Development and Comparison of Generalized Linear Models for Predicting Avoidable Mortality: Fixed vs. Mixed Effects Approach

Abstract

Publicly available data were obtained from the OECD database to develop a model that could help understand the factors contributing to avoidable mortality across different countries. Avoidable mortality refers to the number of deaths that could have been prevented through public health and prevention interventions, and is expected to vary based on demographic, economic, and health factors. Features that were selected as possibly contributing to the variability in avoidable mortality included health-economic factors (public and private health expenditure), health-care availability factors (number of physicians and nurses, and pharmaceutical consumption), and demographic factors (country and year). Furthermore, all variables were standardized based on population size.Two generalized linear regression models were developed and compared. Fixed model corresponded to multiple linear regression and mixed model corresponded to the random effect model with variable Country as a random intercept. The mixed model showed better performance in both training and testing statistics, indicating that the differences between countries account for most of the variation in avoidable mortality.Furthermore, both models were applied to complete cases and imputed cases using multiple imputation. However, imputing the missing cases did not result in any significant improvement in the training and testing statistics for either of the models.