in Supplemen al Fig. 1, and due
o he skewness, we log- ans o med hem.
2.2.2. U ban physical exposome
We assigned 145 indica o s o u ban physical exposu es o he esi-
den ial add ess o each s udy pa icipan . De ailed desc ip ion and
summa y s a is ics o hese indica o s a e p esen ed in Supplemen al
Table 2. The u ban physical exposome se comp ehensi ely depic ed he
u ban en i onmen including aspec s such as a ic, s ee s, land use,
g een (i.e. pa ks, o es s, and ields) and blue (i.e. lakes and seas) spaces,
and so on. The compu ing and en iching p ocess was on he geocode
le el and de i ed om mul iple open sou ces, desc ibed in Supple-
men al No e 2 and elsewhe e (Milias and Psyllidis, 2021; Teeuwen e al.,
2023; an Kamp e al., 2022; Wang e al., 2023). Mos u ban physical
exposu es we e measu ed o modelled in 2018 and 2023, and he pe -
cen age o a ea co e ed by ees was measu ed in 2015. We used he
esiden ial his o y p o ided by he Digi al and Popula ion Da a Se ices
Agency, Finland be ween bi h and 2020 o me ge he u ban physical
exposu es by EUREF-FIN geocodes. Exposu es a ailable in 2018 o 2015
we e me ged wi h esiden ial add esses o pa icipan s in 2018 o 2015,
while exposu es a ailable in 2023 we e me ged wi h esiden ial ad-
d esses in 2020. The e o e, he exposu es should be ega ded as he
exposu es in 2015, 2018, o 2020 in he e m o empo ali y.
2.2.3. O he measu es
Fi e sociodemog aphic a iables we e iden i ied a p io i: sex (ca e-
go ical, emale s. male), age (con inuous, yea ), wo k (ca ego ical, no
wo king o o he si ua ion s. cu en ly wo k), educa ion (ca ego ical,
pos -seconda y o lowe s. bachelo /equi alen o abo e), and li ing
s a us (ca ego ical, wi h a spouse o a pa ne s. wi h a spouse o a
pa ne and child( en), s. o he si ua ion). The la e h ee we e sel -
epo ed a he ea ly midli e ollow-up. Sex was based on he egis e
in o ma ion ob ained when he coho was es ablished, while age was
compu ed om he di e ence be ween he da e o esponse and he da e
o bi h. The e we e ano he h ee beha io al a iables: illici subs ance
use (ca ego ical, ne e s. a leas once), e e smoke (smoked o e 100
Fig. 1. S udy low.
Z. Wang e al.
Heal h and Place 94 (2025) 103495
3
ciga e es li e ime) (ca ego ical, no s. yes), and alcohol d inking (ca -
ego ical, mon hly o less o e en ne e s. 2–4 imes a mon h o mo e),
inqui ed also a he ea ly midli e ollow-up. Adul leisu e- ime physical
ac i i y was associa ed o mos o he sociodemog aphic and beha io al
a iables, as shown in p e ious esea ch (Abu-Oma e al., 2021; Poo -
inga, 2007; Samdahl and Jekubo ich, 1997; Thompson e al., 2020).
To depic he social en i onmen , ou neighbo hood social a iables
a he pos al code le el we e de i ed om S a is ics Finland in 2018: he
p opo ion o esiden li ing alone (single household), o esiden s wi h
he lowes educa ion le el, o esiden s wi h he lowes income qua ile,
and o unemployed esiden s. A neighbo hood dep i a ion sco e was
gene a ed om he la e h ee social a iables (Ki im¨
aki e al., 2020).
We i s s anda dized he h ee a iables o z-sco es, and hei mean
alue was he dep i a ion sco e. Using a median spli , we hen ca ego-
ized neighbo hoods whe e pa icipan s li ed in 2018 in o wo le els:
low- and high-dep i ed. Thus, he wo neighbo hood social a iables:
he p opo ion o esiden li ing alone and dep i a ion le el we e
me ged ia esiden ial his o y in 2018 oo.
2.3. Analysis
2.3.1. Da a p ocessing
A e excluding hose people who did no ha e in o ma ion on
leisu e- ime physical ac i i y, sociodemog aphic, beha io al, and
neighbo hood-le el social a iables, 394 win indi iduals esiden in
hese u ban a eas we e included in his s udy. Gi en ha he e we e only
44 win pai s wi h bo h co wins sa is ying he inclusion c i e ia, we did
no conside zygosi y as a co a ia e and did no pe o m any pai wise
win analysis. The dis ibu ion o sociodemog aphic and beha io al
a iables among included and excluded pa icipan s a e p esen ed in
Supplemen al Table 3. The e we e signi ican di e ences be ween
included and excluded pa icipan s in educa ion, illici subs ance use,
and alcohol d inking.
2.3.2. Clus e ing analysis
The k-p o o ypes clus e analysis was employed o dis inguish
dis inc pa e ns in he u ban en i onmen . I combines dissimila i y
measu es om bo h k-means and -modes algo i hms o mixed ypes o
exposu es, and has shown o ha e a good pe o mance (Huang, 1998;
P eud’homme e al., 2021). Con inuous exposu es we e s anda dized by
s anda d de ia ion (SD). All 145 u ban physical exposu es we e
included in he clus e ing algo i hms. The Silhoue e me hod was used
o p e-speci y he numbe o clus e s (Al- Zoubi and Rawi, 2008).
