Network-Adjusted Empirical Bayes Estimation of Age-Specific Adult Mortality from DHS Sibling Survival Histories: Evidence from Nigeria

Abstract Background Weak civil registration coverage in Nigeria means adult mortality relies heavily on sibling survival histories (SSHs) from Demographic and Health Surveys (DHS). SSH-based estimates suffer from differential visibility and from high sampling variability arising from sparse age-sex-specific death counts. Successive Nigerian DHS rounds have documented an apparent decline in adult mortality, but whether this reflects genuine improvement or is partly an artefact of uncorrected visibility bias has not been formally tested. Methods We developed a network-adjusted empirical Bayes framework integrating individual-level visibility correction with heteroscedastic SURE-based shrinkage, propagating uncertainty end-to-end via bootstrap re-estimation of the shrinkage hyperparameters. The framework was applied to sibling histories from the 2024 Nigeria DHS and, using an identical pipeline, retrospectively to the 2018 NDHS to enable a temporal comparison free of estimator-choice confounding. Estimates were further evaluated through internal consistency diagnostics, sensitivity analyses, and comparisons with conventional DHS aggregate estimator, Global Burden of Disease and UN World Population Prospects reference series. Results Bias-corrected mortality was lower in 2024 than in 2018 across all fourteen age-sex cells, with declines generally larger from the mid-30s onward and statistically distinguishable from sampling variability at three cells. The conventional and bias-corrected estimators agreed on the direction of decline in thirteen of fourteen cells (mean absolute difference, 3.9 percentage points). The exception, women aged 15–19, showed an aggregate-estimator increase against a bias-corrected decline, coinciding with the cell showing the highest invisible-population fraction and a failed internal consistency check. Corrected 2024 estimates were lower than GBD in 13 of 14 cells and lower than WPP in all cells, with WPP values 2.4–3.7 times higher than the corresponding corrected estimates. Conclusion Integrating network-based visibility adjustment with empirical Bayes smoothing provides a robust framework for estimating age-specific adult mortality from DHS sibling survival histories. The method reduces the effects of visibility bias and sampling variability, requires no external calibration schedule, and is readily transferable to other DHS datasets, making it well suited for mortality surveillance in low-registration settings.