Modeling determinants of time to death among low birth weight infants born between 2010 and 2021 in the Iganga-Mayuge population cohort: Uganda

Abstract

Introduction: Low birth weight (LBW) is a major predictor of mortality and morbidity. In Uganda, it was reported that LBW neonates were 3.8 times more likely to die compared to those of normal weight (Arunda et al., 2018). However, only 68.4 % of infants who were 5 years born prior to the 2016 Uganda Demographic and Health Survey (UDHS) had their birth weight reported. Excluding these infants’ data in the estimation of prevalence, survival and associated factors of LBW may lead to loss of study power and biased estimates. Methods: Using data from the IMHDSS open population cohort, the study used multiple imputations to handle missing data. The Kaplan-Meier estimate of the survival function was used to assess overall survival of LBW children and estimate median time to death. Multivariable data analyses were conducted using Cox-Proportional Hazards (PH) and parametric survival analysis models. The parametric model with the least Akaike Information Criteria (AIC) was selected as the best parametric model and used to determine factors associated with time to death. Results: The final sample size comprised 2,385 infants among whom 181 died. On fitting the multivariable Cox-PH model, the proportional hazards assumption was violated so the Exponential and Weibull PH models were not fit. The log-normal model had the least AIC value of among all parametric AFT models. Type of birth (p-value=<0.001) and residing in rural areas (p values=0.022) were the factors associated with time to death. Multiple births were associated with 99.8% decrease in survival compared to singular births (adjusted HR = 0.002, 95% CI = 0.000, 0.025) whereas residing in rural areas was associated with a 92.2% decrease in survival compared to residing in peri-urban areas (adjusted HR = 0.078, 95% CI = 0.009, 0.695). Conclusions: Parametric AFT models fit the data better than the semi-parametric model.