Path analyses of risk factors for linear growth faltering in four prospective cohorts of young children in Ghana, Malawi and Burkina Faso

1 P adoEL, e al. BMJ Glob Heal h 2019;4:e001155. doi:10.1136/bmjgh-2018-001155 Pa h analyses o isk ac o s o linea g ow h al e ing in ou p ospec i e coho s o young child en in Ghana, Malawi and Bu kina Faso Elizabe h L P ado,1 Elizabe h Yakes Jimenez,2 S ephen Vos i,3 Robe S ewa ,4 Ch is ine P S ewa ,1 Jé ôme Somé,1,5 Anna Pulakka,6 Jean Bosco Ouéd aogo,5 Ha ie Ok onipa,1 Eugenia Ocansey,1 B ie a Oaks,1,7 Kenne h Male a,8 Anna La ey,9 Emma Ko ekangas,10 Sonja Y Hess,1 Kenne h B own,1,11 Jaden Bendabenda,8 Ulla Asho n,10 Pe Asho n,10,12 Ma y A imond,13 Se h Adu-A a wuah,9 Souheila Abbeddou,1 Ka h yn Dewey1 Resea ch To ci e: P adoEL, Yakes JimenezE, Vos iS, e al. Pa h analyses o isk ac o s o linea g ow h al e ing in ou p ospec i e coho s o young child en in Ghana, Malawi and Bu kina Faso. BMJ Glob Heal h 2019;4:e001155. doi:10.1136/ bmjgh-2018-001155 Handling edi o Sanni Yaya ►Addi ional ma e ial is published online only. To iew please isi he jou nal online (h p:// dx. doi. o g/ 10. 1136/ bmjgh- 2018- 001155). Recei ed 4 Sep embe 2018 Re ised 6 No embe 2018 Accep ed 17 No embe 2018 Fo numbe ed a ilia ions see end o a icle. Co espondence o D Elizabe h L P ado; elp ado@ ucda is. edu © Au ho (s) (o hei employe (s)) 2019. Re-use pe mi ed unde CC BY. Published by BMJ. Key ques ion Wha is al eady known? ►P e ious c oss-sec ional s udies ha e iden i ied key isk ac o s o s un ed g ow h, including being small o ges a ional age a bi h, poo sani a ion, child- hood dia hoea, and low die a y di e si y and in e- quen in an eeding. Wha a e he new indings? ►We ound consis en associa ions o 18-mon h leng h- o -age z-sco es (LAZ) wi h ma e nal heigh and ma e nal body mass index. ►The ac o s wi h he s onges associa ions wi h 18-mon h LAZ we e leng h o ges a ional age z-sco e (LGAZ) a bi h, ma e nal heigh and ges a- ional age a bi h. ►O he ac o s ha showed signi ican associa ions wi h 18-mon h LAZ (al hough wi h smalle coe i- cien s), we e imp o ed household wa e sou ce, child die a y di e si y, childhood dia hoea incidence and 6-mon h o 9-mon h haemoglobin concen a ion. Wha do he new indings imply? ►In e en ions a ge ing hese ac o s associa ed wi h LAZ may accele a e p og ess owa ds educ- ing s un ing; howe e , much o he a iance in linea g ow h s a us emained unaccoun ed o by indi id- ual-le el ac o s, sugges ing ha communi y-le el changes may be needed o achie e subs an ial p og- ess and u he esea ch is needed o unde s and he causes o s un ing. AbsT ACT S un ing p e alence is an indica o o a coun y’s p og ess owa ds Uni ed Na ions’ Sus ainable De elopmen Goal 2, which is o end hunge and achie e imp o ed nu i ion. Accele a ing p og ess owa ds educing s un ing equi es a deepe unde s anding o he ac o s ha con ibu e o linea g ow h al e ing. We conduc ed pa h analyses o ac o s associa ed wi h 18-mon h leng h- o -age z-sco e (LAZ) in ou p ospec i e coho s o child en who pa icipa ed in ials conduc ed as pa o he In e na ional Lipid-Based Nu ien Supplemen s P ojec in Ghana (n=1039), Malawi (n=684 and 1504) and Bu kina Faso (n=2619). In wo coho s, women we e en olled du ing p egnancy. In wo o he coho s, in an s we e en olled a 6 o 9 mon hs. We examined he associa ion o 42 indica o s o en i onmen al, ma e nal, ca egi ing and child ac o s wi h 18-mon h LAZ. Using s uc u al equa ion modelling, we examined di ec and indi ec associa ions h ough hypo hesised media o s in each coho . Ou o 42 indica o s, 2 we e associa ed wi h 18-mon h LAZ in h ee o ou coho s: ma e nal heigh and body mass index (BMI). Six ac o s we e associa ed wi h 18-mon h LAZ in wo coho s: leng h o ges a ional age z-sco e (LGAZ) a bi h, p egnancy du a ion, imp o ed household wa e , child die a y di e si y, dia hoea incidence and 6-mon h o 9-mon h haemoglobin concen a ion. Di ec associa ions we e mo e p e alen han indi ec associa ions, bu 30%–62% o he associa ions o ma e nal heigh and BMI wi h 18-mon h LAZ we e media ed by LGAZ a bi h. Fac o s ha we e no associa ed wi h LAZ we e ma e nal i on s a us, illness and in lamma ion du ing p egnancy, ma e nal s ess and dep ession, exclusi e b eas eeding du ing 6 mon hs pos pa um, eeding equency and child e e , mala ia and acu e espi a o y in ec ions. These indings may help in iden i ying in e en ions o accele a e p og ess owa ds educing s un ing; howe e , much o he a iance in linea g ow h s a us emained unaccoun ed o by hese 42 indi idual-le el ac o s, sugges ing ha communi y-le el changes may be needed o achie e subs an ial p og ess. InT oduCTIon