Joint Modeling of Mortality Incidence and Survival

Abstract

Analysis of medical data mostly consider survival and mortality count as response variables in identifying factors that are associated with the survival times and with the death count of patients. However, it is quite possible and the literature has shown evidence for these two responses to be correlated and share common factors. Therefore, joint modeling of these two responses simultaneously in one model can provide improved results than fitting two univariate models since the correlation between the two responses can be captured in a joint model. The literature did not consist of any such situation where joint modelling of survival time and the death count was considered. This manifested the objective of developing a method for jointly modeling survival and count responses for which a bivariate Poisson model was proposed. This method was facilitated by the equivalence of the loglikelihoods of survival and Poisson models. The suggested method was fitted for a data set of Dengue patients where factors associated with survival times of dengue patients and death count of patients were identified by the joint model. For comparing the performance of the proposed joint model with two univariate models that can be fitted separately for the two responses, the Akaike Information Criterion (AIC) was used. It was confirmed that the performance of the joint model surpasses the fit of two univariate models since the AIC of the joint model was lower than the total of the AICs of the two univariate models.