| Title: | GRaph-based analysis for stroke prediction (GRASP): A multi-modal model for identifying first ischemic stroke in high-risk population using UK biobank |
| Journal: | Computers in Biology and Medicine |
| Published: | 19 Apr 2026 |
| Pubmed: | https://pubmed.ncbi.nlm.nih.gov/42008901/ |
| DOI: | https://doi.org/10.1016/j.compbiomed.2026.111674 |
Publication 17941
WARNING: the interactive features of this website use CSS3, which your browser does not support. To use the full features of this website, please update your browser.
Abstract
BACKGROUND: Ischemic stroke accounts for 3.71 million deaths annually worldwide. Yet current risk prediction models demonstrate modest discrimination with reported C-statistics typically ranging from 0.63 to 0.75 and cannot reliably identify which individuals within high-risk populations will experience first ischemic stroke (FIS). We developed GRaph-based Analysis for Stroke Prediction (GRASP), a novel multimodal approach to predict FIS likelihood among individuals at risk who share similar vascular risk profiles.</p>
METHODS: We analyzed n = 1226 UK Biobank participants having at least one established vascular risk factor associated with FIS, including 317 (25.9%) who developed FIS during follow-up (mean time to FIS onset: 9.5 years post-recruitment). Participants were stratified into at-risk with future ischemic stroke (AR-FIS) and at-risk with no ischemic stroke (AR-NIS). GRASP employed Graph Attention Network architecture incorporating 136 clinically relevant variables across demographic, clinical, lifestyle, and neuroimaging modalities. Performance was compared against baseline vascular risk factor models and conventional machine learning classifiers.</p>
RESULTS: GRASP achieved superior performance in distinguishing AR-FIS from AR-NIS populations with AUC-ROC of 0.82 (95% CI: 0.78-0.87) and PR-AUC of 0.73 (95% CI: 0.67-0.80). This represented a 15.3% improvement in AUC-ROC and 27.9% improvement in PR-AUC over the best conventional multimodal classifier (XGBoost: AUC-ROC 0.69, PR-AUC 0.60) and substantial improvements over baseline vascular risk factor models (AUC-ROC 0.55, PR-AUC 0.28). Age-stratified analysis revealed optimal performance in younger participants (≤55 years: AUC-ROC 0.82) compared to older participants (>55 years: AUC-ROC 0.75).</p>
CONCLUSIONS: GRASP demonstrated substantial improvements in FIS risk prediction among at-risk populations through integration of multimodal data within a graph-based framework. GRASP also significantly outperformed traditional risk-factors based models in distinguishing which individuals with similar vascular risk profile will develop FIS, offering enhanced clinical risk stratification capabilities.</p>
6 Authors
- Juhi Desai
- Marwa Ismail
- Vivek Prabhakaran
- Nagesh Adluru
- Veena A Nair
- Pallavi Tiwari
1 Application
| Application ID | Title |
|---|---|
| 101084 | Quantitative tools for identifying sex-specific risk and protective factors in the progression to VCID |
Enabling scientific discoveries that improve human health