Linea g ow h al e ing du ing ea ly li e is associa ed wi h la e heal h condi ions, such as ca diome abolic disease,1 and poo cogni- i e and school pe o mance.2 S un ing occu s when child en al e in linea g ow h on 30 Ap il 2019 by gues . P o ec ed by copy igh .h p://gh.bmj.com/BMJ Glob Heal h: i s published as 10.1136/bmjgh-2018-001155 on 13 Janua y 2019. Downloaded om 2P adoEL, e al. BMJ Glob Heal h 2019;4:e001155. doi:10.1136/bmjgh-2018-001155 BMJ Global Heal h ea ly in li e and is de ined as heigh - o -age >2 SD below he median o he WHO Child G ow h S and- a ds.3 S un ing p e alence is an impo an indica o o e alua e a coun y’s p og ess owa ds Uni ed Na ions’ Sus ainable De elopmen Goal 2, which is o end hunge and achie e imp o ed nu i ion by 2030. F om 2000 o 2017, he es ima ed numbe o s un ed chil- d en unde 5 yea s o age globally dec eased om 198 million o 151 million.4 Accele a ing p og ess owa ds educing he p e alence o s un ing equi es a mo e ad anced unde s anding o he ac o s ha con ibu e o ea ly linea g ow h al e ing. Se e al p e ious e iews ha e summa ised hese isk ac o s,5–7 iden i ying en i onmen al ac o s such as household ood insecu i y and poo quali y wa e and sani a ion; ma e nal ac o s such as sho s a u e and poo nu i ion and heal h du ing p egnancy; ca e- gi ing ac o s, such as in equen eeding and low die a y di e si y; and child ac o s such as being bo n p e e m, small o ges a ional age and childhood dia - hoea incidence. F amewo ks ha desc ibe he de e - minan s o childhood malnu i ion5 6 ecognise ha dis al ac o s (eg, communi y, household le el) ope a e h ough mo e p oximal ac o s (eg, ma e nal, child le el) in hei in luence on g ow h al e ing. Howe e , mos s udies ha ha e a emp ed o quan i y he ela- i e con ibu ion o isk ac o s o linea g ow h al e ing ha e used c oss-sec ional analyses, such as he es ima ion o eg ession coe icien s8–10 o popula- ion-a ibu able ac ion.11 12 These me hods desc ibe di ec associa ions, bu do no accoun o he indi ec associa ions o dis al ac o s h ough p oximal ac o s. In he cu en s udy, we pe o med a se o pa h anal- yses o ac o s associa ed wi h linea g ow h s a us a age 18 mon hs in ou longi udinal coho s o child en (n=5846) who pa icipa ed in he In e na ional Lipid- Based Nu ien Supplemen s (iLiNS) P ojec in Ghana, Malawi and Bu kina Faso. The i s objec i e was o iden i y he ac o s associa ed wi h 18-mon h linea g ow h s a us, e lec ing cumula i e linea g ow h du ing ges a ion and he i s 18 mon hs a e bi h, in each iLiNS coho . The second objec i e was o iden i y he pa hways h ough which hese ac o s ope a e. The p ospec i e design enabled examina ion o ac o s asso- cia ed wi h linea g ow h h oughou mos o he i s 1000 days a e concep ion. Ha monisa ion o many o he a iables ac oss coho s allowed examina ion o he same ac o s and pa hways in di e en con ex s. Finally, ou hi d objec i e was o de e mine whe he a pooled analysis examining p edic o s o 18-mon h leng h- o -age z-sco e (LAZ) ac oss coho s accoun ed o mo e a iance in LAZ han he wi hin-coho models. E idence sugges s ha linea g ow h al e ing is a whole-popula ion phenomenon. In popula ions wi h a high p e alence o s un ing, i is no he case ha a subg oup o ha popula ion is al e ing while he es a e g owing no mally, bu ins ead he e is usually a downwa d shi in he en i e heigh dis ibu ion o he popula ion.13 I be ween-popula ion di e ences a e mo e meaning ul han wi hin-popula ion di e ences be ween indi iduals, hen we would expec he pooled analysis o accoun o mo e a iance han he analyses o each sepa a e coho . MeTHods Based on se e al p e ious amewo ks,5 6 14 we de el- oped a concep ual pa h model o po en ial in luences on 18-mon h linea g ow h s a us ( igu e 1). We es ed he ollowing pa hways, which co espond o he labels o he a ows in igu e 1. Pa hway (1) A an indi idual le el, ma e nal heigh may be ela ed o he child’s gene ic po en ial o adul heigh ha can be a ained. In popula ions wi h a high p e alence o s un ing, ma e nal heigh is also pa ly a e lec ion o g ow h es ic ion expe ienced by he mo he du ing ea ly li e. The e o e, inclusion o ma e nal heigh in he model se ed wo pu poses: i s , o adjus o a p oxy o gene ic po en ial, and second, o es he pa hway ha in e gene a ional e ec s o ma e nal g ow h du ing ea ly li e, e lec ed by ma e nal adul heigh , on child linea g ow h may be media ed by (1.1) socioeconomic condi ions o he cu en gene - a ion, (1.2) ma e nal adul ac o s (nu i ional s a us, illness, s ess, dep ession, cogni ion), (1.3) ca egi ing p ac ices, (1.4) child ac o s o (1.5) may di ec ly a ec linea g ow h (a ows o each indi idual pa hway no d awn