A Comparative Analysis of Logistic Regression and Decision Trees for Mortality Risk Prediction using Laboratory Data

Predicting clinical outcomes, particularly mortality risk, is a crucial aspect of patient care and management in the medical field. Numerous studies have employed various methods to improve the accuracy and reliability of these predictions. Among these methods, Logistic Regression has been widely used for its simplicity and interpretability. However, recent advancements suggest that alternative methods like Decision Trees might offer improved performance and interpretability. In this study, we compare the performance of Logistic Regression and Decision Trees in predicting mortality risk among patients diagnosed with heart failure. The data for this analysis was sourced from the MIMIC-III database, encompassing a total of $\mathbf{1, 1 7 7}$ patients. Only laboratory data from the patients’ blood tests were considered. We employed the 5 -fold cross-validation technique to evaluate the models. The performance of each model was assessed using the area under the ROC curve (AUC) or c-index. The Logistic Regression model achieved a c-index of $0.743 \pm 0.048$ on the testing dataset, while the Decision Trees model performed slightly better, with a c-index of $0.787 \pm 0.040$. Based on the experimental results and a thorough assessment of the merits and limitations of each approach, we conclude that Decision Trees not only provide more intuitively comprehensible models but also yield relatively favorable outcomes in terms of discrimination, as measured by the cindex, compared to Logistic Regression.