Cardiovascular disease (CVD) is the leading cause of death for adults in the United States [1], and 49.2% of adults (126.9 million) were living with some form of CVD in 2018 [2]. The primary type of CVD that drives CVD-related morbidity and mortality is atherosclerotic CVD (ASCVD), which is characterized by buildup of plaque in arteries and includes conditions such as coronary heart disease, cerebrovascular disease, peripheral artery disease, and aortic atherosclerotic disease [3]. Despite effective evidence-based strategies for preventing or managing ASCVD, millions of Americans have risk factors that place them at increased risk for having a cardiovascular event [1]. Thus, the prevention and management of CVD is a top public health priority.

The primary method of preventing CVD is promoting a healthy lifestyle throughout life. The American College of Cardiology (ACC) and American Heart Association (AHA) guidelines on the primary prevention of CVD [3] recommend that all adults should maintain a healthy weight, consume a healthy diet, engage in regular physical activity, avoid tobacco use, and practice good sleep hygiene. These lifestyle habits are also recommended for the secondary prevention of CVD. The ACC and AHA guidelines also recommend that adults who are aged 40 to 75 years and are being evaluated for CVD risk should undergo a 10-year ASCVD risk estimation at the initial evaluation and at each subsequent follow-up appointment.

Despite these guidelines, maintaining a healthy lifestyle is difficult for many adults [4]. As such, a key method of CVD prevention and management is participation in behavioral interventions focused on lifestyle modifications. The US Preventive Services Task Force advises that behavioral interventions focused on diet and physical activity provide cardiovascular benefits to adults both with and without known CVD risk [5-7]. Interventions and lifestyle change programs can be focused on primary prevention of CVD and secondary prevention after a cardiac event and can be delivered by a variety of modalities (eg, in person, by telephone, and digitally) but have traditionally been delivered in person [7,8].

Although in-person programs can be very effective for increasing healthy behaviors [5,6], they also present many challenges. Such programs involve significant human-to-human contact (eg, nurses and coaches), making them expensive to implement. In addition, requiring in-person attendance makes it difficult or inconvenient for some individuals to attend [9]. These barriers to implementation and participation make it challenging to scale lifestyle interventions to large populations. Given the large number of adults in the United States at risk for CVD, there is a clear need for scalable and accessible lifestyle change programs for preventing CVD.

There is a growing number of digital lifestyle change programs for primary and secondary CVD prevention [10-12] that can be delivered through a variety of digital modalities (eg, SMS text messaging, smartphone apps, and web-based tools). Digital health programs can avoid many of the challenges posed by in-person programs, offering benefits such as on-demand support, access after standard business hours, lower cost, and increased scalability to larger populations [8]. Furthermore, digital health programs for primary and secondary CVD prevention have been shown to improve individual behaviors and show promise in delivering care that is accessible, cost-effective, and patient focused [13].

There has been prior research on digital health interventions for the prevention of CVD [13-15]. A meta-analysis by Widmer et al [15] of digital health interventions for both the primary and secondary prevention of CVD summarized many of these findings in an assessment of 24,054 participants across 51 studies. This meta-analysis found that digital health interventions significantly reduced negative CVD outcomes, such as CVD events, hospitalizations, and all-cause mortality. Widmer et al [15] also observed reductions in risk factors, including weight, BMI, and Framingham risk score, compared to usual care. A more recent analysis that focused on health behaviors also indicated that digital interventions can improve physical activity, diet quality, and medication adherence [16]. Overall, these studies show a net benefit of digital health interventions on CVD outcomes and risk factors.

As demonstrated by these prior studies, there is evidence for the benefits of digital health interventions for the primary and secondary prevention of CVD. However, there has been considerable variation in the delivery mode of available digital solutions, namely in the involvement of humans as coaches or health care providers. A common feature across most digital health programs is the inclusion of human coaching or care provision delivered via telephone calls, SMS text messaging, or email [17]. Moreover, a recent review of the literature indicated that very few (4/31, 13%) cardiovascular health–related digital health interventions are fully automated, requiring no specific personnel [11]. Human interaction elements in digital health, similar to in-person programs, require significant resources, limiting their scalability [18]. As a result, this pilot study examined a fully digital solution (ie, no human-to-human coaching) for primary and secondary CVD prevention called Lark Heart Health.

