Predicting Cardiovascular Disease Risk Using Transformer Networks and Electronic Health Records
Dilli Ganesh, T J Nandhini, Prashant Johri, Mahmoud Mahfuri
Abstract
cardiovascular disease (CVD) is one of the biggest killers globally, and it is crucial to predict CVD risk at the initial stage to reduce the mortality rate and morbidity of the disease. Thus, this research develops a deep learning approach based on the transformer architecture to predict CVD risk by analyzing EHRs. It is worth noting that conventional models are somewhat challenged with increased amounts and chronological characteristics inherent in-patient data. To cope with this challenge, we utilize transformer models that can model dependencies in the data for EHR and capture complex patterns. Further, the proposed approach converts structured and unstructured data, such as patient demographics, laboratory tests, prescriptions, and notes from electronic health records, into a risk assessment model. The transformer network successfully captures numerous dependencies of the features encountered in the medical field and trends in the medical history compared to traditional models like logistic regression or recurrent neural networks. Based on experimental results, the proposed framework exhibits better predictive performance, reliability, and interpretability to be a useful scientific tool for usage in clinical decision-making. By using deep learning with EHR analytics, this framework helps in early risk prediction for disease diagnosis, which has a positive impact on disease management. It also advances accuracy, extendibility, and flexibility, which paves the way for new AI-facilitated models in the healthcare environment. The future groundwork will entail improving the efficiency of the model and confirming its effectiveness in various populations as well as real-life practice.