Multi-modal Predictive Model for MACE Risk Estimation in Patients with Migraine

Clinical research has indicated a link between mi-graine related characteristics and risk of adverse cardiovascular and cerebrovascular events. We propose a fusion predictive model with graph convolution as backbone, for estimating risk of such events among migraine patients using their neurology consultation notes and history of prescription medication. We propensity matched our dataset on all known cardiovascular disease risk factors so that our model can learn relationships be-tween migraine related features and cardiovascular disease. Our fusion based risk prediction model outperformed all comparative models in terms of prediction accuracy. Post-hoc analysis shed light onto migraine characteristics related to vascular disease risk.