in he igu e). Pa hway (2) Socioeconomic dispa i ies and o he en i onmen al e ec s on child linea g ow h may be media ed by (2.1) ma e nal ac o s, (2.2) ca egi ing p ac ices, (2.3) child ac o s o (2.4) may di ec ly a ec linea g ow h. Pa hway (3) E ec s o ma e nal ac o s on child g ow h may be media ed by (3.1) child ac o s, (3.2) ca egi ing p ac- ices o (3.3) may di ec ly a ec child g ow h. Pa hway (4) E ec s o in an eeding p ac ices on child g ow h may be media ed by child ac o s (4.1) o ca egi ing p ac- ices may di ec ly a ec child g ow h (4.2). Pa hway (5) E ec s o p e e m bi h o in au e ine g ow h es ic- ion on la e child linea g ow h s a us may be medi- a ed by (5.1) pos na al child ac o s (appe i e, illness, haemoglobin (Hb)/i on s a us, physical ac i i y, s ess) o (5.2) may be di ec e ec s. Pa hway (6) E ec s o child ac o s (appe i e, illness, Hb/i on s a us, physical ac i i y, s ess) on child g ow h (6.1) on 30 Ap il 2019 by gues . P o ec ed by copy igh .h p://gh.bmj.com/BMJ Glob Heal h: i s published as 10.1136/bmjgh-2018-001155 on 13 Janua y 2019. Downloaded om P adoEL, e al. BMJ Glob Heal h 2019;4:e001155. doi:10.1136/bmjgh-2018-001155 3 BMJ Global Heal h Figu e 1 Concep ual model. may be media ed by ca egi ing beha iou in esponse o hese ac o s, o (6.2) may be di ec e ec s. iLiNS p ojec ial designs In he iLiNS-DYAD-G ial in Ghana (n=1320) and he iLiNS-DYAD-M ial in Malawi (n=869), p egnan women we e en olled a ≤20 weeks o ges a ion. In he iLiNS- DOSE ial in Malawi (n=1932) and iLiNS-ZINC ial in Bu kina Faso (n=3220), in an s we e en olled a age 6 and 9 mon hs, espec i ely. All pa icipan s we e assigned o ecei e a ious doses and o mula ions o lipid-based nu ien supplemen s (LNS), o o con ol g oups un il age 18 mon hs, when leng h was measu ed.15–18 The e ec s o he in e en ions on 18-mon h child leng h- o LAZ di e ed ac oss ials, wi h posi i e e ec s in Bu kina Faso17 and Ghana,15 bu no in Malawi.16 18 Fo u he in o ma ion, see supplemen al me hods. Pa icipan s In he pa h analyses epo ed he e, we included all child en o whom LAZ a age 18 mon hs was a ailable, comp ising 1039 child en in iLiNS-DYAD-G, 684 in iLiNS-DYAD-M, 1504 in iLiNS-DOSE and 2619 child en in iLiNS-ZINC. P ocedu e De ailed epo s o he da a collec ion p ocedu es in each ial ha e been published elsewhe e15–18; he e- o e, we summa ise he p ocedu es o collec ion o a iables ha we e used in he analyses p esen ed he e. Online supplemen a y able 1 p esen s u he de ails o he da a collec ion p ocedu es and a iable de ini- ions. Figu e 2 shows he da a collec ion schedule o each a iable in each coho . Da a on socio-demog aphic cha ac e is ics and ma e nal an h opome ic s a us we e ga he ed a en ol- men . In he wo DYAD ials, ma e nal p e-p egnancy body mass index (BMI) was es ima ed based on BMI and ges a ional age a en olmen . Capilla y o enous blood samples we e collec ed om mo he s and/o child en a mul iple ime poin s o he assessmen o on 30 Ap il 2019 by gues . P o ec ed by copy igh .h p://gh.bmj.com/BMJ Glob Heal h: i s published as 10.1136/bmjgh-2018-001155 on 13 Janua y 2019. Downloaded om 4P adoEL, e al. BMJ Glob Heal h 2019;4:e001155. doi:10.1136/bmjgh-2018-001155 BMJ Global Heal h Figu e 2 Da a collec ion imeline in each coho . (1) mala ia using a apid diagnos ic es , (2) Hb concen- a ion (g/L), (3) bioma ke s o i on s a us, including zinc p o opo phy in (ZPP) concen a ion (μmol/mol heme), and soluble ans e in ecep o (sT R, mg/L), and (4) bioma ke s o in lamma ion, including alpha- 1-acid glycop o ein (AGP; g/L) concen a ion. In ZINC, known HIV in ec ion was an exclusion c i e ion, bu HIV was no es ed; he e o e, HIV s a us o women who we e en olled was unknown. In DOSE, HIV s a us was also unknown. In DYAD-G, HIV in ec ion was known om an ena al ca ds and HIV-posi i e women we e excluded. In DYAD-M, women we e es ed o HIV a en olmen bu we e no excluded. Ma e nal and/o child sali a samples in DYAD-M and DYAD-G we e collec ed a se e al ime poin s o he measu emen o co isol concen a ion (nmol/L). Ma e nal sel - epo ed s ess was measu ed in DYAD-M a mul iple ime poin s using he Pe cei ed S ess Scale.19 20 Mo he s we e in e iewed ega ding dep es- si e symp oms a 6 mon hs pos pa um in DYAD-M using a locally alida ed adap a ion o he Sel -Re- po ing Ques ionnai e and in DYAD-G wi h he Edin- bu gh Pos -na al Dep ession Scale.21 In DYAD-M, a 6 mon hs pos pa um, ma e nal cogni ion was assessed using digi span o wa d and backwa d, e bal luency, men al o a ion and unc ional heal h li e acy es s.22 Child en we e isi ed weekly o mo bidi y su eil- lance. A hese isi s, ca egi e s we e asked whe he he child