Evaluation of Machine Learning Approaches for Predicting Warfarin Discharge Dose in Cardiac Surgery Patients:...

Area of Science:

• Cardiovascular Surgery
• Pharmacology
• Machine Learning

Background:

• Cardiac surgery patients exhibit increased warfarin sensitivity, raising risks of adverse events.
• Accurate warfarin dosing is crucial for effective anticoagulation post-cardiac surgery.
• Predictive algorithms are needed to optimize warfarin management in this population.

Purpose of the Study:

• To develop and validate machine learning algorithms for predicting optimal warfarin dosage.
• To achieve a therapeutic international normalized ratio (INR) at discharge for cardiac surgery patients.
• To compare different machine learning models for warfarin dose prediction accuracy.

Main Methods:

• Utilized data from 1031 cardiac surgery encounters initiating warfarin.
• Compared penalized linear regression, k-nearest neighbors, random forest, gradient boosting, MARS, and an ensemble model.
• Developed separate models for target INRs of 2.0-3.0 and 2.5-3.5, using cross-validation and evaluating mean absolute error (MAE).

Main Results:

• Random forest regression best predicted doses for target INR 2.0-3.0 (MAE 1.13 mg, 39.5% within 20% of target).
• An ensemble model excelled for target INR 2.5-3.5 (MAE 1.11 mg, 43.6% within 20% of target).
• Implementation of algorithms increased therapeutic INR discharge rates from 47.5% to 61.1%.

Conclusions:

• Machine learning models effectively predict warfarin doses using routine clinical data.
• These algorithms can guide initial warfarin dosing in cardiac surgery patients.
• Optimized anticoagulation management is achievable through predictive modeling.