Machine Learning-Based Prediction of Drug-Induced QTc Changes in a Large Finnish Biobank Cohort

Area of Science:

• Cardiology
• Pharmacology
• Bioinformatics

Background:

• QT interval prolongation is a risk factor for cardiac arrhythmias and sudden cardiac death, often linked to medication use.
• Current methods for evaluating drug-induced QT prolongation rely on literature synthesis and may not capture real-world data promptly.
• Machine learning (ML) offers potential for automated risk detection by analyzing real-world data.

Purpose of the Study:

• To evaluate if an ML model trained on real-world genomic and medication data can identify drug-QTc associations.
• To assess the potential for automated risk detection in clinical workflows using ML.

Main Methods:

• Utilized data from 10,208 individuals in the FinnGen biobank, including prescription records, clinical variables, and genetic information.
• Developed an ML framework using nested cross-validation to predict QTc duration based on clinical factors, medication purchases, and polygenic risk scores.
• Conducted linear regression analyses to validate ML findings.

Main Results:

• Few ML-identified drug-QTc associations matched known effects from expert-curated databases.
• Observed several false positives, and true positives like amiodarone showed clinically insignificant effect sizes in the ML model (+1 ms) compared to linear regression (+49 ms).
• ML models did not reliably identify medications associated with QT-interval prolongation.

Conclusions:

• ML models face challenges in detecting clinically meaningful drug effects on the QT interval.
• The use of QTc as an intermediate marker for electrophysiological vulnerability in this ML framework was limited.
• Systematic evidence reviews by clinical pharmacology experts are currently indispensable for medication safety assessment.