The primary purpose of the Heart Health program is to help participants make and maintain meaningful evidence-based healthy lifestyle changes, learn about their CVD risk factors, and acquire appropriate self-management skills. The Lark Heart Health program is an artificial intelligence (AI)–driven mobile coaching solution that provides synchronous CVD risk counseling anytime, anywhere. The Heart Health program is for primary prevention in individuals without a major ASCVD event or secondary prevention for those in a stable condition after an ASCVD event that occurred ≥6 months prior. Through an ASCVD Risk Estimator survey, evidence-based educational curriculum, and real-time personalized coaching, the Lark Heart Health program provides members with the tools they need to make meaningful lifestyle changes that can help them better prevent and manage ASCVD and coronary artery disease through heart health–specific digital nutrition coaching; medication adherence counseling; and personalized guidance on weight management, activity, stress, and sleep. In addition, one of the goals of the Heart Health program is to provide a user-friendly platform for participants to complete physical and mental health screening surveys that can be time consuming and burdensome for both patients and providers in primary care settings [5,19]. This is particularly important because the completion of health screeners is associated with improved patient outcomes and clinical care [20].

Experts designed the Heart Health program in accordance with guidelines from the AHA; the ACC; and the National Heart, Lung, and Blood Institute. The American Diabetes Association 2020-2025 Dietary Guidelines for Americans, the American College of Sports Medicine, and the American Academy of Sleep Medicine informed the additional nutrition, activity, and sleep recommendations. The development of the Heart Health program was a collaboration between Lark Health and Roche Diagnostics and fits within the World Heart Federation Roadmap for Digital Health in Cardiology [21]. Full details on Lark’s other digital programs have been previously reported [22,23].

Coaching in the Lark Heart Health program is completely powered by AI. The fully digital interface enables maximum program scalability and the advantage of being accessible to participants 24/7 with synchronous coaching, feedback, and encouragement delivered on demand. Feasibility and acceptability studies are a key component of evidence generation for new programs of any kind and provide a crucial preliminary step before conducting randomized controlled trials and commercialization. Similar published studies provide examples of the overall scope, design, and execution of digital programs for cardiovascular health, CVD, and other chronic diseases [24-26]; for example, Lunde et al [24] conducted a feasibility study of an app for cardiac rehabilitation, setting predefined criteria for success on participant satisfaction, recruitment rates, and app adherence to help predict possible pitfalls before conducting a large randomized controlled trial. Despite these inherent advantages, there is little evidence supporting the feasibility and user acceptability of fully digital programs for cardiovascular health. Thus, the goal of this pilot study was to provide preliminary evidence that participants are willing and able to participate in the Heart Health program.

This is a real-world single-arm 3-month pilot study of the feasibility and acceptability of the Heart Health program. The primary objectives and corresponding success criteria determined a priori are outlined in the following subsections.

As part of the Heart Health pilot study, we also evaluated the predictors of recruitment and enrollment as well as changes in cardiac self-efficacy; the results from these analyses are reported elsewhere [28] because they do not focus on the acceptability and feasibility component of the study.

All participants provided informed consent to participate in the study through an electronic consent form. The pilot study received approval from the Advarra Institutional Review Board (Pro00061694). Study personnel instituted appropriate safeguards to prevent any unauthorized use or disclosure of personal health information and implemented administrative, physical, and technical safeguards to protect the confidentiality, integrity, and availability of protected health information. All study data presented in this manuscript are deidentified. Lark is compliant with Health Insurance Portability and Accountability Act (HIPAA) privacy and security rules and all applicable regulations. In addition, Lark is Service Organization Control 2 (SOC 2) and Health Information Trust Alliance (HITRUST) certified.

All study participants received the Heart Health app and a cellular smart scale. There were opportunities for additional incentives throughout the study. At the start of the study, participants could receive a US $50 gift card upon ordering their cellular scale; this incentive was implemented to encourage participant use of the cellular scale for regular weigh-ins for higher data integrity compared to manual entry of weight data. There were two additional opportunities for participants to earn gift card incentives for completing (1) a telephone call regarding customer satisfaction and the usability of the Heart Health app (US $25) and (2) a telephone call about experiences taking surveys in the app (US $100). Additional details on these telephone calls are provided in the Participant Satisfaction Surveys subsection. Finally, participants could receive a Fitbit device upon study completion.

As this pilot study focused on feasibility and acceptability, the results for primary outcomes are largely descriptive in nature, and we evaluated success based on success criteria established a priori, described in the Objectives subsection. Retention rate, engagement, and screener submission rates were summarized using percentages and 95% CIs for proportions using normal approximation. Participant satisfaction was expressed using means and 95% CIs. For the secondary analyses on weight nadir, we report on the weight nadir across the sample, as described in the previous subsection. In addition we assessed the relationships between (1) weight nadir date and weight loss and (2) baseline BMI and weight loss using Pearson correlations.

