Integrating Multimodal Patient Data into Attention-Based Graph Networks for Disease Risk Prediction

In the era of digital medicine, electronic health records (EHRs) encompass a vast array of patient data from diverse sources. Harnessing this readily available information is crucial for personalized medicine and predictive healthcare. This study explores the potential of integrating multimodal data into an attention-based graph network for enhancing disease risk prediction. Leveraging the intrinsic structure of medical data, we propose a novel method called MiGAT, which fuses information from various modalities, including clinical events and sequencing data. By applying graph attention networks (GAT), our model effectively captures the intricate inter-relationships among patients, particularly emphasizing genetic similarity. This enables our model to assess and prioritize disease risks based on the genetic closeness of patients. Results indicate a substantial improvement in predictive accuracy compared to baseline models, reaffirming the potential of our multimodal integration strategy. Our findings underscore the importance of synergizing diverse EHR data sources through advanced computational models for better-informed clinical decision-making and more effective patient care.