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

Stroke remains a leading cause of morbidity and mortality worldwide, with a significant portion of cases resulting in long-term disability or death. Early identification of high-risk individuals is critical for timely intervention and prevention. While modifiable risk factors such as hypertension, smoking, and diabetes account for nearly 80% of stroke cases, genetic predisposition also plays a crucial role in stroke susceptibility. This study explores the impact of genetic risk by integrating polygenic risk scores (PRS) with traditional non-genetic (NG) clinical variables in machine learning (ML) models. Leveraging data from the UK Biobank (UKB), we assess how PRSs enhance predictive accuracy in estimating stroke risk compared to NG factors alone. PRSs have been instrumental in refining disease prediction across various conditions, including cardiovascular disease and type II diabetes, suggesting their potential in stroke risk modeling. By optimizing model performance to maximize the area under the receiver-operating characteristic curve (AUROC), this study aims to improve early recognition of high-risk individuals and contribute to more effective, personalized stroke prevention strategies. In this study, we developed ML models to predict the 10-year incident risk of stroke and examined how biomarkers, demographic factors, diagnoses, medications, and other variables differed among cohorts. Specifically, we analyzed how the model's predicted probability of stroke changed with the inclusion of PRSs alongside NG data, identifying the factors most influential in driving these shifts.Clinical RelevanceIncorporating PRSs in ML models shows added value over NG factors alone in stroke prediction, and certain cohorts can be shown to have various other characteristic differences for certain biomarkers, demographic factors, diagnoses, medications, and other key variables that can propel future stroke research. These findings reinforce the case for implementing PRSs in routine medical practice with established NG factors currently used to enhance early diagnosis and patient outcomes for stroke.