Multilevel Survival Modelling of Neonatal Mortality under Some Prognostic Factors in Uttar Pradesh

Abstract

Introduction: Neonatal mortality rate is a critical indicator of a nation's healthcare system and socio-economic development. While standard survival models assume independence among observations this assumption is inadmissible for hierarchical data structures, limiting their applicability. To address this limitation, this study employed parametric multilevel mixedeffects survival models that explicitly account for clustering and unobserved heterogeneity.

Methods: This study employed unit level data from National Family Health Survey (NFHS-5), 2019-2021 for Uttar Pradesh, which employs a stratified multistage sampling design. Multilevel mixed-effects survival models were fitted (Weibull and exponential distributions) with random intercepts at the Primary Sampling Unit (PSU) and district levels to account for hierarchical clustering. Model fit was assessed using the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), while Likelihood Ratio Tests (LRT) compared nested specifications. The Intra-Class Correlation (ICC) was computed to quantify unobserved heterogeneity attributable to clustering at the district and PSU levels.

Results: The Multilevel Weibull Mixed-effects model provided the best fit, revealing that unobserved heterogeneity was predominantly clustered at the PSU level (variance: 1.238) rather than the district level (0.023). Advanced maternal age (45–49 years) was a critical risk factor (HR = 2.943, 95% CI: 1.729–4.713), as was the use of Smokey cooking fuels (HR = 1.304, 95% CI: 1.014–1.677). Neonates not weighed at birth faced a significantly higher mortality risk (HR = 1.612, 95% CI: 1.287–2.019) compared to those with normal birth weight. Maternal education, religion, and place of delivery were also identified as significant determinants of neonatal outcomes.

Conclusion: The study recommends that efforts to reduce neonatal mortality should address not only individual-level risk factors but also community-level disparities in healthcare access and quality.