Development and Validation of an Interpretable Machine Learning Model for Predicting Thrombocythemia Risk During...

Area of Science:

• Pharmacology
• Medical Informatics
• Machine Learning

Background:

• Third-generation cephalosporins are crucial for severe infections but can cause thrombocythemia, complicating treatment.
• Existing tools for predicting this risk lack accuracy and clinical interpretability.
• There is a need for reliable methods to stratify thrombocythemia risk in patients receiving cephalosporins.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for predicting thrombocythemia risk in patients treated with third-generation cephalosporins.
• To enhance clinical decision-making by providing accurate and understandable risk stratification.
• To identify key predictors of thrombocythemia in this patient population.

Main Methods:

• A retrospective cohort of 25,707 adult patients receiving third-generation cephalosporins was analyzed.
• Machine learning algorithms (XGBoost, Random Forest, LightGBM) were trained and tested, with performance evaluated using ROC-AUC and Brier score.
• SHAP analysis was employed for model interpretability, identifying key predictive factors.

Main Results:

• The XGBoost model achieved superior performance with an AUC of 0.858 and a Brier score of 0.0088.
• Key predictors included baseline platelet count, red blood cell count, creatinine, daily usage frequency, and sex.
• Malignancies increased risk, while female sex showed a protective effect.

Conclusions:

• An interpretable ML framework was successfully developed for precise thrombocythemia risk prediction during cephalosporin therapy.
• This model balances high algorithmic performance with clinical actionability, aiding therapeutic decisions.
• The findings offer a valuable tool for managing potential adverse events associated with cephalosporin treatment.