Development of a Risk Prediction Model for Linezolid‐Induced Thrombocytopenia Based on the Machine Learning Algorithm

Abstract The research aimed to develop a validated model for predicting the risk of linezolid‐induced thrombocytopenia (LIT). An XGBoost model and SelectFromModel method were used to screen the important factors. Based on the selected features, five models—Logistic Regression, XGBoost, Random Forest, Naive Bayes, and Support Vector Machine—were established. Finally, the model results were interpreted using SHAP. In this retrospective study, 187 patients were enrolled, and the incidence of LIT was 35.8%. An XGBoost model was established with good performance, in which the AUCs of the training set and validation set were all 0.9. The duration of linezolid treatment, ICU admission time, low baseline platelet level, shock, and concomitant use of piperacillin‐tazobactam were significant risk factors for LIT. A moderately raised level of platelet‐large cell ratio, total bilirubin, and weight may help reduce the incidence of LIT.