We used a smartwatch app and a smartphone app to collect data on a patient’s daily PA, heart rate, eating episodes, and mood. Data from the sensors embedded in the smartwatch are interpreted as step counts and heart rate and are sent to the smartphone via Bluetooth and then to a cloud drive via Wi-Fi or 4G. All data uploaded to the cloud are then downloaded immediately and uploaded to a server over a secured virtual private network connection through the public internet. Data in the server are analyzed and projected on the dashboard for the health coach to view. The old data are archived and then refreshed by the most recent data on the dashboard every 10 min.

The HerBeat prototype included a wrist-worn smartwatch (Moto 360 2nd Gen, Android Wear OS 2.0) and a smartphone (Samsung Galaxy S6, Android 7.0), with the app installed on both devices and a web-based dashboard for monitoring participant data. The 4 features of the prototype included (1) goal setting, (2) progress, (3) ecological momentary assessment (EMA) surveys, and (4) videos (see Figure 1). The goal setting feature allowed participants to set multiple walking goals for up to 60 min each. Study participants were tasked with setting their own PA goals in terms of the number of minutes walked. Participants were also prompted to report their readiness to begin PA and their current level of energy on a scale of 1 to 10. After setting a PA goal, each participant was sent a motivational message that encouraged exercise. Data about the participant goal setting and subsequent PA performance were monitored through a web-based dashboard in real time by a trained professional.

The progress function permitted participants to review the number of minutes walked, number of steps taken, and distance covered in miles. If participants had not completed their goal when seeking progress, they were presented with the number of minutes remaining to goal completion. If a goal was completed, the participant was sent a gender-specific graphic user interface (GUI) with a congratulatory message for achieving their goal. The EMAs are described in the Measurement section. The final feature provided participants access to 9 customized short videos, developed by the principal investigator (PI) with expertise in behavioral medicine and women’s CV health, on healthy eating behavior and on guidelines for safe PA. The app also sent two types of behavior change intervention messages. If the participant had not set a PA goal by 4 PM daily, a message prompting them to exercise was sent. If participants were proactive in setting and achieving walking goals, they were sent a positive reinforcing message. Figures 2-5 show some of the examples of GUIs of intervention screens. The dashboard was used by the health coach to remotely monitor participants’ PA (step count), heart rate, goal setting behavior, responses to the EMA surveys, and frequency of accessing the health videos. The health coach, via the dashboard, also monitored episodes of Wi-Fi and Bluetooth disconnections. The health coach sent a personalized, encouraging message to engage with HerBeat to the participant’s smartphone about once a week.

After obtaining approval from the university institutional review board and using a single-group, pretest, posttest design, we recruited participants from a university-affiliated outpatient cardiology clinic between May 2018 and August 2018. The cardiology clinic is part of an academic medical center and is staffed by faculty members who are physicians in the Division of Cardiovascular Sciences. The clinic provides state-of-the-art services and treatment options. Participants were recruited by the PI (health coach), who had access to the clinic’s electronic health record system. Potential participants who were scheduled to see their health care provider were approached for inclusion in the study after they had completed their clinic visit. Women were eligible for the study if they were aged 50 years or older; diagnosed with an acute coronary syndrome or coronary revascularization in the last 10 years; able to read, speak, and understand English; and able to participate in a PA, such as walking, unaided. We also sought verbal clearance from their cardiologist to participate in the study. Study exclusion criteria included residing outside a 50-mile radius of the study site; a psychiatric condition including dementia, delirium, or schizophrenia or actively undergoing acute psychiatric treatment; prior neurological brain disorders; current use of illicit drugs and/or chronic alcohol use at the discretion of the PI; or life-limiting comorbid conditions (eg, metastatic cancer).

The informed consent form clearly explained that study participation was voluntary and participants could withdraw from the study at any time without jeopardizing their health care. Their decision to withdraw had no impact on their relationship with their cardiologist. If they wished to withdraw study participation, they needed only to inform the PI, and no further data would be collected from that time onward.

After baseline assessment was completed, the participants were trained by a graduate student with technical expertise to use the smartwatch and the smartphone that were provided for the duration of the study. Technical questions were answered by one of the study personnel via telephone or in person. The participants were then asked to use the prototype for 12 weeks and return for a follow-up visit when data collection was completed and the hardware was returned. We did not explicitly request that participants improve their health behaviors because our primary focus was the usability of the HerBeat prototype.

