Development and Validation of Machine Learning Algorithms to Predict 1-Year Ischemic Stroke and Bleeding Events in Patients With Atrial Fibrillation and Cancer

OBJECTIVES: Current risk assessment tools demonstrate low predictive performance or have not been validated in patients with atrial fibrillation (AFib) and cancer. Thus, we developed and validated new machine learning (ML)-based risk assessment tools for stroke and bleeding prediction among this patient population.

METHODS: We conducted a retrospective cohort study including patients aged ≥66, newly diagnosed with AFib with a record of cancer from the 2012-2018 Surveillance, Epidemiology, and End Results (SEER)-Medicare database. Patients were required to continuously enroll in Medicare Parts A, B, and D for at least 12 months before and after AFib diagnosis. Patients with valvular diseases, repair or replacement, venous thromboembolism, or joint replacement at baseline or receiving oral anticoagulants during the study period were excluded. We developed and validated the ML algorithms separately for stroke and major bleeding by fitting elastic net, random forest (RF), extreme gradient boosting, support vector machine, and neural network models with 10-fold cross-validation (train:test=7:3). We obtained area under the curve (AUC), sensitivity, specificity, and F2 score. Model calibration was assessed using the Brier score. In sensitivity analysis, we resampled the data using the Synthetic Minority Oversampling Technique (SMOTE).

RESULTS: Among 18,388 eligible patients, 523 (2.84%) had ischemic stroke and 221 (1.20%) developed major bleeding after one year. In predicting ischemic stroke, RF significantly outperformed other ML models [AUC=0.916, 95% CI=0.887-0.945, sensitivity=0.868, specificity=0.801, F2 score=0.375, Brier score=0.035]. However, the performances of ML algorithms in predicting major bleeding were low (RF’s AUC=0.623, 95% CI=0.554-0.692). SMOTE did not improve the performance of the ML algorithms.

CONCLUSIONS: Our study demonstrated a promising application of ML in stroke prediction among patients with AFib and cancer. This tool may be leveraged to assist clinicians to identify patients at high risk of stroke in order to optimize treatment decisions.

See COI here.