Can machine learning algorithms accurately predict multi-class coronary artery calcium (CAC) risk categories from electronic health record data in asymptomatic patients?
Machine learning algorithms using routine electronic health record data can effectively predict coronary artery calcium risk categories in asymptomatic patients, potentially aiding in non-invasive cardiovascular risk stratification.
Coronary artery calcium (CAC) is an established surrogate marker for coronary atherosclerotic disease (CAD) burden. The CAC score is also an independent predictor of adverse events with significant incremental prognostic value over traditional/clinical risk stratification algorithms. The objective of this study was to examine the prognostic ability of Machine learning (ML) based algorithms to predict multi-class CAC (0: normal; 1-100: low risk CAD; 101-400 Intermediate risk CAD; >400 severe/high risk CAD) from available electronic health record (EHR) data. A retrospective observation study of 60,923 asymptomatic patients with clinically evaluated CAC score along with sixty five clinical and laboratory parameters were included in developing the ML algorithm (data split into 70% training and 30% test). In addition, a separate cohort of 7,552 patients was used to externally validate the developed ML algorithm. Classification performance was assessed using the area under the receiver operating curve (AUC). The prediction algorithm derived from the ML method showed high predictive value for CAC risk category.
Kolli et al. (2019) studied this question.