Pharmacogenomics driven decision support prototype with machine learning: A framework for improving patient care.

Area of Science:

• Bioinformatics
• Clinical Informatics
• Pharmacogenomics

Background:

• Electronic health records (EHRs) are increasingly used for clinical decisions, but often lack genetic data integration.
• Few healthcare systems effectively incorporate pharmacogenomic data into routine care.
• Informed pharmacogenomic decision-making is crucial for optimizing patient outcomes.

Purpose of the Study:

• To develop and validate informatics methods and predictive models for real-time pharmacogenomic decision support.
• To integrate electronic health record (EHR) and genetic data for predicting adverse drug events.
• To create a prototype system using Department of Veterans Affairs (VA) data.

Main Methods:

• Utilized informatics and predictive modeling to create algorithms integrating EHR and genetic data.
• Developed a prototype using EHR and genetic data from 2,600 HIV patients and controls.
• Mapped patient medications to Clinical Pharmacogenomics Implementation Consortium (CPIC) level A guidelines to identify potential adverse events.

Main Results:

• The predictive model incorporating genetic data achieved AUC scores of 0.972 and F1 scores of 0.97.
• Models without genetic data integration showed significantly lower performance (AUC 0.766, F1 0.73).
• XGBoost demonstrated superior performance in predicting preventable adverse events (PAE) when genetic data was included.

Conclusions:

• Integrating genetic data into EHR systems substantially enhances the accuracy of predicting preventable adverse drug events.
• The developed framework and XGBoost model offer a promising approach for real-time, informed pharmacogenomic decision-making.
• This study highlights the potential of leveraging pharmacogenomics for safer and more effective patient care.