Bayesian Survival Analysis Details Under-Five Child Mortality Patterns Across Ethiopia

A new Bayesian survival analysis of nationally representative demographic data from Ethiopia shows that child mortality remains heavily concentrated in the earliest stages of life, with 91.0 percent of under-five deaths occurring within the first twelve months. Drawing on birth-history records from the 2019 Ethiopia Mini Demographic and Health Survey, the study evaluated the timing, determinants, and geographic distribution of under-five mortality across the country to examine why child survival risks vary across demographic groups and administrative areas.

To capture the temporal dynamics of child survival, the researchers evaluated a cohort of 5,770 children born during the five years preceding the survey, among whom 342 deaths occurred before age five. The individual survival histories were restructured into a longitudinal dataset comprising 165,749 person-month observations. The investigation applied a Bayesian discrete-time survival model using a complementary log-log link function, incorporating age-specific baseline hazards, individual-level covariates, and normalized Bayesian pseudo-likelihood weighting to account for the complex survey design.

The adjusted survival models revealed distinct demographic differences in child mortality risk, particularly regarding the child’s sex and maternal age at childbirth. Female children experienced a lower mortality hazard than male children, with an adjusted hazard ratio of 0.77 and a 95 percent credible interval of 0.60 to 0.98. In addition, children born to mothers aged 20 years and older had substantially lower hazards of death compared with children born to adolescent mothers aged 15 to 19 years.

Specifically, the hazard ratios across maternal age groups demonstrated consistent advantages over adolescent motherhood after controlling for other covariates in the model. Compared with adolescent mothers, the hazard ratio was 0.46 (95 percent credible interval 0.27 to 0.78) for mothers aged 20 to 24 years, 0.39 (95 percent credible interval 0.22 to 0.69) for those aged 25 to 29 years, 0.33 (95 percent credible interval 0.17 to 0.64) for those aged 30 to 34 years, and 0.37 (95 percent credible interval 0.18 to 0.75) for those aged 35 to 39 years. Conversely, socioeconomic and demographic characteristics including maternal education, household wealth index, urban or rural residence, and birth-order categories did not demonstrate clear independent associations with under-five mortality after multivariable adjustment.

The modeling framework also investigated geographic variation by incorporating a Besag-York-Mollié (BYM2) spatial effect across Ethiopia’s 11 administrative regions, alongside random effects across 305 primary sampling units. Regional spatial effects varied across the country, with the Somali region recording the highest posterior median spatial hazard multiplier of 1.28 (95 percent credible interval 0.93 to 2.29). However, because the 95 percent credible intervals for all eleven regions encompassed 1, residual variation between primary sampling units proved more pronounced than structured regional differences.

Model performance and statistical adequacy were evaluated using the Deviance Information Criterion, the Watanabe-Akaike Information Criterion, posterior predictive checks, and prior-sensitivity analyses. These diagnostics supported the model specification while affirming that local cluster-level variation contributed more to unexplained heterogeneity than broad administrative boundaries, underscoring substantial statistical uncertainty in regional spatial estimates.

Based on these findings, the study highlights key operational priorities for public health planning, emphasizing targeted support for neonatal and infant survival as well as maternal health programs directed toward adolescent mothers. The authors caution that because the analysis relies on an observational survey design, the observed associations should not be interpreted as definitive causal effects, recommending that future resource allocation and region-sensitive surveillance be guided by continued population-based monitoring.

Subject of Research: Bayesian discrete-time survival analysis of under-five mortality in Ethiopia

Article Title: Under-five mortality in Ethiopia: a Bayesian multilevel discrete-time survival analysis of the 2019 Ethiopia Mini Demographic and Health Survey

Article References: Joche, M. C. (2026). Under-five mortality in Ethiopia: a Bayesian multilevel discrete-time survival analysis of the 2019 Ethiopia Mini Demographic and Health Survey. Archives of Public Health. https://doi.org/10.1186/s13690-026-02083-y

Image Credits: AI Generated

DOI: 10.1186/s13690-026-02083-y

Keywords: under-five mortality, Ethiopia, Bayesian survival analysis, infant mortality, child health, Demographic and Health Survey, maternal age, spatial epidemiology, Under-five, mortality, Bayesian, multilevel