Data collected from self-report questionnaires and physical assessments were maintained in a database using Research Electronic Data Capture software. The smartwatch streamed step count and heart rate data continuously between 6 AM and 10 PM daily every 3 min via Bluetooth and Wi-Fi to a Health Insurance Portability and Accountability Act–compliant server. We archived participants’ data by a study identification number to protect their identity. Daily backend jobs processing all collected information were automated on the server side. The resulting information was stored in a Structured Query Language (SQL) format for easy retrieval. The web portal dashboard was created from the SQL data to present the data to the health coach.

Data analysis techniques are applied to gain insights into the patient’s activity, heart rate, and EMA survey response data. Patients’ physical activities and EMA responses are analyzed through a decision rule-based expert system as well as by the health coach. These data are used to send standard preprogrammed intervention messages to the patients by the system and to help the health coach to customize intervention messages to send to patients through the dashboard at the right time to maximize their impact.

Descriptive statistics (eg, univariate graphical and numerical statistics, bivariate distributions, scatterplots, and counts/percentages) were generated and summarized for all study data. Paired t tests were used to compare continuously measured variables from baseline to the 12-week posttest measures. Given the small sample size, we were generally underpowered to perform parametric statistics. The α level was set at P≤.05. Qualitative field notes were summarized across all participants for themes.

A total of 11 participants signed the informed consent form, and 10 participants completed data collection. Of the 10 participants, 2, both recently experiencing traumatic life events, engaged very little with HerBeat. Most of the participants were white (8/10, 80%), married, or partnered (6/10, 60%) women with a mean age of 64 years (range 53-75 years; SD 6 years; see Table 1). The majority of participants had health insurance and an income of at least US $40,000 annually; 5 participants worked full time. All participants had CHD, with 2 participants diagnosed with a myocardial infarction and one with heart failure. Moreover, 50% (5/10) of participants had undergone a percutaneous coronary intervention, and 20% (2/10) of participants had undergone a coronary artery bypass graft surgery. None of the participants had ever attended a CBCR program.

The participants had multiple comorbidities including diabetes mellitus (3/10, 30%), osteoarthritis (4/10, 40%), and orthopedic disorders (2/10, 20%), and one participant was being treated for skin cancer (1/10, 10%; see Table 2). The participants exhibited traditional CV disease risk factors, including dyslipidemia, hypertension, physical inactivity, familial heart disease, and being overweight. Most participants had never used tobacco; former smokers had a mean of 23.75 (SD 19.3) pack-years of smoking. Participants were prescribed numerous evidence-based CV medications to treat their chronic conditions.

Table 3 summarizes the baseline and 12-week follow-up physiological and psychosocial participant characteristics. Although we observed no changes in BP, the participants had statistically significant improvements in waist circumference (P=.048), weight (P=.02), and BMI (P=.01). Furthermore, participants’ depressive symptoms significantly improved from baseline (P=.04).

SE for exercise, diet, and managing chronic illness was not statistically significantly different from baseline, although it trended in the desired direction. Participants also demonstrated nonsignificant improvements in REAP-S scores and perceived stress.

Over the course of the study, participants (n=8) collectively set 132 goals, with a mean of 16.5 (SD 17.3) goals per participant for a collective total of 3335 min of walking, with a mean of 34.72 (SD 41.68) min per participant (see Table 4) per week. Most of the walking goals were set between 9 AM and 11 AM and 5 PM and 6 PM. Over the course of the study, smartwatches allocated to the participants collectively recorded 4933 min of walking, with a mean of 22.02 (SD 35.32) min per participant per day. That is, the participants walked more than they intended when setting a goal. Over 12 weeks, each participant walked a mean of 28 days (out of a possible 90 days) and took a mean of 3718.8 (SD 3826.0) steps per day. The group responded to 830 EMA surveys and accessed 8 health educational videos 165 times during the study. The participants accessed more videos related to healthy eating behavior (137/165, 83%) than those related to PA (28/165, 17%). The participants received a total of 265 automated intervention messages based on their progress toward their goals.

The mean score on the SUS was 83.60 (SD 16.4). Participants generally found HerBeat to be easy to learn and use. They also found the functionalities to be well integrated, and they felt confident in using HerBeat. The participants did not find it unnecessarily complex or cumbersome to use (see Table 5).

