Pooling data (or NOT): an empirical examination of the relationship between maternal education and child morbidity in Central Africa using Demographic and Health Surveys
Auteur(s): Tsala Dimbuene Zacharie, Nyirenda Elizabeth T, Kabore Philippe, Kalinga Janet, Boti Sidamo Negussie, Achaliwie FranciscaNom de la revue/Journal: BMJ Public Health
Mois: octobre
Volume: 4
Numéro: 4
Année: 2026
pages: 1-7
DOI: 10.1136/ bmjph-2026-004892
Résumé
<p>Introduction Pooling datasets is increasingly used in global health research to increase statistical power, improve precision and enable cross- country comparisons. Yet, combining data from heterogeneous settings may obscure important contextual differences and lead to conclusions that diverge from those obtained through country- specific analyses. Using maternal education and child morbidity as a case study, this paper examines how pooled vs individual datasets can yield different interpretations, highlighting the implications for research and policy. ,5 Francisca Achaliwie 6 ,3 WHAT IS ALREADY KNOWN ON THIS TOPIC ⇒ Pooling multiple datasets increases statistical power, enabling more precise estimates and greater ability to detect meaningful associations that may be missed in smaller, isolated samples. ⇒ Larger combined samples also improve the stability of model estimates, especially in subgroup analyses or geographically disaggregated settings. WHAT THIS STUDY ADDS ⇒ By pooling datasets while explicitly addressing heterogeneity and contextual policy differences, this study provides new insights to better support evidence- based decision- making tailored to diverse local policy environments. Methods We conducted a secondary analysis of Demographic and Health Surveys (DHS) from five Central African countries: the Democratic Republic of Congo, Cameroon, Republic of Congo, Gabon and Chad. The study population included children aged 0–59 months. Child morbidity outcomes—fever, diarrhoea and cough in the 2 weeks preceding the survey—were analysed using descriptive statistics and multivariate logistic regression models applied separately to pooled and country- specific datasets. Results The final analytic sample comprised 58 079 children under 5 years of age. Educational attainment varied substantially across countries, with 66.0% (96% CI 64.0% to 67.9%) of women in Chad reporting no formal education—far higher than in the other countries. Overall prevalence of cough, diarrhoea and fever was 6.6% (95% CI 6.2% to 7.1%), 19.3% (95% CI 18.6% to 19.9%) and 26.1% (95% CI 25.2% to 26.9%), respectively, with notable cross- country variation. Multivariate analyses revealed that the association between maternal education and child morbidity differed between the pooled dataset and the individual country models, demonstrating how aggregation can mask local heterogeneity. Conclusion Pooling datasets can be a powerful analytical strategy, particularly for increasing statistical power, improving the precision of estimates and enabling cross- country comparisons. Using DHS data from five Central African countries, this study demonstrates that while pooled analyses identify broad regional patterns, they can mask important country- specific heterogeneity, which is of paramount importance to devise national and local policies.</p>