Prediction of hemorrhagic transformation after thrombolysis based on machine learning models combined with platelet...

Hemorrhagic transformation (HT) is a common and potentially serious complication following intravenous thrombolysis (IVT) in patients with acute ischemic stroke (AIS). Despite its high incidence, there remains a lack of simple and effective tools for predicting HT risk.

This study aimed to develop an interpretable machine learning (ML) model incorporating the platelet distribution width to platelet count ratio (PPR) to predict HT occurrence in AIS patients after IVT.

We included AIS patients who underwent IVT at the First Affiliated Hospital of Kunming Medical University between July 2019 and April 2024. Four ML models-logistic regression (LR), random forest (RF), support vector machine (SVM), and extreme gradient boosting (Xgboost)-were constructed using 5-fold cross-validation, with HT after IVT as the outcome. Feature selection was performed using least absolute shrinkage and selection operator (LASSO) regression. Model performance was evaluated based on the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, and balanced F-score. The best-performing model was selected for interpretability analysis, and feature importance was assessed.

LASSO regression identified six predictive features with non-zero coefficients: age, diabetes, malignancy, onset-to-treatment time (OTT), baseline National Institutes of Health Stroke Scale (NIHSS) score, and PPR. Among the models, LR demonstrated the highest predictive performance, achieving an optimal AUC of 0.919, along with average accuracy, sensitivity, and specificity of 0.825, 0.830, and 0.832, respectively. Feature importance in the LR model ranked as follows: baseline NIHSS score, diabetes, PPR, malignancy, age, and OTT.

The LR-based model incorporating PPR effectively predicts HT risk in AIS patients after IVT, providing clinicians with a rapid and accurate tool to assess thrombolytic hemorrhage risk and support treatment decision-making.