Table 1 contains basic descriptive statistics of the enrolled study sample. On average, Heart Health study participants were aged 60.9 (SD 10.3; range 40.5-75.9) years and had a BMI indicating class I obesity (mean 32.6, SD 5.9 kg/m²; range 24.5-49.6 kg/m²). Of the 489 participants, 298 (60.9%) were female; 33 (6.7%) identified as Black, 354 (72.4%) identified as White, and 95 (19.4%) reported belonging to another racial group. An assessment of basic health history indicated that more than half of the sample (286/489, 58.5%) had a history of treatment for high blood pressure, and 19% (93/489) had a history of type 2 diabetes. On the basis of reports of a cardiac event or CVD diagnosis history, 89.8% (439/489) of the participants were in primary prevention, and 10.2% (50/489) were in secondary prevention (ie, history of a cardiac event or CVD diagnosis). ASCVD risk scores for participants in primary prevention who provided complete surveys indicated that 63.8% (118/185) of the participants in this group were borderline risk or higher for developing CVD or having a cardiac event in the next 10 years.

Regarding health behaviors, a little more than half of the sample (263/489, 53.8%) reported never smoking, while approximately a third of the sample (169/489, 34.6%) were former smokers, and 8.8% (43/489) were current smokers. The majority of the sample (382/489, 78.1%) reported being mainly sedentary or engaging in mild low-effort exercise during their leisure time. All enrolled participants reported at least 1 unhealthy dietary habit, including eating salty food daily, eating fried food ≥3 times per week, eating meat ≥2 times per day, not eating fruit daily, or not eating vegetables daily.

The majority of the participants (481/489, 98.4%) in this sample were obese. Specifically, 36.4% (178/489) of the participants were overweight, 34.4% (168/489) were classified as obese class I, 14.3% (70/489) as obese class II, and 13.3% (65/489) as obese class III. Of the 489 participants, 8 (1.6%) reported a BMI between 24 and 25 kg/m² (ie, normal weight) at baseline; we considered these individuals eligible (ie, BMI ≥25 kg/m²) during prescreening because they provided a slightly higher weight on the prescreener survey.

In month 3, study retention based on providing a weigh-in after day 60 was 80% (407/509; 95% CI 76.2%-83.4%). This month 3 retention rate is comparable to benchmarks from non-AI digital health programs [38].

For objective 1, we evaluated participant satisfaction with the Heart Health program. Overall, participants were highly satisfied with the overall app experience (Table 2). On average, participants scored the Lark Heart Health app experience ≥4 (favorable) for each question asked. Members were also highly satisfied with their experience taking the ASCVD survey in the app (Table 2). On average, participants provided high ratings of their comfort with, and understanding of, the ASCVD survey; participants also considered the survey length to be appropriate (422/458, 92.1%, 95% CI 89.3%-94.4% gave it a rating of 3 [just right]).

For objective 2, we found that participants were highly engaged with coaching in the Heart Health app (Table 3). In month 1, of the 489 participants, 483 (98.8%) met the success criteria for being highly engaged (ie, ≥5 coaching conversations); in month 2, there were 445 (91%) highly engaged participants; and in month 3, there were 408 (83.4%) highly engaged participants. When examining the total number of coaching conversations over the course of the program, participants averaged 129.8 (SD 110.1; median 104.0; 95% CI 120.0-139.6) coaching conversations across the 90-day study period, or 1.4 conversations per day. There was substantial variability in the number of coaching conversations across participants, as indicated by the large SD in total coaching conversations. The median number of coaching conversations was lower for the total number of conversations, indicating that the means may be skewed by extremely engaged users. However, the median values still indicate high engagement.

We also tracked the number of days with meals logged, days with weigh-ins, and total number of educational lessons completed across the 90-day study period, all of which occur within coaching conversations with the Lark coach. On average, participants logged meals on 57.0 (SD 30.1; median 68.0; 95% CI 54.3-59.7) days, nearly two-thirds of the days in the study. Regarding weigh-ins, the Lark coach encourages participants to weigh themselves at least once per week. On average, participants completed a weigh-in on 24.3 (SD 19.1; median 18.0; 95% CI 22.6-26.0) days, or approximately 2 times per week over the course of the study. Of these weigh-ins, participants provided the vast majority (435/489, 88.9%) by stepping on their cellular scale, with only 11% (54/489) self-reported within the app. Participants completed an average of 7.1 (SD 4.1; median 9.0; 95% CI 6.7-7.5) educational lessons during the 90-day study.