expe ienced any illness symp oms, including e e , dia hoea, omi ing, cough, nasal discha ge, espi a o y dis ess o poo appe i e du ing he pas se en days and/o da a collec o s measu ed he child’s au icula empe a u e. Longi udinal p e alence and/ o incidence o dia hoea, e e , mala ia, acu e espi- a o y in ec ion and/o poo appe i e we e calcula ed (online supplemen a y able 1). In DOSE and DYAD-M, physical ac i i y a age 18 mon hs was measu ed o e 1 week wi h he hip-wo n Ac iG aph GT3X+accele om- e e (Pensacola, Flo ida, USA).23 In an eeding p ac ices we e assessed a mul iple ime poin s h ough quali a i e 24 hou s and/o 7-day die a y ecall ques ionnai es.24 In all ou ials, de el- opmen al s imula ion was measu ed a age 18 mon hs using he Family Ca e Indica o s in e iew.25 The mo he was in e iewed wi h ega d o he a ie y o play ma e ials and ac i i ies ha adul s used o engage wi h he child in he pas h ee days. In DYAD-G and DYAD-M, ges a ional age a en olmen was mainly de e mined by ul asound and his was used o calcula e ges a ional age a bi h. In DYAD-G, in an weigh and leng h we e measu ed wi hin 48 hou s o bi h o be ween 3 and 14 days a e bi h o 87 (9.4%) when he o me was no possible. In DYAD-M, in an weigh and leng h we e measu ed wi hin 6 weeks o bi h. We es ima ed leng h and weigh a bi h based on LAZ and WAZ measu ed wi hin 6 weeks o bi h, on 30 Ap il 2019 by gues . P o ec ed by copy igh .h p://gh.bmj.com/BMJ Glob Heal h: i s published as 10.1136/bmjgh-2018-001155 on 13 Janua y 2019. Downloaded om P adoEL, e al. BMJ Glob Heal h 2019;4:e001155. doi:10.1136/bmjgh-2018-001155 5 BMJ Global Heal h Figu e 3 Media ion analysis. assuming ha LAZ and WAZ did no change om bi h o he ime o measu emen . Leng h and weigh o ges a ional age a bi h z-sco es we e hen calcula ed based on he INTERGROWTH-21s s anda ds.26 In all ou coho s, leng h was measu ed a age 18 mon hs. All leng h measu emen s we e conduc ed o he nea es 1 mm by eams o wo ained and s anda dised an h o- pome is s using leng h boa ds. Analysis We examined he dis ibu ion o each independen a iable sepa a ely by coho . We log- ans o med skewed a iables and winso ised ou lie s o he 1s and 99 h pe cen ile. I ans o ma ion did no esul in a no mal dis ibu ion, we c ea ed a bina y a iable. All con inuous a iables we e s anda dised o SD uni s by sub ac ing he mean and di iding by SD. We pe o med explo a o y media ion analyses acco ding o he ollowing s eps. We e e o igu e 3 o desc ibe each s ep in he media ion analysis. S ep 1: a iable selec ion The i s condi ion o inclusion in ou media ion analysis was ha X is associa ed wi h Y, ep esen ed by c in igu e 3, so we examined independen associ- a ions be ween each p edic o and 18-mon h LAZ and d opped any ha we e no associa ed a p<0.05. The second condi ion was ha X is associa ed wi h M, ep e- sen ed by a. The hi d condi ion was ha M is associ- a ed wi h Y, ep esen ed by b. In his i s s ep, we es ed he signi icance o b and c. Second, we educed he numbe o a iables measu ing he same cons uc by elimina ing a i- ables ha we e collinea . I wo a iables we e highly collinea ( >0.6), we d opped he one ha was less s ongly associa ed wi h 18-mon h LAZ. Thi d, we examined ou mul i a ia e models wi h each ca e- go y o ac o s oge he p edic ing 18-mon h LAZ and d opped any ha we e no associa ed a p<0.05. I , a his s ep, no a iable was signi ican ly associa ed when con olling o he o he s, we e ained he one ha was mos s ongly associa ed wi h 18-mon h LAZ. The ou mul i a ia e models we e (1) all en i onmen al ac o s oge he , (2) all ma e nal ac o s oge he , excluding ma e nal heigh , (3) all ca egi ing ac o s oge he and (4) all child ac o s oge he , excluding ges a ional age a bi h and LGAZ a bi h. S ep 2: pa h selec ion Fo each pa hway in igu e 1, we examined he asso- cia ion be ween each pai o a iables on he pa hway o de e mine which a iables we e po en ial media o s. In his second s ep, we es ed he signi icance o a in igu e 3 and d opped any pa hways o which p>0.05. In igu e 1, unidi ec ional a ows ep esen pa hways es ed. Bidi ec ional a ows ep esen associa ions ha we e checked o collinea i y, bu we e no o he wise modelled in he pa h analysis. All analyses up o his poin we e conduc ed using SAS V.9.4 (SAS Ins i u e). Nex , o each independen a iable wi h po en ial media o s, we es ed he media ion model using S a a V.14.1 (S a aCo p) bina y media ion p og am. We an he mul iple media ion model including all po en ial media o s oge he , a he han es ing each media o one by one in sepa a e models. In he inal pa h model, we included all pa hways o which he indi ec asso- cia ion o X wi h Y h ough M