One-Year Risk of Bleeding in Patients on a Stable Warfarin Dose After Prosthetic Heart Valve Surgery.

Purpose: Anticoagulation therapy is required to prevent thromboembolic complications in patients with heart valve surgery (HVS). However, caution must be taken due to the risk of bleeding. This study aimed to identify bleeding risk factors in patients with stable warfarin therapy and develop a predictive tool for high-risk patients.

Methods: This study is a nested case-control design using the Korean National Health Insurance Service-National Sample Cohort Data. We identified patients who underwent HVS and were prescribed warfarin within 1 week after the procedure. Of these, patients with the last two identical warfarin prescriptions within 6 months before the bleeding events were defined as the case group, while patients with no bleeding events within 6 months after HVS and two consecutive identical warfarin prescriptions were defined as the control group. Three machine learning models-logistic regression, support vector machine, and random forest-were trained and scored by fivefold validation to validate our feature selection processes. We developed a risk scoring system using adjusted odds ratios from multivariate logistic regression.

Results: Of 1 137 861 subjects, 1093 patients were eligible for the study cohort; 173 and 298 were selected as the case and control groups, respectively. After a series of machine learning processes, eight features were identified as significant risk factors for bleeding events.

Conclusion: Our finding suggests that furosemide, spironolactone, lacrimal system disorders, ursodeoxycholic acid, captopril, chronic kidney disease, zolpidem, and valsartan are the most important features for predicting bleeding events in patients taking a stable warfarin dose after HVS.