Dynamic mortality prediction using machine learning techniques for acute cardiovascular cases

Abstract This paper represents the research results of applying machine learning methods for early predicting of cardiovascular patients mortality. The classification task is solved by analyzing the dynamics data from electronic health records of the patients with the acute coronary syndrome, infarction, and stable angina. Moreover, the approach for identification of model components and their connection is developed. The model structure identification assimilates the patient condition, treatment phases in treatment dynamic. It provides the prediction of better model structure on following steps and request for necessary data to improve the forecast for more informed decision-making. The dynamic data extracted directly from medical information system were analyzed, that is very close to the real process. Using machine learning methods it is possible to make an early prediction of mortality risks. The prediction of laboratory results allows saving the resources. Jointly, both can be offered to clinicians as support for accurate, reasonable saving clinical decisions with minimization risks for patient’s health. The simple lab test results like hemoglobin (HGB), red blood cells (RBC), alanine transaminase (ALT), aspartate transaminase (AST), glucose, platelet (PLT), creatinine levels are used as a predictor. Such a simple approach to solving critical tasks can make the method widely used in clinical practice. The identification of the patient groups’ individuality into account the dynamics probably can contribute to E-science.