They also reported not requiring the support of a technical person to use HerBeat. Only one patient required a home visit to address a technical issue. Participants’ themes derived from field notes mostly involved technical issues. The most frequent complaint was the short battery life of the smartwatch. We rectified this problem after valuable participants’ input. Some working participants found it difficult to carry both a personal phone and a study phone and respond to EMA surveys during the day. A participant who worked in a library sought permission from her supervisor to carry the study phone and respond to the EMA surveys. One participant requested taking HerBeat with her to Europe to allow her to track her activity while on vacation.

Participants’ feedback also led to the redesign of some of the GUIs of the EMA survey. Although there was minimal contact between the health coach and the participants during the 12 weeks and participants went on vacation during the study, they voiced reassurance that their progress was being monitored by the health coach via the dashboard. Participants had no adverse events during the study, and there were no issues raised about privacy concerns.

Data captured during our study suggest that at least one of the participants set a walking goal of 1 min and at least one of the participants watched no health-related videos during the study. We probed the corresponding participants during the final debriefing session about these data. For the first observation, the participant suggested that the walking goal of 1 min was mistakenly set while exploring the goal setting function at the very beginning of the study. The participant’s intention was to navigate further inside the goal setting function. Regarding the second observation, the participant chose not to watch any health-related videos because she felt well informed about these health behaviors.

The primary aim of this study was to determine the usability of our mHealth system, HerBeat, with a cohort of women with CHD before proceeding with the development of a comprehensive home-based secondary prevention intervention. Our secondary aim was to evaluate the influence of HerBeat on various psychosocial and health behaviors of the participants. To our knowledge, this is the first study to evaluate the usability of a gender-specific mHealth app for secondary prevention of CHD in women. The main finding of the study was that the system was acceptable and usable in its prototypic form. The level of engagement of participants with HerBeat was greater than anticipated, given the relatively primitive features. We developed HerBeat to avoid high data entry burden and designed gender-specific GUIs to foster engagement. Given that 80% of health-related apps are abandoned after only 2 weeks [67], the engagement of the participants with our prototype was good, particularly when they were given little prodding for using the technology. We viewed this as an encouragement to proceed with the expanded version of HerBeat, with increased involvement of the health coach.

Comparisons of user engagement with mobile apps of participants with characteristics similar to the participants in our study are difficult to make because usability was defined differently in these studies [68-71]. Some described metrics such as app usage frequency, duration, data registration, or responsiveness of the user to daily tasks. In addition to the often low participant numbers, dropouts, and short study duration, conclusions about engagement are difficult to draw. Completion of tasks within the app, such as completion of an education module, was a typical measure of use in studies with a focus on healthy lifestyle. Forman et al [68] gauged engagement by patient completion of at least one prescribed daily task. In other studies, emphasis was placed on logging medication intake or physical measurements [69,70]. The authors did not report the acceptability of a data entry requirement. We made the decision early during development, based on numerous interviews with patients, to avoid the requirement of data entry to reduce respondent burden. In an uncontrolled single-group, pretest, posttest design [72], participants were required to log daily BP measurements for 55 days; however, it was unclear whether all patients logged BP on each of the 55 days. Patients in one small study of both heart failure and CHD participants [73] appreciated medication reminders and PA information. However, they felt that daily requirements for data entry or other responses were inconvenient. Clearly, high data entry burden is a usability issue [74].

We did not see evidence of the message fatigue reported by others [75]. The fact that the participants responded to 830 EMA surveys over 12 weeks was, in our opinion, quite remarkable. Although the number of EMA survey responses was greater than expected, the responses declined over time. Educational videos on healthy eating behavior were viewed more often than videos related to physical activities, presumably because eating a healthy diet is often a daily or hourly struggle between reflex and self-control. Participants may have viewed the videos to seek assistance with making healthy eating decisions. Eating and body weight regulation is a complex process that involves both metabolic and hormonal control mechanisms and neurocognitive processes involved with memories, expectations, and evaluation of food and the consequences of eating [76]. The decisions about what and when to eat are a balance between reflexive behavior and higher-level cognitive processes. Eating can be reflexive and automatic by the mere smell of a favored food [77]. This reflexive eating can be opposed by dietary restraint of choosing a healthy food that involves higher-level cognitive processes to counter the power of tempting environmental stimuli [78].