One-s ep impu a ion wi hin he algo i hm was applied o missing
alues (Aschenb uck e al., 2023). Since k-p o o ypes clus e analysis is
sensi i e o ou lie s, he p incipal componen analysis (PCA) o mixed
da a was conduc ed be o e. Th ee pa icipan s whose i s o second
p incipal componen s (PCs) ell ou side he ange o i e s anda d de-
ia ions we e iden i ied as ou lie s (Jolli e, 2002) and excluded om
he clus e analysis.
Nex , hie a chical linea eg ession was pe o med o he ela ion-
ship be ween he u ban clus e and leisu e- ime physical ac i i y mea-
su es wi h h ee adjus men plans o co a ia es: 1) sociodemog aphic
a iables, 2) sociodemog aphic and beha io al a iables, and 3) socio-
demog aphic, beha io al, and neighbo hood social a iables. The clus-
e e ec o sampling based on amilies o win pai s was con olled by
he obus s anda d e o .
2.3.3. Machine lea ning analysis
Be o e explo ing he complexi y wi hin he u ban en i onmen ia a
plu alis ic analysis pla o m, gene alized linea eg ession models wi h
he obus s anda d e o we e epea edly pe o med be ween each
leisu e- ime physical ac i i y measu e ( o al LTPA, LTPA, and
commu ing ac i i y) and each u ban physical exposu e (missing alues
we e impu ed). The exposu es we e no placed in a single eg ession
model. The a p io i signi ican h eshold o 0.01 was used o selec
no ewo hy candida es and acknowledge he co ela ion be ween ex-
posu es. Dimensional educ ion inc eases he model s abili y o subse-
quen analysis.
Then, we pe o med he eX eme G adien Boos ing (XGBoos )
model o assess he impo ance o u ban physical exposu es on each
leisu e- ime physical ac i i y measu e, unco e in e ac ions, and iden-
i y nonlinea ela ionships (Chen and Gues in, 2016). I is an op imized
dis ibu ed g adien boos ing lib a y designed o e icien and scalable
aining o machine lea ning models, wi h g adien -boos ed decision
ees algo i hm (Chen and Gues in, 2016). The hype pa ame e s we e
uned h ough he 5- old c oss- alida ion g id sea ch (Yu and Zhu,
2020). The pa icipan s we e andomly spli in o aining and es ing
subse s in a a io o 3:1. The model pe o mance was e alua ed by
oo -mean-squa e e o (RMSE). Selec ed u ban physical exposu es,
sociodemog aphic, beha io al, and neighbo hood social a iables we e
included in he model. A e hype pa ame e uning, he model was
epea ed wo addi ional imes wi h di e en seeds o esul obus ness.
Due o he lack o coun e ac ual design, he cu en cons uc ion o
XGBoos models did no conside any causal in e ence.
To inc ease model anspa ency, he SHapley Addi i e exPlana ions
(SHAP) alue was used o in e p e and isualize he esul s om he
XGboos model, which ea u ed he exposu es’ impo ance on he
ou come based on he coope a i e game heo y (Lundbe g and Lee,
2017). I s di ec ion sugges s he e ec ’s di ec ion on p edic ion, leading
he model o p edic ei he a highe o lowe alue o ou comes. I s
magni ude is a measu e o how s ong he e ec is. We quan i ied
pai wise in e ac ion SHAP alues be ween included a iables and
summed hei absolu e alue o all pa icipan s, wi h a high alue
indica ing a s ong in e ac ion and syne gis ic e ec (Lundbe g e al.,
2020). Addi ionally, G oup-Lasso INTERac ion-NET was pe o med o
in e ac ion o compa e wi h he XGBoos ’s esul (Lim and Has ie,
2015).
2.3.4. Sensi i i y analysis
Due o missing alues in u ban physical exposu es, we addi ionally
pe o med sensi i i y K-p o o ype clus e analysis and epea ed gene -
alized linea eg ession models be ween each u ban physical exposu e
and each ou come, a e emo ing pa icipan s wi h missing alues (n =
13).
Then, we used PCA o mixed da a again, as ea u e ex ac ion, o
ans o m he high-dimensional exposome da a (all 145 exposu es) in o
in o ma i e, compac indices. We ex ac ed he op n PCs which
explained o e 80 % o he o al a iance. We hen p e o med XGBoos
models including selec ed PCs, sociodemog aphic, beha io al, and
neighbo hood social a iables o all h ee ou comes.
3. Resul s
3.1. Desc ip ion o pa icipan s
O he 394 included pa icipan s (mean age: 37, SD: 1.5) (Table 1),
mo e indi iduals we e emale (55 %). Al oge he , 87 %, 79 %, and 38 %
o pa icipan s we e employed, had a leas bachelo -le el educa ion,
and li ed wi h a spouse/pa ne and child( en), espec i ely. In hei
ea ly midli e, mo e han hal o he pa icipan s d ank alcohol a leas
2–4 imes a mon h (58 %), bu ewe we e e e smoke (45 %) o had
used illici subs ances such as ma ijuana a leas once (48 %). Be o e log-
ans o ma ion, he means o o al LTPA, LTPA, and commu ing ac i i y
(uni : MET hou s/day) we e 5.4 (SD: 4.7), 4.3 (SD: 4.4), and 1.1 (SD:
1.0), espec i ely. A e log- ans o ma ion, Spea man co ela ions be-
ween o al LTPA and LTPA, be ween o al LTPA and commu ing ac-
i i y, and be ween LTPA and commu ing ac i i y we e 0.9, 0.3, and 0.1,
espec i ely.
