Application of Random Survival Forests andAccelerated Failure Time Shared Frailty Models in Understanding Under-Five Child Mortality in Kenya

Abstract

Background: Under-five mortality rates is one of the health indicators of great importance for any country. Kenya is among those nations in the sub-saharan part of Africa which has high under-five deaths, and thus it will be of importance to apply best statistical approaches to establish which factors have influence on child mortality, this will assist to plan for the interventions. Approach: Our study employed use of Random Forest for Survival Regression and Classi- fication to analyze the Kenya Demographic Health Survey (KDHS) 2014 data to do selection of the risks factors for the under-five mortality. Akaike Information Criterion (AIC) statistics was employed to select most appropriate accelerator failure time (AFT)-shared frailty model. Results: The results gotten through fitting the AFT-shared frailty model was that there was presence of unmeasured factors at community cluster while at household cluster there was no evidence suggesting existence of the unmeasured factors. Log-logistic AFT-model showed that the sons who have died, daughters who have died, duration of breastfeeding, and months of breastfeeding were found to be having significant influence on the under-five mortality (p < 0:05). Log-logistic AFT model with Gaussian frailty was the most appropriate model for under-five child mortality due it’s least Akaike Information Criterion (AIC) statistic. Conclusion: Our study found out that there was presence of unobserved heterogeneity at community clusters, this means that there are other influences that do affect mortality at community clusters which the variables alone in the model cannot explain. On the other hand there was no presence of the unobserved heterogeneity at household clusters, implying that factors influencing under-five deaths in the households can be clarified just by using the covariates in the model without the inclusion of household cluster term.