Decision Tree Modification Based on Multi-level Data Categorization

The study is aimed on modifying a decision tree (DT) for predicting adverse events in clinical medicine by including risk factors (RF) in its structure, identified using multi-level categorization of predictors. A retrospective cohort study was conducted using data from 4,673 electronic medical records of patients with a diagnosis of ST-segment elevation myocardial infarction (STEMI) who underwent percutaneous coronary intervention (PCI). Patients were divided into two groups; the first group consisted of 318 (6.8%) patients who died in the hospital, the second group included 4,359 (93.2%) patients with a favorable outcome of PCI. DT method and multimetric categorization of predictors were used to create prognostic models for in-hospital mortality (IHM). The performance of the models was assessed using 6 quality metrics. The study endpoint was the all-cause IHM in patients with STEMI after PCI. A modified DT method has been developed on the basis of multi-level categorization of predictors and identification of RF for IHM. A comparative analysis of the quality of models based on the CART and modified DT algorithms showed higher performance of the second (AUC 0.813 vs 0.765, p-value = 0.003). The advantage of this method is the ability to extract production rules that ensure transparency of the generated predictive solutions. Conclusions. A model based on a modified DT algorithm is an effective prognostic tool allowing high performance estimation of IHM probability and clinical interpretation of the prognostic results.