Z. Wang e al.
Heal h and Place 94 (2025) 103495
4
3.2. Resul s om clus e ing and hie a chical eg ession
The Silhoue e me hod iden i ied he op imal numbe o clus e s o
be h ee (la ges Silhoue e index, o al wi hin-clus e sum o squa es:
42323.69). Using he map o Helsinki and Espoo and he spa ial laye o
cen e s and shopping a eas in 2019 om he communi y s uc u e
moni o ing sys em, Finnish En i onmen Ins i u e (Viinikka e al.,
2023), we classi ied Clus e 1, 2, and 3 as he “O iginal ci y cen e ”,
“New ci y cen e ”, and “Subu ban” clus e s, espec i ely, based on he
pa icipan s’ esidence in 2018, as he u ban clus e a iable (Fig. 2).
The naming p ocess was solely based on isual obse a ion.
A e ully adjus ing o sociodemog aphic, beha io al, and neigh-
bo hood social a iables, compa ed o pa icipan s who li ed in he
“o iginal ci y cen e ” clus e , pa icipan s who li ed in he “subu ban”
clus e we e associa ed wi h signi ican ly lowe log- ans o med sco es
o o al LTPA (be a: −0.12, 95 % CI: −0.22, −0.03) and LTPA (be a:
−0.16, 95 % CI: −0.28, −0.05) (Table 2). The e ec sizes did no change
subs an ially a e adjus men o sociodemog aphic a iables only and
adjus men o bo h sociodemog aphic and beha io al a iables.
Rega dless o adjus men plans, he e was no signi ican associa ion
be ween he u ban clus e and commu ing ac i i y (Table 2). The e was
no signi ican di e ence in any ou come be ween pa icipan s who li ed
in he “subu ban” and “new ci y cen e ” clus e s. The powe s o ull-
adjus ed models o o al LTPA, LTPA, and commu ing ac i i y we e all
1.0.
3.3. Resul s om XGBoos
Based on he epea ed gene alized linea eg ession, he e we e 25
u ban physical exposu es signi ican ly associa ed wi h o al LTPA and
24 wi h LTPA (Supplemen al Table 4). No u ban physical exposu e me
he h eshold p- alue o 0.01 o associa ion wi h commu ing ac i i y
(Supplemen al Table 4), so he e was no XGBoos analysis o i .
In he XGBoos model o o al LTPA including selec ed u ban phys-
ical exposu es, sociodemog aphic, beha io al, and neighbo hood social
a iables, he op h ee impo an u ban physical exposu es we e he
coun o any ype o oad junc ions wi hin a 500 m bu e (in s_500), he
dis ance o he closes oad ( ega dless o oad ype) (dis _any oad), and
he 5-yea s mo ing a e age o No malized Di e ence Vege a ion Index
(NDVI), an indica o o gene al g eenness, wi hin a 500 m bu e a ound
he home du ing whole yea (nd i_5y s_all_500) (Fig. 3A). In depen-
dence plo s, a U-shaped ela ionship was obse ed be ween he SHAP
alue and he coun o any ype o oad junc ions wi hin a 500 m bu e
(Fig. 3B). When he coun was be ween 15 and 50, i p edic ed a lowe
log- ans o med o al LTPA (nega i e SHAP alues). I i was lowe o
highe han he ange, i p edic ed a highe log- ans o med o al LTPA
(posi i e SHAP alues). Likewise, he e was a non-linea ela ionship
be ween he dis ance o he closes oad and he SHAP alue, wi h bo h
W-shaped pa e ns (Fig. 3C). In Fig. 3D, when he 5-yea s mo ing
a e age o NDVI wi hin a 500 m bu e du ing whole yea was below 0.2,
he SHAP alue was app oxima ely 0.005. When i was be ween 0.2 and
0.4, SHAP alues anged om 0.001 o 0.002. When i was o e 0.4,
SHAP alues became nega i e, bu when i was u he o e 0.6, SHAP
alues u ned posi i e again.
In he XGBoos model o LTPA (Fig. 3E), he op h ee mos impo an
u ban physical exposu es we e he coun o pocke pa ks wi hin an 800
m walking dis ance (coun _pocke pa ks_800), he 5-yea s mo ing
a e age o Modi ied Soil Adjus ed Vege a ion Index (MSAVI), ano he
indica o o gene al g eenness, wi hin a 500 m bu e a ound he home
du ing whole yea (msa i_5y s_all_500), and he o al a ea o all in e -
connec ed pocke pa ks wi hin an 800 m walking dis ance (suma -
ea_pocke pa ks_800). The e was a shi om nega i e o posi i e
p edic ions o log- ans o med LTPA, when he coun o pocke pa ks
wi hin an 800 m walking dis ance exceeded wo (Fig. 3F). Fo he 5-
yea s mo ing a e age o MSAVI wi hin a 500 m bu e a ound he
home du ing whole yea , SHAP alues we e posi i e below 0.018 and
nega i e abo e 0.18 (Fig. 3G). Gene ally, when he o al a ea o all
in e connec ed pocke pa ks wi hin an 800 m walking dis ance was
below 0.01 km
2
, SHAP alues we e nega i e, and when i was o e 0.01
km
2
, SHAP alues we e posi i e. Fig. 3G u he sugges ed ha he
exposu e looked like a ou -g oup ca ego ical a iable.
