Mortality Prediction of Patients With Cardiovascular Disease Using Medical Claims Data Under Artificial Intelligence Architectures: Validation Study (Preprint)

Abstract

BACKGROUND As stated by WHO, Cardiovascular disease (CVDs) are the number 1 cause of death globally, which means more people die annually from CVDs than from any other cause. An estimated 17.9 million people died from CVDs in 2016, representing 31% of all global deaths. Of these deaths, 85% are due to heart attack and stroke. In this study, we present a benchmark comparison of various Artificial Intelligence (AI) architectures on predicting mortality of CVD patients using the structured medical claims data. OBJECTIVE This study mainly aims to support health clinicians to accurately predict mortality among patients with CVD using only claims data before a clinic visit. METHODS The used dataset was joined from Medical Benefits Scheme (MBS) and Pharmaceutical Benefits Scheme (PBS) service information in the period between 2004 and 2014, released by the Department of Health Australia in 2016. It includes 346,201 records corresponding to 346,201 patients. A total of five AI algorithms including four classical Machine Learning (ML) algorithms (Logistic Regression (LR), Random Forest (RF), Extra Trees (ET) and Gradient Boosting Trees (GBT)) and a deep learning algorithm which is a densely connected neural network (DNN) were developed and compared in the study. In addition, due to the minority of ‘deceased’ patients in the data set, a separate experiment using Synthetic Minority Oversampling Technique (SMOTE) was conducted to enrich the data. RESULTS Regarding model performance, in terms of discrimination, GBT and RF are the models with highest AUROC (97.8% and 97.7% respectively), followed by ET (96.8%) and LG (96.4%) while DNN is the least discriminative (95.3%). In terms of reliability, LG predictions are the least calibrated compared to those of four algorithms. In this study, despite increasing training time, SMOTE is proved to further improve model performance of LG while other algorithms, especially GBT and DNN, work well with class imbalanced data. CONCLUSIONS Compared to other research in the clinical literature involving AI models using claims data to predict patient health outcomes, our models are more efficient since we only utilize a smaller number of features but still achieve high performance. And this study could support health professionals to accurately choose AI models to predict mortality among patients with CVD using only claims data before a clinic visit.