PREDICTIVE MODELING OF CHILD MORTALITY IN MIGORI AND NYAMIRA COUNTIES USING INDIRECT METHODS

Abstract

Child mortality remains a critical public health challenge, particularly in developing countries like Kenya, where disparities in healthcare are stark across different regions. In counties such as Nyamira and Migori, persistent high rates of under-five child mortality demonstrate the need for more precise statistical predictions for and targeted interventions. Traditional methods for estimating child mortality, such as those derived from household surveys, are often hampered by issues like missing data and survivor bias, leading to inaccurate mortality estimates. This study sought to develop a comprehensive predictive model for under-five child mortality in Migori and Nyamira counties, Kenya, by incorporating temporal patterns and social determinants of health. Utilizing a retrospective cohort design, the study analyzed historical data from health records, census reports, and household surveys spanning 34 years (1989-2022). The analysis incorporated indirect estimation techniques to address data gaps and employed multiple linear regression, gradient boosting regressor, and spatio-temporal modeling to capture temporal and seasonal trends in child mortality. The multiple linear regression model was significant, explaining 89.9% of the change in neonatal mortality in Migori County and 80.6% of the variation in Nyamira County. Gradient boosting regressor performed optimally, accounting for 80.9% of the change in child mortality, indicating good predictive capability and suggesting that the chosen independent variables effectively capture the complexity of the response variable. Spatio-temporal modeling log-likelihood value of -111.87 indicated a relatively good fit, capturing the observed data well (pseudo-R-squared = 0.9415). Results indicated that infant mortality rates in both counties have fluctuated historically, with distinct seasonal trends influenced by factors such as disease prevalence and access to healthcare services. The temporal and seasonal analysis revealed that periods of increased respiratory complications and malaria prevalence corresponded with higher mortality rates. The study provides a methodological framework that can be adapted to other regions with comparable challenges. By addressing the limitations of traditional mortality estimation methods and leveraging advanced predictive modeling techniques, the study contributes to the ongoing efforts to improve child health outcomes in Kenya and beyond.