Supplemen al Fig. 2 displays pai wise SHAP in e ac ion alues in he
XGBoos model o o al LTPA, and he e was some pai wise in e ac ion
be ween u ban physical exposu es bu wi h low in e ac ion SHAP alue.
Simila ly, he XGBoos model o LTPA (Supplemen al Fig. 3) indica es
less in e ac ions also wi h e y low alues. G oup-Lasso INTERac ion-
NET models also did no cap u e any s ong pai wise in e ac ion o
nei he physical ac i i y measu e analyses.
The lea ning cu es in aining and es ing subse s o models o o al
LTPA and LTPA a e p esen ed in Supplemen al Fig. 4. Fo he XGBoos
model o o al LTPA, he RMSE is 0.23 in he aining subse and 0.29 in
he es ing subse . In he wo ex a es s, he aining RMSEs a e bo h
0.24 and he es ing RMSEs a e 0.27 and 0.29. Compa ing he epo ed
es wi h wo ex a es s, he impo ance ank a ied, bu he coun o
any ype o oad junc ions wi hin a 500 m bu e was always he mos o
hi d mos impo an (Supplemen al Table 5). Fo he XGBoos model o
LTPA, he RMSE is 0.32 in bo h aining and es ing subse s. In he wo
ex a es s, he aining RMSEs a e bo h 0.31 and he es ing RMSEs a e
0.33 and 0.34. The impo ance ank also a ied be ween epo ed esul s
and wo ex a es s, bu he mos impo an u ban physical exposu e, he
coun o pocke pa ks wi hin an 800 m walking dis ance, was in he op
wo in one ex a es (Supplemen al Table 6).
3.4. Sensi i i y analysis
A e excluding 13 pa icipan s wi h missing alues in some u ban
physical exposu es, he Silhoue e me hod iden i ied wo clus e s. In he
ollowing ully adjus ed linea eg ession models, no signi ican di e -
ences in any o he leisu e- ime physical ac i i y measu es we e ound
be ween he clus e s. Repea ed gene alized linea eg ession analyses
e ealed 25 u ban physical exposu es signi ican ly associa ed wi h o al
LTPA, consis en wi h he analysis using impu ed da a, and 26 exposu es
wi h LTPA, wo mo e han he analysis wi h impu ed da a. S ill, no
Table 1
Cha ac e is ics o sociodemog aphic, beha io , and neighbo hood social a i-
ables (pa icipan s n =394).
Cha ac e is ics N. (%)/Mean (SD)
Sex
Male 179 (45.4)
Female 215 (54.6)
Wo k
No wo king o o he si ua ion 51 (12.9)
Cu en ly wo k 343 (87.1)
Educa ion
Pos -seconda y o lowe 84 (21.3)
Bachelo /equi alen o abo e 310 (78.7)
Li ing s a us
Wi h a spouse o a pa ne 118 (30.0)
Wi h a spouse o a pa ne and child( en) 151 (38.3)
O he si ua ion 125 (31.7)
Age (yea s) 37.1 (1.5)
Illici subs ance use
No 204 (51.8)
Yes 190 (48.2)
E e smoke (smoked o e 100 ciga e es li e ime)
No 216 (54.8)
Yes 178 (45.2)
Alcohol
Mon hly o less, o e en ne e 164 (41.6)
2-4 imes a mon h o mo e 230 (58.4)
Dep i a ion le el
Low 225 (57.1)
High 169 (42.9)
The p opo ion o single households in he neighbo hood 50.0 (10.8)
Z. Wang e al.
Heal h and Place 94 (2025) 103495
5
u ban physical exposu e eached he 0.01 P- alue h eshold o associ-
a ion wi h commu ing ac i i y.
A o al o 28 PCs explained o e 80 % o he o al a iance o 145
u ban physical exposu es. The esul s o XGBoos models o o al LTPA,
LTPA, and commu ing ac i i y, which included PCs we e p esen ed in
Supplemen al Fig. 5. PC1 was he mos impo an o p edic bo h o al
LTPA and LTPA, and he ege a ion indices and he pe cen age o build-
up a eas had high coe icien s on PC1. PC2 and PC15 we e he second
impo an PCs o p edic o al LTPA and LTPA, espec i ely. PC25 and
PC27 we e he op wo impo an PCs o p edic commu ing ac i i y.
Fig. 2. Twin pa icipan s’ esidence in he Helsinki and Espoo a ea in 2018 colo ed by clus e
No e: The g ay laye shows cen e s and shopping a eas in 2019.
Table 2
Resul s o he linea eg ession be ween he u ban clus e and physical ac i i y measu es.
