Prediction of Cardiovascular Disease Risk and Risk Factor Assessment in Diabetic Patients Based on Deep Learning and Interpretability Methods

Background Diabetes and cardiovascular diseases (CVD) are two major causes of death in the United States. Therefore, accurately predicting the likelihood of cardiovascular disease in diabetic patients and identifying high-risk individuals is of great importance. Methods and Results Our sample includes 1,080 adult diabetic patients from the National Health and Nutrition Examination Survey dataset in the United States, with ten features such as gender, age, body mass index(BMI), history of hypertension, and history of high cholesterol. This paper develops a new deep learning framework that addresses the common issue of sample imbalance in the medical field. A Attentive Interpretable Tabular Learning model(TabNet) was constructed to predict the risk of cardiovascular disease in individual diabetic patients, achieving an accuracy of 0.74 and an AU-ROC value of 0.804. SHapley Additive exPlanations methods(SHAP) were used to assess the risk factors for cardiovascular disease in diabetic patients, identifying waist circumference, age, and body mass index as the top three influencing factors. Conclusion Deep learning methods have great applicability and high accuracy in predicting cardiovascular disease risk in diabetic patients. Elderly and obese diabetic patients, especially those with abdominal obesity, are at high risk for cardiovascular disease. Diabetic patients with a history of high cholesterol also face an increased risk of cardiovascular disease.