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. This study aimed 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 public database, patients admitted to the intensive care unit (ICU) following initial CPB were included. Clinical demographics, outcomes, and biochemical marker data were collected to identify key variables affecting TP. Prediction models were constructed with four different ML approaches, and model performance was externally validated. Both ICU stay 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 model performed the best, demonstrating strong performance in both internal validation (area under the curve [AUC] = 0.841) and external validation (AUC = 0.795). Platelet count at ICU admission and lactate levels were among the key factors influencing TP risk.