The hidden bias of missing data: multiple imputation reveals geographic disparities and nutrition-WASH synergy as drivers of childhood diarrhea in Nigeria- a cross-sectional analysis of 33,924 children

OBJECTIVE: To investigate the determinants of childhood diarrhea in Nigeria, examining socio-economic, geographic, nutritional, and WASH factors, while systematically comparing Multiple Imputation by Chained Equations (MICE) versus Complete Case Analysis (CCA) to demonstrate how missing data methods bias conclusions. METHODS: A secondary cross-sectional analysis was conducted using the latest Nigerian Demographic and Health Survey (2018 NDHS) data on 33,924 under-5 children. The primary outcome was a caregiver-reported diarrhea incidence. Independent predictors were identified using multivariable logistic regression. Robustness was tested by sensitivity analyses through comparison between MICE, CCA and testing principal results under Missing Not at Random (MNAR) assumptions. Multi-dimensional risk heatmaps and correlation networks illustrated interactive risks. RESULTS: Diarrhea prevalence was 12.0%. Profound geo-economic and geographical disparities were uncovered; residing in the North-East was the strongest risk factor (aOR = 2.61), while wealth and maternal education demonstrated protective associations. There was a very high synergistic interaction between poor nutrition and WASH, with combined high-risk categories having the greatest prevalence of diarrhea (15.4%). Intake of eggs had a strong protective association (aOR = 0.87) that persisted under MNAR models. Sensitivity analyses revealed CCA-generated biased estimates, masking protective effects of education and inducing spurious risks for vitamin A and fruit consumption. CONCLUSION: Childhood diarrhea in Nigeria is driven by a complex interplay of interacting factors. Effective control requires a mix of integrated, multi-sectoral interventions involving WASH, nutrition, and poverty concurrently, focusing on high-burden zones. Methodologically, advanced management of missing data is critical to produce valid evidence.