was signi ican . I he in e ac ion be ween X and M was signi ican a p<0.05, we s a i ied he sample a he median o he inde- penden a iable and es ed he indi ec associa ion o X h ough M a bo h high and low le els o X. I he indi ec associa ion was signi ican a bo h high and low le els o X, hen we e ained he pa hway in he model; o he wise, we emo ed ha pa hway. Finally, we an he inal pa h model using he sem command in S a a wi h he mlm op ion o es ima e he model on he ull da a se using maximum likelihood es ima ion o missing alues. All models con olled o h ee co a ia es: andomly assigned ial g oup (LNS s no LNS), child sex and child age a 18-mon h LAZ assessmen . In he inal models, we co ec ed p alues o mul iple compa isons using he Benjami- ni-Hochbe g co ec ion,27 which is ecommended o con olling he alse disco e y a e in s uc u al equa- ion models.28 We applied he co ec ion sepa a ely o each model (ie, o each coho ). I any pa hway was no signi ican a co ec ed p<0.05, hen we did no d aw ha pa hway in he pa h diag am. Fo objec i e 3, we examined he 16 a iables ha we e a ailable in all ou coho s: household asse index, household ood insecu i y access index, ma e nal and pa e nal educa ion, household wa e and oile , ma e nal age, heigh , and BMI, child dia - hoea and e e p e alence, child 6-mon h o 9-mon h on 30 Ap il 2019 by gues . P o ec ed by copy igh .h p://gh.bmj.com/BMJ Glob Heal h: i s published as 10.1136/bmjgh-2018-001155 on 13 Janua y 2019. Downloaded om 6P adoEL, e al. BMJ Glob Heal h 2019;4:e001155. doi:10.1136/bmjgh-2018-001155 BMJ Global Heal h Hb and ZPP, mean die a y di e si y ac oss ime poin s, a ie y o play ma e ials and ac i i ies wi h ca egi e s a age 18 mon hs. We epo he R2 in he models wi h hese 16 p edic o s in each coho sepa a ely and in he pooled model o de e mine whe he he pooled anal- ysis accoun ed o mo e a iance in 18-mon h LAZ han he wi hin-coho models. esulTs Summa y s a is ics o all independen a iables a e p esen ed in online supplemen a y able 2. The a iable selec ion esul s a e shown in online supplemen a y able 3-6. The pa hway selec ion esul s a e shown in online supplemen a y able 7-10. A age 18 mon hs, he mean (SD) LAZ in each coho was −0.8 (1.0) in DYAD-G, −1.7 (1.1) in DYAD-M, −1.9 (1.1) in DOSE and −1.5 (1.1) in ZINC. The pe cen age o child en who we e s un ed (LAZ < −2) a age 18 mon hs in each coho was 12% in DYAD-G, 31% in DYAD-M, 43% in DOSE and 32% in ZINC. Among en i onmen al a iables, 18-mon h LAZ was posi i ely associa ed wi h highe household asse s (DYAD-G), household being loca ed close o he ma ke (DOSE), highe ma e nal educa ion (DOSE, ZINC), highe pa e nal educa ion (DYAD-M) and an imp o ed wa e sou ce (DOSE, ZINC). O he ma e nal a iables, we obse ed highe LAZ among child en o mo he s who we e alle in all ou coho s and among child en o mo he s who had highe BMI in h ee coho s (DYAD-M, DOSE, ZINC). In DYAD-G, highe LAZ was associa ed wi h highe ma e nal Hb a ≤20 weeks ges a ion. Among he ca egi ing a iables, we obse ed highe LAZ among child en who had g ea e die a y di e si y (a 15 mon hs in DOSE and a 9 mon hs in ZINC), highe 18-mon h a ie y o play ma e ials (DYAD-M, ZINC) o highe 18-mon h ac i i ies wi h ca egi e s (DYAD-M). O he child a iables, highe LAZ was associa ed wi h highe Hb a 6 (DYAD-M, DOSE) o 9 mon hs (ZINC), g ea e posi i e change in Hb om 9 o 18 mon hs (ZINC), lowe 18-mon h AGP (DYAD-M), lowe dia - hoea incidence om 6 (DOSE) o 9 (ZINC) o 18 mon hs, lowe p e alence o poo appe i e om 6 o 18 mon hs (DYAD-G) and highe ac i i y le els as measu ed by accele ome e mean ec o magni ude a 18 mon hs (DOSE). In he wo coho s en olled be o e bi h, bo h ges a ional age a bi h and LGAZ a bi h we e s ongly associa ed wi h 18-mon h LAZ. All coe icien s in he inal models a e p esen ed in igu e 4 and online supple- men a y able 11 . Rega ding pa hway (1), examina ion o gene ic e ec s o in e gene a ional e ec s o ma e nal g ow h du ing ea ly li e, e lec ed by ma e nal adul heigh , on child 18-mon h LAZ consis en ly showed ha his was la gely a di ec associa ion (66%–99% o he associa ion be ween ma e nal heigh and LAZ ac oss he ou coho s) a he han ia media ion by o he ac o s. In he wo coho s in which leng h a bi h was measu ed, LGAZ a bi h media ed a subs an ial p opo ion o his asso- cia ion (34% in DYAD-G, 30% in DYAD-M). A e y small p opo ion o his associa ion was media ed by child Hb a 6 mon hs (2% in DOSE) o die a y di e si y in in an eeding (1% in ZINC). Fo pa hway (2), he ex en o which socioeconomic dispa i ies we e media ed by o he ac o s a ied ac oss coho s and independen a iables, om 0% o 43%. In ZINC, he associa ion