Ou come (log- ans o med) Cha ac e is ics Model 1
a
Model 2
b
Model 3
c
Be a (95 % CI) R
2
Be a (95 % CI) R
2
Be a (95 % CI) R
2
To al LTPA U ban clus e
1 (o iginal ci y cen e ) Re . 0.08 Re . 0.08 Re . 0.09
2 (new ci y cen e ) −0.06 (−0.13, 0.01) −0.06 (−0.13, 0.01) −0.05 (−0.13, 0.02)
3 (subu ban) −0.13 (−0.21, −0.06)* −0.13 (−0.21, −0.05)* −0.12 (−0.22, −0.03)*
LTPA U ban clus e
1 (o iginal ci y cen e ) Re . 0.08 Re . 0.08 Re . 0.08
2 (new ci y cen e ) −0.07 (−0.15, 0.01) −0.06 (−0.15, 0.02) −0.07 (−0.16, 0.03)
3 (subu ban) −0.16 (−0.25, −0.07)* −0.16 (−0.25, −0.06)* −0.16 (−0.28, −0.05)*
Commu ing ac i i y U ban clus e
1 (o iginal ci y cen e ) Re . 0.03 Re . 0.06 Re . 0.06
2 (new ci y cen e ) −0.01 (−0.06, 0.03) −0.01 (−0.05, 0.04) 0.00 (−0.04, 0.05)
3 (subu ban) −0.03 (−0.08, 0.02) −0.03 (−0.08, 0.02) −0.01 (−0.07, 0.05)
*P <0.05.
a
Adjus ed o age, sex, educa ion, wo k, and li ing s a us.
b
Based on model 1, addi ionally adjus ed o smoking, alcohol d inking, and illici subs ance use.
c
Based on model 2, addi ionally adjus ed o neighbo hood dep i a ion le el and he p opo ion o single households in he neighbo hood.
Z. Wang e al.
Heal h and Place 94 (2025) 103495
6
Fig. 3. Resul s o XGBoos models o o al leisu e- ime physical ac i i y ( o al LTPA) and leisu e- ime physical ac i i y wi hou commu ing ac i i y (LTPA)
No e: The SHAP ba plo s show he in luence o each a iable: o al LTPA (a) and LTPA (e). The SHAP dependence plo s show how a single indi idual in luences he
XGboos p edic ion on o al LTPA (b, c, d) and LTPA ( , g, h). in s_500 is he coun o any ype o oad junc ions wi hin a 500 m bu e ; suma ea_pocke pa ks_800 is
he o al a ea o all in e connec ed pocke pa ks wi hin an 800 m walking dis ance; nd i_5y s_all_500 is he 5-yea s mo ing a e age o No malized Di e ence
Vege a ion Index wi hin a 500 m bu e du ing whole yea ; coun _pocke pa ks_800 is he coun o pocke pa ks wi hin an 800 m walking dis ance. Abb e ia ion:
leisu e- ime physical ac i i y (LTPA); SHapley Addi i e exPlana ions (SHAP).
Z. Wang e al.
Heal h and Place 94 (2025) 103495
7
4. Discussion
We used clus e ing analysis and XGBoos o simul aneously and
comp ehensi ely s udy he e ec o 145 u ban physical exposu es on
leisu e- ime physical ac i i y in 394 Finnish adul s in hei ea ly midli e.
Th ough ans o ming he high-dimensional exposu e da a in o in o -
ma i e and compac clus e index, we ha e iden i ied h ee clus e s
named: “o iginal ci y cen e ”, “new ci y cen e ”, and “subu ban”, based
on he isual obse a ion om he map. We ound people li ing in
subu ban a eas had a lowe le el o physical ac i i y in leisu e ime
compa ed o hose li ing in he o iginal ci y cen e . The e was no di -
e ence be ween “o iginal ci y cen e ” and “new ci y cen e ” clus e s.
XGBoos models e ealed a complex ela ionship be ween he u ban
physical exposome and leisu e- ime physical ac i i ies, in which
impo an exposu es showed non-linea i y and looked like h eshold
a iables. Mo e oad junc ions and sho e dis ances o he closes oad
co ela ed wi h highe le els o o al LTPA (i.e. he highe mean MET
hou s pe day). NDVI wi hin a 500 m bu e below 0.4 o o e 0.6 was
co ela ed wi h highe le els o o al LTPA, as well. Mo eo e , highe
amoun s o ege a ion g eenness (indica ed by MSAVI) and smalle and
less pocke pa ks wi hin 800 m walking dis ance we e associa ed wi h
lowe le els o LTPA. Compa ing esul s o o al LTPA and LTPA, he
oad cha ac e is ics became mo e impo an when conside ing
commu ing ac i i y in leisu e ime. We did no ind any conside able
in e ac ion be ween u ban physical exposu es con ibu ing o leisu e-
ime physical ac i i ies. These indings we e co ela ional wi hou
u he causal assessmen .
P e ious esea ch has documen ed he ela ionship be ween
di e en le els o u baniza ion and physical ac i i y among he u ban
esiden s bu wi h inconsis en indings ega ding he di ec ion o e -
ec s. A c oss-sec ional s udy in Shanghai, China wi h 327 esponden s
(mean age: 40) simila ly epo ed highe leisu e- ime physical ac i i y
among down own esiden s compa ed o subu ban dwelle s, which
adhe ed o ou indings (Zhou e al., 2013). A Finnish s udy indica ed
ha child en li ing in u ban a eas had a lowe isk o o e weigh and
obesi y, a consequence o ac i e physical ac i i y (M¨
aki e al., 2023).
