Umeå University's logo

umu.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Machine learning techniques to model child low height-for-age in the northern province of Rwanda: the role of climatological and environmental factors and their interactions
Umeå University, Faculty of Medicine, Department of Clinical Sciences, Paediatrics. University of Rwanda, College of Medicine and Health Sciences, School of Public Health, Kigali, Rwanda.
Department of Physical Geography and Ecosystem Science, Centre for Geographical Information Systems, Lund University, Lund, Sweden; University of Rwanda, College of Sciences and Technology, Centre for Geographic Information Sciences, P.O. Box 4285, Kigali, Rwanda.
Umeå University, Faculty of Medicine, Department of Clinical Sciences, Paediatrics.ORCID iD: 0000-0001-6328-1098
Department of Physical Geography and Ecosystem Science, Centre for Geographical Information Systems, Lund University, Lund, Sweden.
2026 (English)In: Clinical Epidemiology and Global Health, E-ISSN 2213-3984, Vol. 37, article id 102284Article in journal (Refereed) Published
Abstract [en]

Objective: Childhood stunting is a significant health issue in Rwanda, particularly within the Northern Province. While demographic and socio-economic factors have been more extensively studied, the impact of environmental and climatic factors on stunting prevalence has received less attention. This study aimed to determine if these factors could be used to better predict localized variations in height-for-age z-scores (HAZ).

Study design: A population-based, cross-sectional study.

Methods: Data were collected on child and maternal characteristics, household socioeconomic status, climate, and environmental predictors. An eXtreme Gradient Boosting (XGBoost) algorithm was used, complemented by GeoShapley for spatial analyses, to explain the spatial variability between low height-for-age and its risk factors.

Results: The model performed well, with the coefficient of determination (R2) value of 0.83, the root mean standardized error (RMSE) of 0.13, and the mean absolute error (MAE) of 0.10. Key predictors of HAZ included rainfall, childcare practices, food insecurity, elevation, and soil fertility. Considering the location feature, environmental and climatic factors significantly contributed to the spatial variability in HAZ.

Conclusion: Many environmental, climatological, and socio-economic factors emerge as predictors for HAZ variability. It is essential to consider their complexity for comprehensive interventions targeting childhood stunting in Rwanda and similar settings.

Place, publisher, year, edition, pages
Elsevier, 2026. Vol. 37, article id 102284
National Category
Epidemiology Public Health, Global Health and Social Medicine
Identifiers
URN: urn:nbn:se:umu:diva-249005DOI: 10.1016/j.cegh.2025.102284ISI: 001665135900001Scopus ID: 2-s2.0-105027443855OAI: oai:DiVA.org:umu-249005DiVA, id: diva2:2034733
Funder
Sida - Swedish International Development Cooperation Agency, 11277
Note

Available from: 2026-02-02 Created: 2026-02-02 Last updated: 2026-05-04Bibliographically approved
In thesis
1. Integrating spatial, classical and social epidemiology, machine learning, and qualitative approaches to understand childhood stunting in the Northern Province of Rwanda
Open this publication in new window or tab >>Integrating spatial, classical and social epidemiology, machine learning, and qualitative approaches to understand childhood stunting in the Northern Province of Rwanda
2026 (English)Doctoral thesis, comprehensive summary (Other academic)
Alternative title[sv]
Hämmad längdtillväxt hos barn i Norra provinsen i Rwanda : integrering av spatial och klassisk epidemiologi, maskininlärning och kvalitativa perspektiv
Abstract [en]

Background: In Rwanda, childhood stunting has declined over the last two decades, yet the Northern Province remains a high-burden area, an apparent contradiction given its fertile soils and robust agricultural activity. This thesis sets out to understand that disconnection. Rather than relying onnational averages or single-discipline explanations, it asks where stunting concentrates, whichfactors are most influential and modifiable, and what the key caregivers' experiences are behindthe fact that available resources and programs do not translate into better child growth.

Methods: A population-based cross-sectional study was conducted among 601 households with children aged 1–36 months across all five districts of Northern Province. Maternal, child, and household characteristics were collected using structured questionnaires. Climatic, environmental, and agroecological data were collected from remote sources. Parents were separately reached through focus group discussions. The design integrated five components: (1) spatial statistics to test randomness and identify sector-level hotspots; (2) machine learning to quantify the roles of climatic and environmental features alongside caregiving proxies; (3) multivariable logistic regression to determine modifiable factors; (4) wealth-inequality analysis to identify contributors to pro-poor inequalities; and (5) reflexive thematic analysis of gender-segregated focus groups to explore the caregivers’ experiences and understanding of childhood stunting risk in the studyarea.

Results: The stunting prevalence was 27.1% among children 1-36 months. Its distribution was non-random and clustered, with statistically significant hotspots in three districts of the province. The machine-learning model indicated that location, in combination with climatic and environmental features, explained a large share of the variability in child low height-for-age. The multivariable regression analysis identified several modifiable household-level risk factors, including male children, maternal stature, inadequate breastfeeding, female-headed household, and inadequate handwashing facilities. Inequality analysis showed that the burden of stunting falls disproportionately on households within the poorest quintiles, with partners involved in non-skilled/unemployment and food insecurity. Exploring parents’ experiences and understanding ofchildhood stunting, four main themes emerged: power struggles and broken care underminingchild nutrition, vulnerabilities across the first 1,000 days shaping stunting trajectories, the paradox of plenty where livestock and food don’t nourish children, and fragile substitute careleaving children behind.

Conclusion: Childhood stunting in the Northern Province of Rwanda remains complex with spatial clustering, many hidden and localised climatic and environmental pressures intersecting with household behaviours, caregiving constraints, and poverty, preventing available resources from reaching the child’s plate. The thesis lists several possible venues to move the needle on stunting prevention in the study area.