o ma e nal educa ion wi h 18-mon h LAZ was media ed by he child’s a ie y o play ma e ials (43%) and he associa ion o imp o ed wa e sou ce wi h LAZ was no media ed by any o he a i- ables. In DYAD-G, he associa ion be ween household asse s and child LAZ was pa ly media ed by ma e nal lowe Hb concen a ion in ea ly p egnancy (17%) and ges a ional age a bi h (11%). In DOSE, h ee en i- onmen al a iables we e associa ed wi h LAZ: dis ance o he nea es ma ke , ma e nal educa ion and an imp o ed wa e sou ce. The di ec p opo ion o hese associa ions anged om 74% o 93%. The associa ion o dis ance o he nea es ma ke wi h LAZ was pa ly media ed by ma e nal BMI (11%) and child dia hoea incidence (5%). The associa ion o ma e nal educa ion wi h 18-mon h LAZ was pa ly media ed by ma e nal BMI (7%). The associa ion o imp o ed wa e sou ce wi h child LAZ was pa ly media ed by child physical ac i i y (26%). In DYAD-M, while he associa ion o pa e nal educa ion wi h LAZ was e ained in he a iable selec ion p ocess, i was no signi ican in he inal model and was no media ed by any a iables. Rega ding pa hway (3), he ex en o which associa ions o ma e nal heal h and nu i ional s a us wi h child LAZ we e media ed e sus di ec e ec s also a ied by coho and independen a iable (0%–56%). In DYAD-M, he associa ion be ween ma e nal BMI and 18-mon h LAZ was la gely media ed by LGAZ a bi h (62%). In he wo coho s in which LGAZ a bi h was no a ailable, he associa ion o ma e nal BMI wi h LAZ was la gely a di ec associa ion (92%–94%), wi h a small p opo ion medi- a ed by he child’s a ie y o play ma e ials (8% in ZINC) o child Hb concen a ion a 6 mon hs (6% in DOSE). In DYAD-G, he associa ion o ma e nal Hb in ea ly p eg- nancy wi h 18-mon h LAZ was no media ed by any o he a iables. Associa ions o die a y di e si y in in an eeding a 9 mon hs in ZINC and 15 mon hs in DOSE wi h 18-mon h LAZ (pa hway 4.1) we e pa ly media ed by child dia - hoea incidence. Ca egi e s who epo ed highe die a y di e si y in in an eeding also epo ed highe dia hoea incidence in he child. The associa ion o die a y di e si y wi h 18-mon h LAZ was also posi i e and he associa ion o dia hoea incidence wi h 18-mon h LAZ was nega i e; he e o e, he p opo ion o he associa ion be ween die a y di e si y and child LAZ ha was media ed by dia - hoea incidence was nega i e (−5% in ZINC and −28% in DOSE). Resul s om bo h ials ha measu ed ges a ional age and leng h a bi h showed ha hese we e di ec on 30 Ap il 2019 by gues . P o ec ed by copy igh .h p://gh.bmj.com/BMJ Glob Heal h: i s published as 10.1136/bmjgh-2018-001155 on 13 Janua y 2019. Downloaded om P adoEL, e al. BMJ Glob Heal h 2019;4:e001155. doi:10.1136/bmjgh-2018-001155 7 BMJ Global Heal h Figu e 4 Pa h diag am o ac o s associa ed wi h18-mon h LAZ in ou iLiNS coho s. 1Measu ed in one coho . 2Measu ed in wo coho s. 3Measu ed in h ee coho s. AGP, alpha-1-acid glycop o ein; BMI, body mass index; Hb, haemoglobin; iLiNS In e na ional Lipid-Based Nu ien Supplemen s; LAZ, leng h- o -age z-sco e. associa ions, unmedia ed by any o he child ac o s (pa hway 5.2). The ollowing esul s om he ZINC coho suppo ed pa hway (6.1): child en wi h lowe Hb a age 9 mon hs we e p o ided wi h lowe a ie y o play ma e ials a 18 mon hs, media ing 12% o he associa ion wi h 18-mon h LAZ. Simila o die a y di e si y, a ie y o play ma e ials was posi i ely associa ed wi h dia hoea incidence, he e o e his media ed −8% o he nega i e associa ion be ween dia hoea incidence and child LAZ. The inal models accoun ed o 34% o he a iance in 18-mon h LAZ in DYAD-G, 37% in DYAD-M, 17% in DOSE and 20% in ZINC. O e all, ma e nal and child ac o s showed s onge associa ions wi h 18-mon h LAZ compa ed wi h en i onmen al and ca egi ing ac o s, e en when conside ing bo h indi ec and di ec e ec s ( igu e 5). Fo objec i e 3, he 16 a iables common ac oss all ou coho s accoun ed o he ollowing pe cen age o a iance in 18-mon h LAZ: 16% in DYAD-G, 21% in DYAD-M, 15% in DOSE, 19% in ZINC, and 25% in he pooled analysis. O hese 16 a iables, ma e nal heigh accoun ed o he la ges p opo ion o a iance in 18-mon h LAZ: 11% in DYAD-G, 14% in DYAD-M, 10% in DOSE, 11% in ZINC and 9% in he pooled analysis. The e o e, excluding ma e nal heigh , en i onmen al, ma e nal, ca egi ing and child ac o s accoun ed o only 5%–8% o a iance in LAZ in he indi idual coho s and 16% in he pooled analysis. on 30 Ap il 2019 by gues . P o ec ed by copy igh .h p://gh.bmj.com/BMJ Glob Heal h: i s published as 10.1136/bmjgh-2018-001155 on 13 Janua y 2019. Downloaded om 8P adoEL, e al. BMJ Glob Heal h 2019;4:e001155. doi:10.1136/bmjgh-2018-001155 BMJ Global Heal h Figu e 5 Coe icien