Signi ican esul s we e also ound o anspo a ion ac i i ies in his
Chinese s udy, while we ound a null esul (Zhou e al., 2013). A Ca-
nadian s udy showed ha he physical ac i i y le el was highe in u ban
han in subu ban among adolescen s om schools in lowe
socio-economic a eas (Shea e e al., 2012). Ne e heless, a sys ema ic
e iew sugges ed ha child en and eenage s who li e in subu ban a eas
we e mo e physically ac i e han in u al and u ban a eas (Sande cock
e al., 2010), and, simila o he Shanghai s udy abo e, a na ionwide
s udy in China showed ha ising u baniza ion co ela es wi h longe
commu ing imes among adul s (mean age: 45) (Zhu e al., 2017). In US,
only male adolescen s li ing in u ban a eas engaged in mo e
mode a e- o- igo ous physical ac i i y han hose li ing in subu ban
a eas (Moo e e al., 2014). The inconsis ency be ween li e a u e and ou
indings may be due o di e en popula ion cha ac e is ics, spo s cul-
u es, coun y con ex s, u ban planning, o u banici y de ini ions.
Ins ead o a p e-de ini ion o (sub)u ban a eas by go e nmen al guide-
lines, we used an unsupe ised da a-d i en clus e ing me hod o
de e mine he e ogeneous u ban en i onmen s wi hin u ban a eas
e lec ing eal-li e exposu e modes and accoun ing o co ela ion, ad-
di i e, and mix u e e ec s (Guillien e al., 2021).
XGBoos models anked he elemen s o oad cha ac e is ics, g een-
spaces, and pocke pa ks, as s ongly associa ed wi h leisu e- ime
physical ac i i ies among ea ly midli e adul s. Mo e oad junc ions
and sho e dis ances o he closes oad sugges be e connec i i y and
accessibili y, c ea ing a mo e con enien en i onmen o people o
walk o bike o hei des ina ions, A e iew o quali a i e s udies indi-
ca ed access o local des ina ions and ac i e anspo a ion co ela ed
wi h physical ac i i y among adul s (Sal o e al., 2018). A Finnish s udy
ound ha he densi y o in e sec ions, de ined as he junc ion o a
minimum o h ee oads, was posi i ely associa ed wi h he numbe o
physical ac i i y bou s and he le el o mode a e o igo ous physical
ac i i y among olde adul s (Keskinen e al., 2020). Zang e al. (2022)
iden i ied he in e sec ion densi y, as well as s ee scape g eene y, as he
mos impo an physical exposu e con ibu ing o ligh physical ac i i y
among olde adul s, by andom o es models. Howe e , he ela ionship
be ween s ee connec i i y, in ol ing he numbe o in e sec ions, and
physical ac i i y in all age g oups o adul s a ied ac oss di e en bu e
a eas in u ban en i onmen s, sugges ing he complexi y o u ban li ing
en i onmen s (McGinn e al., 2007). Connec i i y we e also shown o be
associa ed wi h anspo a ion (McCo mack and Shiell, 2011), al hough
we did no ind any impo an ac o s o commu ing ac i i ies, he ole o
oads became mo e p ominen when commu ing ac i i ies we e
included in leisu e- ime physical ac i i ies. Whe e he associa ion o
g eenspace wi h physical ac i i y is ela i ely inconsis en (B owning
e al., 2022), ou indings show an associa ion in which su ounding
g eenness is posi i ely associa ed wi h LTPA up un il a h eshold o 0.4
NDVI, wi h highe NDVI ela ing o lowe LTPA. High le els o g een-
space migh e lec subu ban li ing o some ex en , and o he g eenspace
indica o s, such as accessibili y, we e no p ominen . The ela ionship
be ween g eenspace and physical ac i i y could be mode a ed by he
le el o u baniza ion (B owning e al., 2022). O he s udies ha e simila
indings on he h eshold e ec . Fo example, he posi i e associa ion o
physical ac i i y wi h mul iple g eenspace uses indica o s eached o
peak when indica o s we e wi hin a 600 m bu e (Ca dinali e al.,
2024). Zang e al. (2022) also ound ha s ee scape g eene y had a
posi i e e ec on ligh physical ac i i y when i anged om 0.12 o 0.15
poin , co esponding o a low le el o isible g eene y. Besides, ano he
Chinese s udy also iden i ied he 0.4 NDVI, co esponding o a eas wi h
spa se o mode a e ege a ion, as he u ning poin o i s associa ion
wi h sel - a ed heal h among he old popula ion (Huang e al., 2022),
and sel - a ed heal h closely co ela ed wi h physical ac i i y (Guan,
2022). Fo pocke pa ks, a na u al expe imen al s udy in low-income
Ame ican neighbo hoods ound a highe le el o leisu e- ime exe cise
among middle-aged esiden s a e pocke pa ks we e cons uc ed
(Cohen e al., 2014). Use s o pocke pa ks, de ined as li ing wi hin a 0.5
mile (~800 m) adius, had highe exe cise le els han adi ional pa k
use s (Cohen e al., 2014). Resea che s u he summa ized ha pocke
pa ks we e cos -e ec i e o p omo ing physical ac i i y in inne -ci y
a eas (Cohen e al., 2014). A s udy in Chongqing, China, u ilizing in-
e iews on concep ual unde s anding o pa k images, e ealed ha he
en i onmen al cha ac e is ics o pocke pa ks con ibu ed o a es o -
a i e e ec in ol ing en e ainmen ac i i ies and elie (Peng e al.,
2023). No ewo hy, a ecen Chinese s udy using Ligh
G adien -Boos ing Machine model ound ha ec ea ional acili ies we e
he mos impo an ac o o walking beha io in old adul s bu he
numbe o pa ks was he leas impo an among 11 ac o s, highligh ing
he speci ic e ec d i en by he con en inside pa ks o ec ea ion a eas
(Yang e al., 2024). This anno a ion on bene i h eshold may p o ide
mo e p ecise guidance on u ban planning, bu mo e eplica ion and
explo a ion is needed o alida e hese indings o coun y- o
Eu opean-wide implica ions.