Abstract [sv]

Bakgrund: I Rwanda har andelen barn med hämmad längdtillväxt minskat under de senaste två decennierna. Den Norra provinsen är dock ett område med fortsatt hög förekomst, en uppenbar motsägelse med tanke på dess bördiga jordar och robusta jordbruksverksamhet. Denna avhandling syftar till att förstå denna skillnad. Snarare än att förlita sig på nationella genomsnitt eller förklaringar från enstaka vetenskapsområden tar den ett multidisciplinärt angrepp och frågar sig var förekomsten av hämmad längdtillväxt är koncentrerad, vilka faktorer som är mest inflytelserika och modifierbara, och vilka de viktigaste vårdgivarupplevelserna är som ligger till grund för det faktum att tillgängliga resurser inte leder till bättre tillväxt i barndomen. 

Metoder: Avhandlingen bygger på en populationsbaserad tvärsnittsstudie bland 601 hushåll med barn i åldrarna 1–36 månader i de fem distrikten i Norra provinsen. Mödra-, barn- och hushållskarakteristika samlades in med hjälp av strukturerade frågeformulär. Klimat-, miljö- och ekologiska data samlades in från externa källor. Föräldrar inbjöds separat till fokusgruppsdiskussioner. Designen integrerade fem komponenter: (1) rumslig statistik för att testa slumpmässighet och identifiera områden med ökad förekomst på sektornivå; (2) maskininlärning för att kvantifiera klimat- och miljöfaktorernas roll tillsammans med data på omständigheter som speglar omsorg om familjens barn; (3) multivariabel logistisk regression för att identifiera modifierbara faktorer; (4) analys av skillnader i ekonomiska resurser för att identifiera bidragande faktorer till omständigheter som missgynnar de fattiga; och (5) reflexiv tematisk analys av könsuppdelade fokusgrupper för att utforska vårdgivarnas erfarenheter och förståelse av risken för hämmad tillväxt i barndomen i studieområdet. 

Resultat: Förekomsten av hämmad tillväxt var 27,1 % bland barn i åldrarna 1–36 månader. Dess fördelning var icke-slumpmässig, utan ansamlad i vissa områden, särskilt i tre distrikt i provinsen. Maskininlärningsmodellen indikerade att plats, i kombination med klimat- och miljöfaktorer, förklarade en stor del av variationen i barns längd i förhållande till ålder. Den multivariata regressionsanalysen identifierade flera modifierbara riskfaktorer på hushållsnivå, inklusive manligt kön, moderns längd, otillräcklig amning, ensamstående mödrar och otillräckliga möjligheter att tvätta händerna. Analysen av ekonomiska skillnader visade att andelen hämmad tillväxt faller oproportionerligt mycket på de fattigaste hushållen, med partners som är arbetslösa eller involverade i okvalificerade arbeten och i hushåll som saknar säker tillgång på mat. En undersökning av föräldrars erfarenheter och förståelser av hämmad tillväxt i barndomen visade fyra huvudteman: maktkamper och sviktande omsorg som undergräver barns näringsintag, sårbarheter under de första 1 000 dagarna som formar hämmad tillväxt, överflödsparadoxen där tillgång till boskap och mat inte ger näring till barnen, och bräckliga alternativ till barnomsorg. 

Slutsats: Hämmad tillväxt hos barn i Rwandas norra provins är en komplex fråga med rumslig anhopning, flera dolda och lokala klimat- och miljöpåverkansfaktorer som skär sig mot hushållsbeteenden, fattigdom samt begränsningar inom barnomsorg, vilket hindrar tillgängliga resurser från att nå barnets tallrik. Avhandlingen listar alternativa sätt att förebygga förekomsten av hämmad tillväxt i studieområdet.

Place, publisher, year, edition, pages
Umeå: Umeå University, 2026. p. 64
Series
Umeå University medical dissertations, ISSN 0346-6612 ; 2425
Keywords
Stunting, undernutrition, spatital, inequalities, machine learning, factors, qualitative, Rwanda
National Category
Public Health, Global Health and Social Medicine Pediatrics
Research subject
Nutrition; Public health; Epidemiology
Identifiers
urn:nbn:se:umu:diva-252777 (URN)978-91-6850-025-6 (ISBN)978-91-6850-026-3 (ISBN)
Public defence
2026-06-12, ULED.A.310 - Triple Helix, Universitetsledningshuset, Umeå University, Umeå, 09:00 (English)
Opponent
Supervisors
Funder
Sida - Swedish International Development Cooperation Agency, 11277
Note

Link to participate via Zoom: https://umu.zoom.us/j/67871831305

Available from: 2026-05-05 Created: 2026-05-04 Last updated: 2026-05-06Bibliographically approved

Open Access in DiVA

fulltext(10602 kB)50 downloads
File information
File name FULLTEXT01.pdfFile size 10602 kBChecksum SHA-512
4419dbbe3589b772ef225f2307bd97c1a96ce22d2719007b3d47f950b4cd73875ec9cde21697e29d7fa1189a1f68f41ebb50d5378112dc2c447dfafd0a9daba4
Type fulltextMimetype application/pdf

Other links

Publisher's full textScopus

Authority records

Ndagijimana, AlbertLind, Torbjörn

Search in DiVA

By author/editor
Ndagijimana, AlbertLind, Torbjörn
By organisation
Paediatrics
EpidemiologyPublic Health, Global Health and Social Medicine

Search outside of DiVA

GoogleGoogle Scholar
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

doi
urn-nbn

Altmetric score

doi
urn-nbn
Total: 3256 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf