PROGNOSTIC REGRESSION MODELING OF UNDER-FIVE MORTALITY: EVIDENCE FROM COMMUNITY-LEVEL DATA IN LAGOS STATE, NIGERIA

Abstract

Under-five mortality remains unevenly distributed across communities in developing economies, reflecting disparities in socioeconomic conditions, maternal characteristics, and access to basic health services. This study examines the determinants of under-five mortality across five administrative divisions of Lagos State, Nigeria, using a prognostic regression framework that aligns with the distributional properties of mortality data. The primary outcome was the number of under-five deaths experienced by a woman, while a secondary outcome captured whether any under-five death occurred. Explanatory variables included socioeconomic status, maternal education, healthcare access, nutritional status, sanitation, immunization coverage, and community of residence. Binary logistic regression was used to model the probability of experiencing any under-five death, while Gaussian and Gamma regression models were applied to mortality severity. Model selection using Akaike and Bayesian Information Criteria identified the Gamma regression model as most appropriate. Results indicate that improved socioeconomic status reduced expected under-five mortality by approximately 68%, while enhanced healthcare access and sanitation reduced mortality by about 57% and 49%, respectively. Each additional year of maternal education reduced expected mortality by about 6%, whereas poor nutritional status significantly increased risk. Marked spatial disparities were observed, with Ikorodu, Badagry, and Epe exhibiting higher mortality burdens. Overall, the findings reveal the importance of distribution-aware prognostic modeling and targeted community-specific interventions to reduce under-five mortality in Lagos State.