Development and Validation of a Machine Learning-Based Prediction Model for Thrombocytopenia Following Cardiopulmonary Bypass

Postoperative thrombocytopenia (TP) is a serious complication of cardiopulmonary bypass (CPB). An accurate prediction model is crucial for the early identification of high-risk patients. Therefore, this study sought to develop and validate a prediction model for TP following CPB using machine learning (ML) algorithms combined with least absolute shrinkage and selection operator (LASSO) regression for variable selection. Using a large multi-center database, patients who were admitted to the intensive care unit (ICU) following initial CPB were included. Clinical demographics and outcomes with biochemical marker data were collected to determine key variables affecting TP. Prediction models were constructed with four different ML approaches, and model performance was externally validated. The duration of both ICU and total hospitalization were longer and the mortality rate was higher in the TP group than in the non-TP group. LASSO regression identified 10 key predictive variables. The Extreme Gradient Boosting (XGBoost) model performed the best, with an area under the receiver operating characteristic curve (AUC) of 0.841 in internal validation, an accuracy of 0.829, and a precision of 0.955. The platelet count at ICU admission and lactate levels were among key factors influencing the risk of TP.