We conducted a retrospective cohort study of adult participants enrolled in CR at University of California, San Francisco (UCSF) CR from May 22, 2020 (when use of the mobile app began) until May 21, 2022. For the primary study analysis, we included all participants aged ≥18 years who were enrolled in CR and downloaded the mobile app. Since the mobile app was only available in English, the study only included English-speaking participants. For the primary outcome analysis, we included only participants with 6MWD measurements at both enrollment and completion. For participants who enrolled in CR multiple times during this period (n=16), we analyzed the first episode of participation in CR.

Participants enrolled in CR at UCSF can participate in in-person, telehealth, or hybrid CR [26]. Use of a mobile app (Chanl Health; Figure 1) is offered to all CR patients in all modes of participation but is optional. The mobile app allows patients to log exercise activities and vitals, view educational materials, set medication reminders, and message CR staff. CR staff include doctors, nurses, exercise physiologists, pharmacists, a nutritionist, and a mental health provider. Concomitantly, CR staff can view patient logs and activities. Eligible patients were informed of the mobile app and taught how to use the app by a CR staff member using task-based training on entering vital signs, entering an exercise session, viewing an education module, messaging with providers, and setting up medication reminders (if desired). There was no cost to the patient associated with using the app.

Patient app interaction was retrieved from mobile app use logs. Patient app interactions included exercise logs, vital sign logs (eg, BP, weight, blood glucose, and oxygen saturation), education material views, and chat with providers. Data were manually entered into the app by patients. Total app interaction was measured as the sum of total exercise logs, vital sign logs, and education material views. For analysis of BP measurements recorded in the app, when multiple measurements were present with the same date and time, BP values were averaged.

The primary outcome was change in functional capacity between enrollment and completion of the CR program, measured by change in 6MWD. At enrollment and completion, participants completed a 6-minute walk test on a standard course at the UCSF CR center [27]. The secondary outcome was change in BP between enrollment and completion of the CR program. At enrollment and completion, systolic and diastolic BP were measured with a standard BP cuff at the UCSF CR center. Measurements of 6MWD and BP were recorded by CR staff in CR Quality Improvement data and entered into the electronic health records, which served as the data sources for outcome measurements.

Participant characteristics were obtained from structured electronic health record data, including age, sex assigned at birth, race, ethnicity, insurance, SES by zip code, and primary reason for CR referral. Sex assigned at birth was used since it was the only available data. Insurance was categorized as either private or public (eg, Medicare, Medicaid, and other city- or state-sponsored health plans). SES by zip code was established using the UCSF Health Atlas, where zip codes across California are rated between 1 (lowest quintile) to 5 (highest quintile) of socioeconomic advantage [28]. The UCSF Health Atlas uses neighborhood data system from a census tract-level data accounting for 7 factors (eg, percentage of people aged ≥25 years with college, high school, or less than high school education level; percentage of people with blue-collar jobs; percentage of people aged ≥16 years in the workforce, but without a job; and percentage of people below 200% of the federal poverty level, median household income, and median rent) [28]. The primary diagnosis associated with the CR referral was used as the primary reason for referral, categorized as coronary artery disease, valvular disease, heart failure, or other diseases. If a coded referral diagnosis was not present, reason from referral was determined from chart review of the patient’s individualized treatment plan for CR.

Clinical CR measurements at enrollment and completion were obtained from CR Quality Improvement data, including BMI, waist-to-hip ratio, General Anxiety Disorder (GAD)-7, and Patient Health Questionnaire (PHQ)-9. BMI was calculated through weight and height recorded in the health electronic record, and waist-to-hip ratio were measured in person by CR staff using standardized methods. PHQ-9 was scored 0-27, with higher scores representing more severe depressive symptoms; GAD-7 was scored 0-21, with higher scores representing more severe anxiety symptoms. Laboratory test values were obtained from structured electronic health record data, including hemoglobin A_1c, total cholesterol, low-density lipoprotein, high-density lipoprotein, and triglycerides. Enrollment laboratory values were defined as the measurements prior to the participant’s first day in the CR program. Completion laboratory values were defined as a recorded measurement 3 to 6 months after enrollment.

