Developing a machine learning model for bleeding prediction in patients with cancer-associated thrombosis receiving anticoagulation therapy

Only one conventional score is available for assessing bleeding risk in patients with cancer-associated thrombosis (CAT): the CAT-BLEED score.Our aim was to develop a machine learning-based risk assessment model for predicting bleeding in CAT and to evaluate its predictive performance in comparison to the CAT-BLEED score.We collected 488 attributes (clinical data, biochemistry, and ICD-10 diagnosis) in 1080 unique patients with CAT. We compared CAT-BLEED score, Ridge and Lasso Logistic Regression, Random Forest, and XGBoost algorithms for predicting major bleeding or clinically relevant non-major bleeding (CRNMB) occurring 1-90 days, 1-365 days, and 90-455 days after venous thromboembolism (VTE).The predictive performances of Lasso Logistic Regression, Random Forest, and XGBoost were higher compared to the CAT-BLEED score in the prediction of bleeding occurring 1-90 days and 1-365 days. For predicting major bleeding or CRNMB 1-90 days after VTE, the CAT-BLEED score achieved a mean AUC of 0.48 ± 0.13, while Lasso Logistic Regression and XGBoost both achieved AUCs of 0.64 ± 0.12. For predicting bleeding 1-365 days after VTE, the CAT-BLEED score achieved a mean AUC of 0.47 ± 0.08, while Lasso Logistic Regression and XGBoost achieved AUCs of 0.64 ± 0.08 and 0.59 ± 0.08, respectively.This is the first machine learning risk model for bleeding prediction in patients with CAT receiving anticoagulation therapy. Its predictive performance was higher compared with the conventional CAT-BLEED score. With further development, this novel algorithm might enable clinicians to perform personalized anticoagulation strategies with improved clinical outcomes.