Abstract 10945: Risk Stratification and Prediction of Acute Myocardial Infarction Events Patients of Connective Tissue Disorders

Abstract

Introduction: Connective tissue disorders (CTD) play an important role in the pathogenesis of atherosclerotic heart disease with a dynamic interplay between inflammatory and traditional cardiovascular risk factors. Hypothesis: We aim to develop, validate, and compare population-level machine learning models to predict the first acute myocardial infarction (AMI) event in CTD. Methods: We extracted patient data from the Healthcare Cost and Utilization Project (HCUP) and identified 62 demographic and clinical variables. We identified those with systemic lupus erythematosus, rheumatoid arthritis, systemic sclerosis, mixed connective tissue disorder, Sjogren’s syndrome, polymyositis, and dermatomyositis. We excluded those with prior history of myocardial infarction, known coronary artery disease, or missing key variables. We split the records randomly into training (70%) and testing (30%) datasets. Keras sequential model with Adadelta optimizer was used to compile the deep neural network model and scikit-learn classifiers were used for machine learning models. We estimated model performance based on the area under the receiver-operator characteristics curve (AUC). Results: 961,405 records were identified, and out of these, 14,961 (1.6%) had the first AMI event. The neural network model performed best in predicting AMI with an AUC:0.96 [A]. Classifier models had varying degrees of success range from 0.71 to 0.86 with Gradient Boosting Classifier, AUC:0.86 [B], being the best performing classifier, followed by Decision Tree, AUC:0.85, Logistic regression, AUC:0.84, Random Forest, AUC:0.84, Naive Bayes, AUC:0.75, Stochastic Gradient Descend, AUC: 0.71. Conclusions: A neural network model was able to predict the first AMI in people with CTD accurately. Developing and implementing machine learning models can help clinicians predict individual patient risk with a high degree of success.