Leveraging Machine Learning to Predict Warfarin Sensitivity in the Puerto Rican Population: A Pharmacogenomic...

Area of Science:

• Pharmacogenomics
• Machine Learning
• Personalized Medicine

Background:

• Warfarin is a widely used oral anticoagulant, but its dosage varies significantly due to genetic and clinical factors.
• Existing pharmacogenomic models often fail to account for the admixed genetic profiles of populations like Caribbean Hispanics.
• Adverse drug events from warfarin are a significant cause of hospitalizations, particularly in older adults.

Purpose of the Study:

• To compare eight machine learning methods for predicting warfarin sensitivity in Puerto Rican patients.
• To assess the utility of machine learning in personalized warfarin therapy for ethno-specific populations.
• To identify potential genetic contributors to warfarin sensitivity in Caribbean Hispanics.

Main Methods:

• Secondary analysis of genetic and clinical data from 146 Puerto Rican patients treated with warfarin.
• Comparison of eight machine learning algorithms, including a gradient boosting classifier.
• Evaluation of model performance using precision, accuracy, and confusion matrixes.

Main Results:

• A gradient boosting classifier achieved the highest accuracy (0.7500) and weighted precision (0.7642).
• The sensitivity for detecting warfarin-sensitive patients remained low across all tested models.
• The genetic variant rs202201137 was suggested as a potential contributor to model predictions.

Conclusions:

• Machine learning shows potential for predicting warfarin sensitivity in Puerto Rican populations, but current models have limitations.
• Improved accuracy and sensitivity are needed for clinical utility in personalized warfarin dosing.
• Further investigation into ethno-specific genetic variants is crucial for refining pharmacogenomic models.