Using wearable and lifestyle data to predict adverse cardiac events in patients with established coronary artery disease

Background The Secondary Manifestation of Arterial Disease (SMART2) risk score that guides secondary cardiovascular (CV) prevention relies on static clinical parameters. However, real-world recovery factors like sleep, activity, and autonomic function also affect recurrent CV risk. Digital health tools enable continuous home monitoring of these factors. Purpose To evaluate and compare the predictive performance of (1) SMART2 risk score clinical parameters, (2) validated lifestyle questionnaires, and (3) digital biomarkers from wearable and chatbot monitoring regarding adverse cardiac events during two years of follow up. Additionaly, to derive a clinical-interpretable scoring system for adverse cardiac events based on the most informative features. Methods Sixty-six patients with coronary artery disease (median age 60 years; 88% male) were monitored for 21 days after coronary revascularisation using a smartwatch and a chatbot-based self-report platform. SMART2 parameters were extracted from the electronic medical record, and lifestyle questionnaires were administered after the procedure. Four predictor sets were compared: SMART2 parameters, lifestyle questionnaires, wearable and chatbot data (activity, sleep, heart-rate metrics and self-reported lifestyle), and all combined. The primary endpoint was adverse cardiac events (CV death, re-event, new CV disease, or unplanned admission). Predictive performance was assessed using nested cross-validation, and the best model informed a simplified elastic-net scoring system. Risk tiers were prespecified on the probability scale as Low <17%, Moderate 17–25%, and High ≥25% (based on the cohort’s predicted-risk distribution and decision-curve thresholds). Results During follow-up, 26 patients (39%) experienced an adverse cardiac event. The wearable and chatbot model achieved the highest discrimination (AUC 0.75) compared with SMART2 parameters (AUC 0.63), questionnaire-based model (AUC 0.54), and their combination (AUC 0.71) (p < 0.01) (Figure 1). Decision-curve analysis showed that the wearable and chatbot model offered the highest clinical benefit for identifying patients with an intermediate predicted 2-year risk (10–30%). Top predictors included sleep restedness, sleep quality, circadian HR amplitude and acrophase, daily steps, HR mesor, HR mean, HR minimum, and grain intake. The scoring system for adverse cardiac events (nine predictors, AUC 0.71) stratified patients into prespecified moderate- (17–24%) and high-risk (25–85%) tiers (Figure 2). Conclusion In patients recovering from coronary revascularization, early post-discharge digital monitoring of sleep, activity, and circadian HR dynamics predicts adverse cardiac events more accurately than post-intervention parameters based on the SMART2 risk score or clinical questionnaire data. This scoring system for adverse cardiac events translates digital recovery signatures into a clinically interpretable framework for precision secondary prevention.