On the basis of decision rules related to participants’ responses employed in HerBeat, some intervention messages were deployed more frequently than others. Most participants exceeded the walking goals they set. In other words, most of the time participants did not abruptly stop their walk after achieving their PA goal but rather exercised beyond the goal. We hypothesize that this may reflect low SE when setting the goal, followed by greater confidence when they surpassed the goal. Although we did not set a target for time spent walking or for step count, the participants’ daily step count was relatively modest. A common goal of 10,000 steps per day has been perpetuated by the lay press and is often used as the default by software programs on wearables and smartphones [79]. In the United States, the average number of steps accrued daily (measured by smartphones) is approximately 4800; worldwide, it is approximately 5000 [80]. There is sparse data on the number of daily steps needed for health [81,82] or clinical outcomes and mortality [83]. In the Women’s Health Study, a cohort of 16,741 women with a mean age of 72 years wore accelerometers to measure their steps per day over 7 days [84]. Women who averaged 4400 steps per day had significantly lower mortality rates during a follow-up of 4.3 years compared with the least active women who took approximately 2700 steps per day. As more steps per day were accrued, mortality rates progressively decreased before leveling at approximately 7500 steps per day [84].

Although we did not expect participant health behaviors, SE, perceived stress, or depressive symptoms to improve with a limited functionality prototype, we nonetheless observed significant reductions in waist circumference, weight, and BMI as well as reduced depressive symptoms after study participation. These improvements were unexpected because the research team had minimal contact with the participants during the 12 weeks, and we did not prompt them to set goals for walking. Participants reported minimal positive changes in their SE for exercise, diet, or managing chronic illness, but scores nonetheless trended in the expected direction. From baseline to the 12-week follow-up, there was a modest increase in the mean minutes of moderate-intensity exercise. There were no reductions in participants’ time spent sitting. The primary purpose of this study was to examine the usability of the system, and secondarily, to examine behavior change after the 12-week study. With a more robust version of the system, we will examine the effectiveness of the system in a randomized clinical trial.

Our findings must be balanced with the limitations of the study. First, this was a small convenience sample from a single study site. With multiple statistical testing, we may have capitalized on chance findings. The generalizability of the findings is limited to women with CHD. Furthermore, we used a nonexperimental design without a control group. Second, this was a usability test of a minimal viable product with minimal contact from the research team. Third, our study was not long enough to evaluate any sustained behavior change. A randomized controlled trial with a larger sample is needed to better understand the optimal way of providing secondary prevention through digital health interventions. However, the aim of our study was to examine the usability, viability, and user requirements for developing a more comprehensive mHealth intervention for women with CHD.

This usability study has encouraged us to develop a comprehensive mHealth behavior change intervention that targets PA, healthy eating, stress management, medication adherence, and smoking cessation. Such a home-based system is not intended to replace CBCR but rather to offer behavior change theory–based interventions in real time to individuals as they live their lives, particularly for those who cannot access CBCR. Evidence for the effectiveness of self-management of multiple health behaviors for improved outcomes will require a larger, randomized controlled trial of a longer duration. A pilot randomized study of the next version of HerBeat is currently underway.

Our formative evaluation of HerBeat helped us to refine our design strategy for the next trial. We plan to incorporate more provision for the user to communicate with the health coach as a group as well as individually. We have expanded the EMA surveys to target more behaviors of relevance to CV health. With feedback from the participants, we have developed many more meaningful BCTs that are deployed using decision rules in response to the participants’ responses to the EMA surveys. We have enhanced the dashboard to be more visually usable by the health coach. Finally, we have resolved some of the problems with the wearable sensor by implementing the use of a different smartwatch that has a long battery life.

CV disease remains the leading cause of death worldwide. Healthy lifestyle behaviors are critical to CV health. We designed a mHealth prototype specifically for women with CHD to assist them with behavioral self-management. The participants found the prototype easy to use over 12 weeks and were receptive to setting walking goals and responding to EMA surveys. Mobile technology is an innovative and scalable approach to reducing the risk factors of CV disease, but evidence related to acceptability remains limited. Our study has contributed to the limited data on the usability of mobile apps for CV disease self-management.