Predicting warfarin dosage from clinical data: a supervised learning approach | JoVE Visualize

Area of Science:

Pharmacology and Clinical Pharmacy
Artificial Intelligence in Medicine
Data Science in Healthcare

Background:

Warfarin, a widely used anticoagulant, presents challenges in dosage determination due to its narrow therapeutic range and potential for adverse drug events.
Accurate warfarin dosing is critical for patient safety and treatment efficacy, as highlighted by regulatory bodies like the Joint Commission on Accreditation of Healthcare Organizations.
Existing methods for warfarin dosage adjustment often require complex clinical judgment, leading to variability and potential errors.

Purpose of the Study:

To develop and evaluate supervised learning models for predicting optimal warfarin dosage.
To compare the predictive accuracy of various machine learning techniques, including ensemble methods, against baseline clinical practice.
To identify the most effective models for assisting clinicians in warfarin dosage decision-making and mitigating adverse drug events.

Main Methods:

Utilized historical clinical data from 587 Taiwanese patients undergoing warfarin treatment.
Investigated supervised learning algorithms: multilayer perceptron, model tree, k-nearest neighbors, and support vector regression (SVR).
Employed ensemble techniques, including homogeneous (bagging) and heterogeneous (voting) methods, to enhance prediction accuracy.

Main Results:

All machine learning models significantly outperformed the baseline clinical dosage (MAE=0.394, σ(E)=0.558).
Bagged Voting (MAE=0.210, σ(E)=0.357) and Bagged SVR (MAE=0.210, σ(E)=0.366) demonstrated the highest prediction accuracy with the lowest mean absolute error and standard deviation of errors.
Ensemble models showed superior performance in predicting warfarin dosage compared to individual algorithms.

Conclusions:

The developed warfarin dosage prediction models offer a valuable tool for clinical decision support.
Implementation of these models can enhance the precision of warfarin dosing, thereby reducing the risk of adverse drug events for patients.
The study highlights the potential of machine learning, particularly ensemble techniques, to improve anticoagulant therapy management.