s o di ec and indi ec e ec s on 18-mon h LAZ in he inal s uc u al equa ion models. BMI, body mass index; Hb, haemoglobin; LAZ, leng h- o -age z-sco e. dIsCussIon In ou p ospec i e coho s o child en, we conduc ed pa h analyses o en i onmen al, ma e nal, ca egi ing and child ac o s hypo hesised o be associa ed wi h 18-mon h linea g ow h s a us. Ou o 42 indica o s examined, only 2 indica o s eme ged as consis en p edic o s o 18-mon h LAZ in 3–4 coho s: ma e nal heigh and ma e nal BMI. Two addi ional ac o s, which we e only measu ed in he wo p egnancy coho s, we e signi ican ly associa ed wi h 18-mon h LAZ in bo h coho s in which hey we e meas- u ed: LGAZ a bi h and ges a ional age a bi h. In he wo coho s en olled a e bi h, ou ac o s we e signi - ican ly associa ed wi h 18-mon h LAZ in bo h coho s: imp o ed household wa e sou ce, die a y di e si y in in an eeding, childhood dia hoea incidence and 6-mon h o 9-mon h Hb. O e all, ma e nal and child ac o s showed s onge and mo e consis en associa- ions wi h child LAZ han en i onmen al and ca egi ing ac o s, e en when conside ing bo h indi ec and di ec e ec s. While speci ic pa hways di e ed ac oss coho s, in gene al di ec associa ions we e s onge and mo e p e - alen han indi ec associa ions. The s onges e idence o media ion was he inding ha 62% o he associa ion o ma e nal BMI wi h 18-mon h LAZ was media ed by LGAZ a bi h in DYAD-M. This unde sco es he impo - ance o ma e nal nu i ional s a us o e al g ow h, wi h las ing consequences o g ow h s a us la e in childhood. LGAZ a bi h also media ed a subs an ial p opo ion o he associa ion o ma e nal heigh wi h 18-mon h LAZ (30%–34%) in bo h coho s in which i was measu ed (DYAD-M and DYAD-G). This pa hway is likely o e lec gene ic in luences, a leas in pa , and he e o e may be only pa ly modi iable. In he wo coho s in which LAZ a bi h was no measu ed (DOSE and ZINC), he associ- a ion o ma e nal heigh wi h 18-mon h LAZ was a di ec associa ion. O he eigh key isk ac o s closely associa ed wi h linea g ow h al e ing ha we ound, h ee we e consis en ly iden i ied in p e ious s udies ha examined he ela i e con ibu ion o isk ac o s o s un ing p e alence glob- ally: e al g ow h es ic ion, dia hoea and low die a y di e si y. Danaei e al used popula ion su eys o es ima e cases o s un ing a ibu able o 18 isk ac o s in 137 de eloping coun ies, including indica o s o ma e nal nu i ion and in ec ion, eenage mo he hood and sho bi h in e als, e al g ow h es ic ion and p e e m bi h, child nu i ion and in ec ion, and en i onmen al ac o s. They ound ha a la ge numbe o s un ing cases we e a ibu able o being bo n small o ges a ional age a e m (10.8 million, 24% o cases), ollowed by poo sani a ion (7.2 million, 16% o cases) and dia hoea (5.8 million, 13% o cases).11 In a sys ema ic e iew, Mosi es e al calcu- la ed he popula ion-a ibu able ac ion o s un ing o i e ca ego ies o isk ac o s: low bi h weigh , insu i- cien die , en e ic dys unc ion, in ec ion and oxin expo- su e. They es ima ed ha 25% o global s un ing could be a ibu ed o child dia hoea incidence. In A ica, he la ges numbe o s un ing cases was a ibu able o low die a y di e si y, low eeding equency, o bo h (30%).12 on 30 Ap il 2019 by gues . P o ec ed by copy igh .h p://gh.bmj.com/BMJ Glob Heal h: i s published as 10.1136/bmjgh-2018-001155 on 13 Janua y 2019. Downloaded om P adoEL, e al. BMJ Glob Heal h 2019;4:e001155. doi:10.1136/bmjgh-2018-001155 9 BMJ Global Heal h Ano he s udy using popula ion-based su eys om 54 coun ies es ima ed ha mo he s in he lowes ca ego y o heigh (<145 cm) we e 2.1 imes mo e likely o ha e s un ed child en han mo he s in he highes ca ego y (≥160 cm).8 The key isk ac o s o linea g ow h al e ing ha we iden i ied we e simila o hose iden i ied in hese s udies despi e se e al di e ences in me hodology, including p ospec i e coho s a he han c oss-sec ional design, examina ion o bo h di ec and indi ec associ- a ions h ough pa h analysis, and examina ion o LAZ as a con inuous a iable a he han s un ing cases. By analysing associa ions wi h LAZ, we we e able o es ima e he di e ence in LAZ wi h each SD change in p edic o a iables. Each SD di e ence in ma e nal heigh (5–6 cm) was associa ed wi h 0.25 di e ence in LGAZ a bi h (0.5 cm) and a 0.2–0.3 di e ence in child 18-mon h LAZ (0.5–0.8 cm). Each SD di e ence in LGAZ a bi h (1.9 cm) was associa ed wi h abou a 0.5 di e ence in child 18-mon h LAZ (1.4 cm). Each SD di e ence in ges a- ional age a bi h (1.8 weeks) was associa ed wi h a 0.13–0.17 di e ence in 18-mon h LAZ (0.3–0.5 cm). The coe icien sizes o he o he key isk ac o s (ma e nal BMI, imp o ed wa e sou ce, die a y di e