INTRODUCTION:In Turkey, there is a scarcity of knowledge about the predictors of health outcomes at a national level, and it is well known that there is a gap between rural and urban parts of developing countries in terms of the level of health outcomes. This study aims to find out predictor factors of the public health outcomes at a province level in Turkey.METHODS:Life expectancy at birth and mortality are used as public health outcome indicators. Logistic regression and Random Forest classification generated by using 50, 100, and 150 trees were used to compare prediction performance of health outcomes. The results of different prediction methods were recorded changing the “k” parameter from 3 to 20 in k-fold cross validation. The Area Under the ROC Curve (AUC) was used as a measure of prediction accuracy. Prediction performance differences were tested using Kruskall-Wallis analysis and visualized on a heatmap. Finally, predictor variables of public health outcomes were shown on a decision tree.RESULTS:Study results revealed that Logistic regression outperformed Random Forest classification. The difference between all prediction methods to predict public health outcome indicators was statistically significant (p<.000). The heatmap shows that AUC values to predict mortality have superior performance when compared with life expectancy at birth. Decision tree graphs present that the most important predictor variables were total number of beds for mortality and percentage of higher education graduates for life expectancy at birth.CONCLUSIONS:The results of this study represent a preliminary attempt to determine public health outcome indicators. It is hoped that the results of this study serve as a basis to understand the determinants of health care outcomes at province level with focus on a developing country. This study illustrates that there is a need to spend extra effort for future studies to analyze public health outcomes to improve social welfare functions in health systems.
Public health outcome of Tuberculosis Cluster Investigations, England 2010–2013