We included all patients within the study period who fit our inclusion criteria; therefore, sample size was not calculated a priori. Patient characteristics, app interactions, and clinical test results were summarized using descriptive statistics, mean (SD) for normally distributed characteristics, and median (IQR) for nonnormally distributed characteristics. Changes in clinical characteristics were calculated by subtracting enrollment values from completion values and compared using a paired t test (2-tailed). Linear regression was used to evaluate the association between change in 6MWD and total app interactions, change in 6MWD and total exercise logs, change in BP and total app interactions, and change in BP and total BP logs. We also examined, among all people enrolled in CR who downloaded the app, whether total app interactions were associated with completion of CR using logistic regression. Regression models were adjusted for confounders including age, sex assigned at birth, race, ethnicity, SES by zip code, insurance, and primary diagnosis. Significance was considered P<.05. With our sample size, we would have >70% power to detect a difference of 25 meters (the minimum clinically important difference) per 100 app uses [29]. All statistical analyses were performed using Stata (version 17.0; StataCorp).

This study was approved by the UCSF Institutional Review Board (21-33754), who waived the requirement of informed consent for this retrospective study. All identifiable data were stored on secure cloud servers. From identifiable raw data, deidentified data sets were created for analysis. There was no compensation to patients.

In this study of 107 patients participating in CR with the use of a mobile app, we found no association between the amount of app use and improvement in clinical outcomes, including 6MWD and BP. We also did not observe an association between app use and completion of CR.

Previous research has shown that CR programs using mobile apps can achieve similar or improved outcomes compared to traditional CR programs [26]. However, there has been a lack of published research on associations between patient interaction with CR mobile apps and clinical outcomes. In their study, Widmer et al [30] found that digital health interventions in which patients interact with the app through reporting dietary and exercise habits showed a nonsignificant reduction in cardiovascular-related rehospitalizations [30]. In a randomized study, Rosario et al [31] found no statistically significant difference in 6MWD, waist measurement, RPE, and resting heart rate comparing hospital-based CR patients with and without a smart phone app, portable BP monitor, and weight scale [31]. In our nonrandomized study, we found no association between total app interactions and change in functional capacity (change in 6MWD) or change in BP.

In a systematic review of 6 randomized control trials of CR mobile apps, Tuttle et al [32] found inconsistent clinical results. Analysis suggested improved outcomes with apps that incorporated automated data collection, such as automatic step counters, automatic information logs, correctional goal setting, and real-time feedback [32]. Meanwhile, apps without automatic data collection and feedback were not successful [32]. The app used by patients in this study required manual data entry and did not provide automated feedback; moreover, its use was not associated with improved outcomes, which is consistent with the systematic review findings.

Nonetheless, with the larger use of smartphones comes an inherent shift of mobile app incorporation into clinical care. In addition, there is a high interest of mobile app use for health care delivery among patients and clinicians [9,33]. With this interest, technologies such as mobile apps can provide flexibility, scalability of care, accessibility of information, and communication among teams and patients [16,20-23]. As we have learned during the COVID-19 pandemic, mobile apps provided a mechanism to facilitate CR care and information exchange between patients and providers in a hybrid and remote setting, with no significant difference in clinical outcomes from in-person CR programs [26]. Patient experience is considered an important predictor for health care quality affecting clinical outcomes, medical adherence, and hospital retention [34-37]. Similarly, provider experience, although less studied, has been shown to affect patient outcomes and successful integration of programs [36,38]. However, little research is done on the role of mobile apps in improving patient and provider experiences [39,40]. This study did not capture measures of patient and provider satisfaction with mobile app use. Future studies should examine the impact of mobile app use on patient and provider experience.

Certain limitations must be considered. Generalizability may be limited since the cohort was enrolled and completed the program during the COVID-19 pandemic. Other factors that may affect generalizability include underrepresentation across sex assigned at birth, race, ethnicity, and SES since the cohort was mostly male, White Caucasian, non-Hispanic, living in neighborhoods associated with high SES, and with private insurance. SES by zip code incompletely captures individual-level SES factors.

This study only examined mobile app use among people who chose to download a mobile app and cannot account for differences between this population and people who did not choose to download the mobile app. Moreover, the patient population was only from a single center with a limited sample size. Additionally, in this observational study, using data collected during the routine course of health care, we could not account for all potential unmeasured confounders.

In a population of patients participating in CR and using a mobile app, this study found no associations between total app interactions and changes in functional capacity, total exercise logs and changes in functional capacity, total app interactions and program completion, total app interactions and changes in BP, and total BP logs and changes in BP. Mobile app use was not associated with completion of the program. Given the lack of association between mobile app use and clinical outcomes, we encourage further study of the impact of mobile app use on patient and provider experience. Importantly, this research should be conducted in a diverse population to address existing disparities.