Machine Learning-Based Prediction for Stroke Patient Classification using Polygenic Risk Scores.

Area of Science:

• Genetics and Genomics
• Biostatistics and Epidemiology
• Machine Learning in Healthcare

Background:

• Stroke is a major global health concern, causing significant disability and death.
• While non-genetic (NG) factors are crucial, genetic predisposition also influences stroke susceptibility.
• Early identification of high-risk individuals is key for effective stroke prevention.

Purpose of the Study:

• To evaluate the added value of polygenic risk scores (PRS) in machine learning (ML) models for stroke risk prediction.
• To compare the predictive accuracy of ML models using PRS plus NG factors versus NG factors alone.
• To identify factors influencing stroke risk prediction shifts with PRS integration.

Main Methods:

• Developed ML models to predict 10-year incident stroke risk using UK Biobank data.
• Integrated polygenic risk scores (PRS) with traditional non-genetic (NG) clinical variables.
• Optimized models for predictive accuracy using AUROC, analyzing variable influence.

Main Results:

• ML models incorporating PRS demonstrated enhanced predictive accuracy for stroke risk compared to NG factors alone.
• Analysis revealed shifts in predicted stroke probability upon PRS inclusion.
• Identified key demographic, biomarker, and clinical factors influencing model predictions.

Conclusions:

• Polygenic risk scores offer significant added value to non-genetic factors in ML-based stroke prediction.
• Integrating PRS into routine practice alongside NG factors can improve early stroke diagnosis and patient outcomes.
• Further research into specific cohorts and variable interactions can refine personalized stroke prevention.