We performed a randomized controlled trial the rationale and design of which, including the sample size and power calculation, have previously been described in detail [9] and are summarized in the Electronic Supplementary Material [ESM] File S1. Patients enrolled to CR at Canisius Wilhelmina Ziekenhuis (Nijmegen, the Netherlands) were eligible if they were > 18 years, had a diagnosis of coronary artery disease, and were able to understand and perform the study procedures. Exclusion criteria included inability to use the smartphone application, contraindications, or severe orthopedic problems that restrict PA. Informed consent was obtained from all patients prior to participation, and the Cardiac RehApp trial protocol adhered to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a priori approval by the Medical Ethics Committee of the Radboud University Medical Center (NL72182.091.19). The CONSORT checklist is available in ESM Table S1.

Inclusion and baseline measurements took place prior to the CR program. Thereafter, they were randomly allocated (1:1) to the intervention or control group using a computerized algorithm with allocation concealment and random block sizes (4-6) stratified for index diagnosis (acute events versus elective procedures). Directly after the 6‑week program, post-CR measurements were performed. Only outcome analyses were blinded.

All patients received usual care consisting of a ~ 6-week comprehensive center-based CR program. The program included consultations focusing on lifestyle improvement, medication adherence and psychosocial well-being. Additionally, patients participated in a supervised exercise program and, if appropriate, a dietary module, psycho-educative prevention module and psychological module.

Participants in the intervention group received a 6-week smartphone training program alongside usual care CR. The participants were instructed to perform daily physical activities in their home situation using the Virtual Training mHealth smartphone application (Welfaster ApS, Denmark) [9]. The personalised home-based program was configured by the research team to ensure safety and feasibility during the trial. The application contained different training programs, including strength and aerobic exercises (e.g., squats, walking, cycling) and synchronous instructions by video, text, and audio. At baseline, the researcher and participant set individual goals based on preferences (e.g., biking or walking) and physical status (i.e., physical limitations, previous exercise experience, age), which were converted into a personalized home-based training program by the researcher. The smartphone program stimulated daily PA through goal-setting, personalised exercise prescriptions, and automated reminders in the home environment. Progression and feasibility were monitored via in-app messaging, and participants could contact the research team through the interactive platform when needed. These elements reflect common behaviour-change techniques such as action planning, prompting, and feedback, which are known to support increases in PA levels [10].

All outcomes were measured at baseline and Post-CR. PA patterns were objectively measured during 8 days with a validated thigh-worn accelerometer (ActivPAL micro, PAL technologies, Glasgow, United Kingdom) [11] and analyzed by a modified version of the script of Winkler et al. [12]. Moderate-to-vigorous PA (MVPA), light intensity PA (LIPA), and sedentary time were expressed in h/day. Step count was expressed as steps/day.

The Åstrand test was performed to examine physical fitness [9]. During the test, heart rate was continuously monitored (Polar V800, Kempele, Finland), and the Borg score was reported during the third and sixth minutes. Patients using heart rate-lowering medication (e.g., beta-blockers) followed an adjusted test [9]. Hand grip strength was assessed in the dominant hand using a dynamometer (Jamar, Jackson, MI, USA). Three measurements were performed, and the maximum strength effort (kg) was used for analysis.

Quality of life was measured using the validated HeartQoL questionnaire [13], and cardiac anxiety was measured using the validated Cardiac Anxiety Questionnaire (CAQ) [14].

All statistical analyses were performed using R version 4.2.1 with packages “lme4” and “Lmmstar” (visualization purposes), and “emmeans”. All tests were two-sided, confidence intervals (CI) were at the 95% level, and P‑values < 0.05 were considered statistically significant. Continuous normally distributed data were presented as mean ± standard deviation (SD), continuous not-normally distributed data as median [interquartile range] (IQR) and categorical variables as number (%). All data were visually inspected for normality.

Primary and secondary outcome analyses were performed on a modified intention-to-treat basis, including all participants with at least one primary outcome measurement at baseline or post-CR. A baseline-adjusted (i.e constrained) linear mixed-model analysis was used to handle missing data and to avoid baseline imbalances between treatment arms [15]. Time (categorical) and the interaction time*group were included in the model.