MedGate-Fusion improves stroke risk prediction from records
MedGate-Fusion: Integrating First-Encounter Semantic Narratives and Physiological Biomarkers for Prospective Stroke Risk Stratification
Artificial Intelligence
Summary
Predicting who might have a stroke is hard because early warning signs are hidden in many types of health information. The authors created MedGate-Fusion, a computer system that combines doctors’ notes from a patient’s first visit with common health measurements to better assess stroke risk over five years. They removed direct mentions of stroke in the notes to avoid cheating the system. This approach uses lots of medical records to find early clues that might normally be missed.
What this means in practice
- •For primary care data teams: Improve five-year stroke risk models by combining first-visit doctor notes with routine biomarker data for earlier and more accurate predictions.
- •For health it developers: Develop clinical decision support tools that integrate multi-modal patient data to identify high stroke risk patients proactively.$Commercial implications: Enables creation of predictive software sold to healthcare providers for stroke prevention.
Authors
Hemn Khdr, Mohammad Noaeen, Karim Keshavjee, Aziz Guergachi, Zahra Shakeri
Abstract
Prospective stroke risk stratification in primary care is challenging because early risk signals are distributed across routine biomarkers and unstructured clinical narratives. We propose MedGate-Fusion, a multi-modal gated architecture that integrates transformer-based embeddings of first-encounter narratives with ten routinely recorded risk markers. We used electronic medical record data from the Canadian Primary Care Sentinel Surveillance Network (CPCSSN). Starting from 808,921 encounter-level observations, we constructed a first-encounter cohort and retained 102,736 unique patient records with non-empty narratives and sufficient data to evaluate a five-year stroke outcome. To reduce explicit target leakage from diagnostic mentions in notes, we applied dictionary-based redaction of stroke-related terms prior to semantic encoding.