The iden i ied exposu es may be elabo a ed by ac o s such as
accessibili y, oppo uni ies, and aes he ic a ibu es indi idual el o
pe cei ed om he be e a ailabili y and quali y o g een ec ea ional
spaces and imp o ed connec i i y. A sys ema ic e iew demons a ed
ha accessibili y o acili ies, he p esence o sidewalks, and aes he ic
ea u es posi i ely co ela e wi h physical ac i i y pa icipa ion,
including bo h leisu e and commu ing ac i i ies, among adul s (Choi
e al., 2017). Pe cei ed neighbo hood en i onmen al ac o s such as
s ee connec i i y we e co ela ed wi h mode a e o igo ous physical
ac i i y and commu ing ac i i y (Ca lson e al., 2018). Social cogni i e
heo y u he sugges s ha indi iduals a e shaped by and espond o
hei social and buil en i onmen s. They also engage in o e hough ,
sel - egula ion, sel - e lec ion, ica ious lea ning, and inno a ion
h ough imagina ion and communica ion (Bandu a, 1986). En i on-
men al ac o s we e co ela ed wi h sel -e icacy belie s (Beauchamp
Z. Wang e al.
Heal h and Place 94 (2025) 103495
8
e al., 2019), which ha e been consis en ly linked o physical ac i i y
(Young e al., 2014). Among adolescen s, sel -e icacy o o e coming
ba ie s was shown o media e he e ec o pe cei ed equipmen
accessibili y on physical ac i i y in a c oss-sec ional s udy (Mo l e al.,
2005). While ou analyses used objec i e en i onmen al measu es,
u u e s udies in eg a ing pe cei ed en i onmen al and in e pe sonal
ac o s could p o ide deepe insigh s in o he mechanisms unde lying
beha io o in o ming e ec i e in e en ions (Beauchamp e al., 2019).
Besides i s s eng h, his s udy is no wi hou limi a ions. Fi s , he
sample size was ela i ely small compa ed o o he exposome s udies.
Al hough he sample size o K-p o o ype clus e ing (o e 10 imes he
numbe o clus e s) and subsequen eg ession seems o be adequa e,
inconsis ency in addi ional XGBoos models highligh s he need o a
la ge sample. Addi ionally, due o he complexi y o he la ge-
dimensional exposome se , he modes sample size made cap u ing
ela i ely small in e ac ions mo e challenging. The win design was also
no employed o con ol o he unobse ed indi idual he e ogenei y
(con ounding). Second, only pa icipan s om he i e la ges ci ies in
Finland we e included, limi ing he gene alizabili y. The included pa -
icipan s ep esen ed 19 % o wins who ha e comple ed ea ly midli e
ollow-up, while he i e ci ies co e ed app oxima ely 28 % o he o al
Finnish popula ion. Besides, we did no include any pa icipan s li ing
in u al a eas. No only he physical en i onmen , bu li es yles may also
di e be ween u ban and u al a eas. The e o e, he in e p e a ion
should be na owed down o speci ic ypes o ci ies. Thi d, u ban
physical exposu es we e based on esiden ial add esses, which o e look
dynamic human beha io s ou side he home, leading o measu emen
e o s. In addi ion, he used esiden ial geocodes co esponded o pa -
icipan s’ esidences in 2017, 2018, o 2020, wi hou accoun ing o
how long hey li ed a hose add esses. Measu emen e o s could skew
ou iden i ica ion o key de e minan s, as exposu es wi h la ge e o s
migh show weake associa ions and be classi ied as less in luen ial,
e en i hey a e ac ually mo e impo an han hose iden i ied as mos
in luen ial. Mo e g anula and accu a e es ima ions o exposu e and
beha io could acili a e he explo a ion in he dynamic in e ac ion
be ween he en i onmen and human beha io (Sonnenschein e al.,
2022). Fou h, some exposu es we e a ailable in 2023 bu me ged wi h
he add ess in 2020, posing a empo ali y issue. The ela i ely slow
u ban enewal and cons uc ion in Finland educed he conce n (E e s
e al., 2024). Fi h, missing alues in exposu es may in oduce bias.
Excluding pa icipan s wi h missing alues al e ed he op imal numbe
o clus e s, while he numbe o signi ican associa ions be ween expo-
su es and ou comes emained simila o he numbe based on impu ed
da a. Gi en ha only abou 3 % o pa icipan s had missing alues, he
e ec is likely modes , bu cau ion is s ill wa an ed. Six h, leisu e- ime
physical ac i i y was sel - epo ed. MET alues used in he cu en s udy
we e gi en by he Compendium o Physical Ac i i y being es ima es o
he es ing oxygen consump ion o an a e age man, e en hough he
ene gy expendi u e is always ela ed o ac o s as body weigh , body
composi ion and mo emen e iciency (Ainswo h e al., 2011; He -
mann e al., 2024). The de ice-based measu emen o leisu e- ime
physical ac i i y would ha e been mo e accu a e. Howe e , he al-
idi y o leisu e- ime physical ac i i y ques ions used in Finnish wins has
been demons a ed (Leskinen e al., 2009; Walle e al., 2008). Se en h,
he naming o clus e s may in oduce subjec i i y and o e simpli y
complex u ban en i onmen al pa e ns, e en hough ou unsupe ised
clus e ing me hod iden i ied no el s uc u es. The agnos ic, da a-d i en
na u e o he esul s should be ca e ully conside ed in he in e p e a ion.
