Optimizing warfarin dosing in diabetic patients through BERT model and machine learning techniques.

This study highlights the importance of evaluating warfarin dosing in diabetic patients, who require careful anticoagulation management. With rising rates of diabetes and cardiovascular diseases, understanding the factors influencing warfarin therapy is vital for improving patient outcomes and reducing adverse events. Data was sourced from the IWPC dataset, examining characteristics such as age, gender, diabetes status, indication for warfarin, weight, and height. We utilized the Bidirectional Encoder Representations from Transformers (BERT) model to analyze therapeutic doses, leveraging its ability to understand contextual relationships in the data. A machine learning approach was essential for predicting appropriate warfarin dosages, employing algorithms like Random Forest, KNN, MLP, Linear Regression, and SVM classification. We allocated 20 % of the data for testing and 80 % for training. Results showed that Linear Regression performed less effectively than MLP, KNN, SVM, and Random Forest in both training and testing. Notably, Random Forest's training MAE was significantly lower, while the other models showed similar performance in predicting warfarin dosages. This study emphasizes the importance of personalized anticoagulation management for diabetic patients on warfarin. The application of the BERT model alongside machine learning algorithms, particularly Random Forest, demonstrated effectiveness in predicting appropriate dosages. These findings suggest that integrating these advanced models into clinical practice can enhance decision-making, optimize patient outcomes, and reduce adverse events.