Development of a Risk Prediction Model for Linezolid-Induced Thrombocytopenia Based on the Machine Learning Algorithm
Jie Chi1, Juan Wang1, Heng Tang1
1Department of Pharmacy, Tongling Municipal Hospital, Tongling, Anhui, People's Republic of China.
Journal of Clinical Pharmacology
|July 4, 2025
Summary
This study developed a predictive model for linezolid-induced thrombocytopenia (LIT). Key risk factors identified include linezolid duration, ICU stay, low platelets, shock, and piperacillin-tazobactam use.
Area of Science:
- Pharmacology
- Clinical Medicine
- Data Science
Background:
- Linezolid-induced thrombocytopenia (LIT) is a significant clinical concern.
- Accurate prediction of LIT risk is crucial for patient management.
Purpose of the Study:
- To develop and validate a predictive model for linezolid-induced thrombocytopenia (LIT).
- To identify key risk factors associated with LIT development.
Main Methods:
- Retrospective study of 187 patients treated with linezolid.
- Feature selection using XGBoost and SelectFromModel.
- Development and comparison of five predictive models: Logistic Regression, XGBoost, Random Forest, Naive Bayes, and Support Vector Machine.
- Model interpretation using SHAP values.
Main Results:
- The incidence of LIT was 35.8% in the study cohort.
- An XGBoost model demonstrated excellent performance with AUCs of 0.9 for both training and validation sets.
- Significant risk factors for LIT included duration of linezolid treatment, ICU admission time, low baseline platelet count, shock, and concurrent use of piperacillin-tazobactam.
- Platelet-large cell ratio, total bilirubin, and weight were identified as potentially protective factors.
Conclusions:
- The developed XGBoost model effectively predicts the risk of linezolid-induced thrombocytopenia.
- Identifying high-risk patients allows for closer monitoring and potential intervention.
- Understanding risk factors can guide clinical decision-making regarding linezolid therapy.
