MACHINE LEARNING PREDICTION MODELS FOR IN-HOSPITAL MORTALITY AFTER ISOLATED ON-PUMP CORONARY ARTERY BYPASS GRAFTING

Abstract

TYPE: Abstract TOPIC: Cardiothoracic Surgery PURPOSE: Study aims to develop prediction models for in-hospital mortality (IHM) after isolated on-pump coronary artery bypass grafting (CABG) based on modern machine learning (ML) methods and to compare their accuracy. METHODS: Data set obtained from 866 electronic medical records of coronary artery disease patients received isolated on-pump CABG surgery from 2008 to 2018. Preoperative clinical, laboratory and instrumental tests results were used as inputs. The overall IHM was 4%. Predictors were identified during multistep selection procedure of statistical hypotheses analysis and weight coefficients calculations. Logistic regression, random forest and artificial neural networks (ANN) were used for models construction and predictors verification. Models cross validation was performed on test samples and control validation - on samples, whose data were not used in models development. RESULTS: Highest predictive potential IHM risk factors were identified: 7 from the EuroSCORE II scale (age, LVEF, recent MI, extracardiac arteriopathy, urgency, NYHA class) and 5 additional (heart rate, systolic blood pressure, aortic stenosis, left ventricular (LV) relative wall thickness index, and LV relative mass index). Developed ML models showed higher AUC values and sensitivity compared to the classical EuroSCORE II scale with the maximal prognostic accuracy for ANN model (AUC 93 %, sensitivity 90 %, and specificity 96 %). The predictive robustness of the models was confirmed by results of the control validation. CONCLUSIONS: Modern ML technologies allowed to develop robust prediction models for IHM after isolated on-pump CABG. CLINICAL IMPLICATIONS: Developed model could be promising for medical decision-making support system and deserves further clinical validation. DISCLOSURE: Nothing to declare. KEYWORD: machine learning