si y, child dia - hoea and Hb) we e smalle in magni ude (0.06–0.12). Despi e he examina ion o 42 en i onmen al, ma e nal, ca egi ing and child ac o s measu ed longi- udinally h oughou mos o he i s 1000 days, ou models accoun ed o a ela i ely small p opo ion o he a iance in 18-mon h LAZ (17%–37%), and a la ge p opo ion o ha was due o ma e nal heigh (10%– 14%). Fac o s ha we e no associa ed wi h LAZ we e indica o s o ma e nal i on s a us, illness and in lamma- ion du ing p egnancy, ma e nal s ess, dep ession and cogni ion, exclusi e b eas eeding du ing 6 mon hs pos pa um, in an eeding equency and child e e , mala ia and acu e espi a o y in ec ions. Va iance ha emained unaccoun ed o could be due o unmeasu ed ac o s, such as en e ic dys unc ion, asymp oma ic in ec ions, mic obiome composi ion, ma e nal HIV s a us o aspec s o he die ha we e no assessed. I is also possible ha popula ion-le el ac o s, such as communi y sani a ion o access o heal hca e, may be s onge d i e s o g ow h s un ing compa ed wi h indi idual o household-le el ac o s. Ou inding ha he pooled analysis accoun ed o mo e a iance in LAZ (25%) compa ed wi h sepa a e anal- yses by coho (11%–21%) also suppo s his a gumen , al hough his di e ence is no e y la ge. Excluding he a iance accoun ed o by ma e nal heigh , he pooled analysis accoun ed o wo o h ee imes he a iance in LAZ (16%) compa ed wi h he sepa a e coho models (5%–8%). This may be due o he lack o inclusion o a coho expe iencing heal hy g ow h. All ou iLiNS coho s expe ienced g ow h al e ing, al hough he coho s in Malawi and Bu kina Faso had al e ed 1.5–1.9 SD below he WHO median on a e age, compa ed wi h he Ghanaian child en whose mean LAZ was only 0.8 SD below he WHO median by age 18 mon hs. Ano he po en ial eason is ha we did no measu e communi- y-le el o social ac o s ha may con ibu e o child en’s g ow h al e ing a a popula ion le el. Fo example, se e al s udies ha e ound ha communi y-le el sani a- ion co e age p edic s child g ow h mo e s ongly han household-le el sani a ion.29 30 S eng hs o he s udy we e he la ge numbe o chil- d en in each coho , he la ge numbe o a iables examined and he a ailabili y o da a om ou coho s o child en in h ee A ican coun ies, wi h many a i- ables ha monised ac oss coho s, enabling he examina- ion o he same ac o s and pa hways ac oss con ex s. Ano he s eng h was ha he coho s we e en olled ei he du ing p egnancy o a 6 o 9 mon hs pos pa um and ollowed p ospec i ely h ough 18 mon hs pos pa um, allowing hese isk ac o s o be examined h oughou a la ge po ion o he i s 1000 days a e concep ion. The p ospec i e design allowed associa ions in he pa h model o be d awn om a iables measu ed a ea lie ime poin s (eg, p egnancy, bi h) o a iables measu ed a la e ime poin s (eg, 6 mon hs, 18 mon hs), educing he isk o e e se causali y. Howe e , his does no p e en he possibili y o esidual con ounding, ha is, unmeasu ed a iables accoun ing o obse ed asso- cia ions, he e o e we a e no able o es ablish causali y. Ano he limi a ion was ha we we e no able o explo e ce ain isk ac o s, such as asymp oma ic in ec ions o en e ic dys unc ion. We epo ed mul iple compa isons; he e o e, may ha e ound some alse posi i e associa- ions due o chance. Howe e , we educed his possibili y by epo ing p alues co ec ed o mul iple compa isons and by basing ou p ima y conclusions on indings ha we e eplica ed in mo e han one coho . These samples may no be ep esen a i e o hei popula ions due o hei pa icipa ion in he nu i ional supplemen a ion ials. Howe e , any bias in oduced by he supplemen a- ion would be expec ed o dec ease a he han inc ease he s eng h o he associa ions.31 The e o e, ou conclu- sions ega ding consis en signi ican associa ions would emain he same. ConClusIon Ou indings may help o in o m he design o in e en- ions o pu sue Uni ed Na ions’ Sus ainable De elopmen Goal 2, o end hunge and achie e imp o ed nu i ion by 2030, wi h he educ ion o s un ing p e alence as a key indica o o p og ess owa ds his goal. The e idence we p esen shows ha child linea g ow h s a us is closely linked o highe ma e nal heigh and BMI, e al g ow h, ull- e m bi h, imp o ed household wa e supply, highe child die a y di e si y, Hb du ing in ancy and educed childhood dia hoea incidence, bu we did no ind associa ions wi h ma e nal i on s a us, illness and in lam- ma ion du ing p egnancy (including ma e nal HIV in ec ion in he one coho in which i was measu ed), ma e nal s ess and dep ession, exclusi e b eas eeding on 30 Ap il 2019 by gues . P o ec ed by copy igh .h p://gh.bmj.com/BMJ Glob Heal h: i s published as 10.1136/bmjgh-2018-001155 on 13 Janua y 2019. Downloaded om