Eigh h, he e was no causal assessmen . Unmeasu ed con ounding
(al hough a wide ange o exposu es and co a ia es included), e e se
ela ionships, and o he issues emain as conce ns; ou pu s may be o e -
o unde -es ima ed. In sho , a complex associa ion was obse ed, bu
due o he co ela ional and obse a ional na u e o he s udy and hese
men ioned limi a ions, i m causal conclusions canno be d awn, and
u he endea o is needed o explo e unde lying mechanisms wi h a
igo ous design.
5. Conclusion
This s udy employed wo analy ical app oaches o explo e he
in ica e co ela ional ela ionship be ween he u ban physical expo-
some and leisu e- ime physical ac i i y in ea ly midli e in Finland.
Clus e ing analysis e ealed h ee he e ogeneous pa e ns o u ban en-
i onmen s. Li ing in subu ban a eas was associa ed wi h lowe le els o
leisu e- ime physical ac i i y han in o iginal ci y cen e a eas. XGBoos
models iden i ied pocke pa ks, oad cha ac e is ics, and g eenspaces as
in luen ial ac o s wi h non-linea ela ionships, which beha ed like
h eshold a iables. Gi en limi a ions in causali y, sample size, gene -
alizabili y, and measu emen g anula i y, we call o u he s udies in
o he se ings o eplica e ou analyses. We s ill ad oca e p esen ing his
e idence o s akeholde s and policymake s o conside he complexi y in
de eloping ailo ed in e en ions on some u ban ea u es o achie e
highe cos -e ec i eness by ocusing on he mos in luen ial de-
e minan s and hei op imal anges in add essing he challenge o he
physically inac i e li es yle in ou apidly u banizing wo ld.
CRediT au ho ship con ibu ion s a emen
Zhiyang Wang: W i ing – o iginal d a , Visualiza ion, Me hodol-
ogy, In es iga ion, Fo mal analysis, Da a cu a ion, Concep ualiza ion.
Sa i Aal onen: W i ing – e iew & edi ing, Resou ces, Me hodology,
In es iga ion, Da a cu a ion, Concep ualiza ion. Roos Teeuwen:
W i ing – e iew & edi ing, Da a cu a ion. Vasileios Milias: W i ing –
e iew & edi ing, Da a cu a ion. Ca men Peu e s: W i ing – e iew &
edi ing, Me hodology. B uno Raimbaul : Da a cu a ion. Teemu Pal-
iainen: W i ing – e iew & edi ing, Da a cu a ion. E in Lumpe:
W i ing – e iew & edi ing, Me hodology. Danielle Dick: W i ing –
e iew & edi ing, Resou ces, Da a cu a ion. Jessica E. Sal a o e:
W i ing – e iew & edi ing, Resou ces, Da a cu a ion. Ma ia Fo as e :
Da a cu a ion, Concep ualiza ion. Payam Dad and: W i ing – e iew &
edi ing, Concep ualiza ion. Jo di Júl ez: W i ing – e iew & edi ing,
Funding acquisi ion. Achilleas Psyllidis: W i ing – e iew & edi ing,
Funding acquisi ion, Da a cu a ion, Concep ualiza ion. I ene an
Kamp: W i ing – e iew & edi ing, P ojec adminis a ion, Funding
acquisi ion, Concep ualiza ion. Jaakko Kap io: W i ing – e iew &
edi ing, Supe ision, P ojec adminis a ion, Funding acquisi ion,
Concep ualiza ion.
Consen o pa icipa e
All pa icipan s and hei pa en s/legal gua dians ga e in o med
w i en consen o pa icipa e in he s udy. The au ho s asse ha all
p ocedu es con ibu ing o his wo k comply wi h he e hical s anda ds
o he ele an na ional and ins i u ional commi ees on human expe -
imen a ion and wi h he Helsinki Decla a ion o 1975, as e ised in
2008.
E hics app o al and consen o pa icipa e
The e hics commi ee o he Depa men o Public Heal h o he
Uni e si y o Helsinki (Helsinki, Finland) and he Ins i u ional Re iew
Boa d o Indiana Uni e si y (Blooming on, Indiana, USA) app o ed he
FinnTwin12 s udy p o ocol om he s a o he coho . The e hical
app o al o he e hics commi ee o he Helsinki Uni e si y Cen al
Hospi al Dis ic (HUS) is he mos ecen and co e s he mos ecen
da a collec ion (ea ly midli e) (HUS/2226/2021, da ed Sep embe 22,
2021).
A ailabili y o da a and ma e ials
The FinnTwin12 da a a e no publicly a ailable due o he e-
s ic ions o in o med consen . Howe e , he FinnTwin12 da a a e
a ailable h ough he Ins i u e o Molecula Medicine Finland (FIMM)
Z. Wang e al.
Heal h and Place 94 (2025) 103495
9
