Skip to main content
ISTH 2023 Main banner
Back

    Poster Networking

    Venous Thromboembolism

    Poster Networking Session

    PB0966 - Machine learning model for bleeding prediction in patients with cancer-associated thrombosis receiving anticoagulation therapy

    Monday, June 26, 2023
    18:30 – 19:30 ET
    Room: Exhibition
    Background: There is only one conventional score for the assessment of bleeding risk in patients with cancer associated thrombosis (CAT), the CAT-BLEED score. There is no machine learning model until

    Aims: Our aim was to develop a risk assessment model for the prediction of bleeding in CAT using machine learning techniques and to assess its predictive performance compared with CAT-BLEED.

    Methods: We collected 488 attributes (clinical data, biochemistry and ICD-10 diagnosis) in 1080 unique patients identified through Østfold Thrombosis Registry.
    We compared CAT-BLEED, lasso logistic regression and XGBoost algorithms for the prediction of major bleeding or clinically relevant non-major bleeding (CRNMB) occurring 1-90 days after VTE and 1-365 days after VTE, respectively. Mean area under the receiver characteristic curve (AUC) and area under the precision recall curve (AUPRC) was determined using 10-fold cross validation.

    Results: The predictive performances of lasso logistic regression and XGBoost were higher compared to the CAT-BLEED score in prediction of bleeding occurring after 1-90 days and 1-365 days (table 1 and figure 1). For the prediction of major bleeding or CRNMB 1-90 days after VTE the CAT-BLEED score achieved a mean AUC=0.48, while lasso logistic regression achieved and XGBoost both achieved AUC=0.64. For the prediction of bleeding 1-365 days after VTE, the CAT-BLEED score achieved a mean AUC=0.47, while lasso logistic regression and XGBoost achieved AUC=0.64 and 0.59, respectively.

    Conclusion(s): This is the first machine learning risk model of bleeding prediction in patients with CAT receiving anticoagulation therapy. Its predictive performance is higher compared with the conventional CAT-BLEED score. With further development, the novel algorithm might enable a clinician to perform a personalized anticoagulation strategy and duration of treatment with better clinical outcomes.