Effects of an AI-powered eHealth intervention on physical capacity and self-management in coronary artery disease: the TIMELY multicenter RCT
B Schmitz, R Garbsch, V Tsakanikas, M Habibovic, C Pena Gil, J R Gonzalez Juanatey, M Schmidt, H Malberg, H Schmidt, D I Fotiadis, W Maerz, D Gatsios, W J Kop, J Bosch, Frank MoorenAbstract
Background
eHealth solutions offer promising avenues for optimizing personalized care and supporting disease management in patients with coronary artery disease (CAD). TIMELY represents the first artificial intelligence (AI)-driven eHealth intervention grounded in cardiac rehabilitation (CR) principles and integrated with Internet of Things (IoT) technology.
Purpose
To report the outcomes of the TIMELY RCT.
Methods
TIMELY is a multicenter RCT conducted in Germany, Spain, and the Netherlands, enrolling 360 CAD patients. Participants randomized to the intervention group (IG, n=180) received a 6-month app-based behavior change program as part of phase III CR aftercare, while control group (CG) participants received usual care. Assessments occurred at baseline, 6, and 12 months. The TIMELY platform supports behavior change and self-management by integrating clinical and psychosocial data, evaluating risk, and delivering personalized feedback via app-based prompts. Physical activity engagement is facilitated through a chatbot and individualized exercise prescriptions derived from weekly activity profiles, utilizing Ecological Momentary Assessment (EMA). Participants were equipped with an activity tracker, blood pressure monitor, and 3-channel Holter monitor, all connected to an integrative cloud-based platform. AI algorithms optimize CAD risk prediction and behavioral recommendations, with data displayed on a dashboard accessible to case managers. Primary outcomes were changes in mortality risk (assessed via a composite biomarker score) and 6-minute walking test (6MWT) distance. Secondary outcomes included physical fitness, physical activity levels, dietary habits, body weight, smoking cessation, medication adherence, and psychological stress.
Results
Mean participant age was 60.5±9.2 years; 34% had a history of MI, and 21% were women. At 6 months, 161 (89%) IG and 156 (87%) CG participants completed primary outcome assessments. Twenty-eight serious adverse events occurred, none attributable to the intervention. Intention-to-treat analysis revealed no significant between-group differences for primary outcomes including biomarker-based risk assessment. However, per-protocol analysis demonstrated that the IG maintained approximately 7% superior 6MWT performance over 12 months (IG=567.5±90.3 m vs. CG=529.6±111.7 m; p<0.05). The IG also exhibited greater improvements in physical fitness (time×group interaction, p=0.027), with higher values at 6 months (IG=144.5±39.6 W vs. CG=130.9±44.0 W) and 12 months (IG=141.9±45.2 W vs. CG=130.3±45.2 W). Medication adherence (MARS-5 score) remained significantly higher in the IG at 12 months (interaction, p=0.0071). No other significant between-group differences were observed for secondary outcomes.
Conclusions
The AI-driven eHealth intervention TIMELY provides personalized support for CAD patients during phase III CR and demonstrates promise for improving physical exercise capacity.