Harnessing Artificial Intelligence and Deep Learning to Determine a Coronary Artery Calcium Estimate in Patients with No History of AtheroSclerotic CardioVascular Disease (HIDDEN-ASCVD) Study

PROJECT SUMMARY/ ABSTRACT Data science advances have led to the emergence of opportunistic imaging screening – applying algorithms to screen for conditions from imaging performed for other purposes. Opportunistic screening for coronary artery calcium (CAC) is an ideal test case. CAC is the strongest predictor of atherosclerotic cardiovascular disease (ASCVD) and knowing one has CAC can be a powerful motivator of preventive behavior. However, <1% of potentially eligible patients receive electrocardiogram (ECG)-gated computed tomography (CT) scans to measure CAC. CAC can be identified on non-gated chest CTs, of which 19 million are performed annually in the U.S. We previously demonstrated that a deep learning (DL) algorithm can accurately quantify CAC on non- gated, non-contrast chest CTs, and notifying patients and providers of the presence of opportunistic CAC significantly increased statin prescription rates compared to usual care. The HIDDEN-ASCVD study will extend validated DL-CAC methods to non-cardiac, contrast-enhanced CT scans and include a factorial randomized controlled trial (RCT) to identify the optimal opportunistic CAC notification strategies to maximize statin prescription rates and patient and provider acceptability within Kaiser Permanente Northern California, an integrated health system providing care to >4.5 million members with broad racial and ethnic diversity. In Aim 1, we will validate the diagnostic accuracy of the existing DL-CAC algorithm and adapt the algorithm to contrast-enhanced CT scans using self-supervised DL and paired non-gated and gated CT scans. We will also describe the epidemiology and outcomes of opportunistic CAC in this diverse health care system. In Aim 2, we will identify 5,760 adults with DL-CAC on chest CT, without known ASCVD, and not receiving statins and efficiently test multiple notification strategies based on the MINDSPACE behavior change principles and the Multiphase Optimization Strategy (MOST) framework for intervention optimization. Through a factorial RCT design, we will determine the most effective notification strategy that maximizes statin therapy and minimizes patient-reported anxiety. Co-design focus groups of providers and patients that received notifications will then provide input on serial adaptation of the strategy based on their experience. The adapted optimal strategy will be tested in a single-arm validation study to determine an unbiased estimate of the net effect of the intervention on statin initiation and patient acceptability. The RCTs will be enriched for historically marginalized racial and ethnic groups to provide generalizable knowledge. The HIDDEN-ASCVD study will be paradigm shifting by identifying the optimal behavioral science-driven notification strategy for opportunistic CAC balancing the benefit of statin therapy with the perceived risks of health-related anxiety. This research will motivate future studies designed to implement opportunistic CAC screening and notification on a health system-wide scale reducing the overall burden of ASCVD and mitigating documented health disparities.