For objective 3, we found that participants were highly engaged with the screeners in the app. Specifically, 89.8% (440/489; 95% CI 87%-92.5%) submitted all 3 screeners (excellent), 94.3% (460/489; 95% CI 91.6%-96%) submitted ≥2 screeners (good), and 97.5% (479/489; 95% CI 96.3%-99%) submitted ≥1 screeners (minimum improvement over usual care). It should be noted that these groups were not mutually exclusive.

We examined peak weight loss among participants who provided at least 2 weights (483/489, 98.8%). Across this sample, the average weight nadir was 3.8% (SD 2.9%; 95% CI 3.5%-4.1%) and occurred on day 51.9 (SD 26.4; 95% CI 49.5-54.3). There was a significant correlation between the day in the study that the weight nadir occurred and percentage weight loss at nadir date (r=0.36; P<.001). This indicates a higher percentage weight loss for weight nadir dates later in the study. Across all participants, the vast majority (482/483, 99.9%) lost or maintained weight, with only 1 (0.2%) of the 483 members gaining weight. The proportion of the sample in 4 different weight loss categories is as follows: 26.5% (128/483) lost weight ≥5%, 44.7% (216/483) lost weight between 2% and 4.9%, 28.6% (138/483) maintained weight –2% to +2%, and 0.2% (1/483) gained weight >2% [39,40]. We also examined weight loss by baseline BMI and found that peak weight loss was not significantly correlated with BMI (r=0.004; P=.41).

The study results indicate that the fully digital AI-powered health coaching app called Lark Heart Health had a high degree of acceptability and feasibility among this pilot study sample. In this 90-day study of the Heart Health program, we found that participants were highly satisfied with their experience using the app, showed a high degree of engagement with health coaching and other app features, and successfully completed screeners on important health indicators during their first month in the program. Secondary analyses further indicated that participants lost weight during the 90-day program. Taken together, the findings from this pilot study met or exceeded all success criteria for the primary study objectives and provided key insights into weight loss as a clinical outcome that can be further examined in future clinical studies. Thus, these findings support the acceptability and feasibility of the Heart Health program. We discuss key insights from each of the primary objectives in the next subsection.

Regarding objective 1, the 2 participant satisfaction surveys indicated that participants were highly satisfied with their experience using the Heart Health program, including their experience taking the ASCVD Risk Estimator survey in the app. Mean scores on all items across the 2 surveys indicated that participants rated their experience favorably across multiple dimensions, such as how easy it was to understand the information in the app and feeling comfortable with taking a survey in the app. The mean score on the overall app experience survey (4.4 out of 5; SD 0.8) also closely aligns with high satisfaction scores in prior feasibility studies on apps for CVD management [24].

For objective 2, we found that participants were highly engaged with coaching in the Heart Health app. In month 1, of the 489 participants, 483 (98.8%) met the success criteria for being highly engaged (ie, ≥5 coaching conversations); in month 2, there were 445 (91%) highly engaged participants; and in month 3, there were 408 (83.4%) highly engaged participants. Engagement with the app far exceeded the expectations set by our success criteria. This insight indicates that engagement thresholds can be set at higher levels in future testing with the Heart Health program. Although monthly engagement rates were higher than anticipated across all 3 months, the results showed a decrease in engagement over time; for instance, 8.6% (42/489) of the participants were minimally engaged in month 3 (the final month in the study), as indicated by ≤1 coaching conversation during this month. This decrease is expected in digital health, with engagement and retention rates typically dropping over time [41,42]. Identifying this decrease in engagement also provides the opportunity to identify when engagement begins to wane and perhaps use targeted incentives at this time to mitigate drop-off. As we focused on coaching conversations as the primary engagement outcome, it is also possible that some users may have favored different features, such as weight logging and device use (eg, cellular scale). The descriptive results on weigh-ins support the assertion that users engaged with this feature.

The results from objective 2 also highlight the high degree of variability in engagement between participants. As with any program or app, different individuals will have different use patterns or connect with different aspects of the program. Indeed, we have shown in previous research of other Lark programs that there are different user personas among individuals who engage with the app, with some users focusing more on data and tracking functionalities or device use, rather than being highly engaged with coaching specifically [43].

In the context of this pilot study, there were some participants who had disproportionately high engagement levels. In our previous work, we refer to these individuals as enthusiasts [43]. These are individuals who log on many times per day and engage with many or all facets of the app. It is important to note that super users are not representative of most users of an app and can skew mean scores; for example, the median values for coaching conversations in this study were lower than the average values, although this still indicated high engagement. That said, designers of apps can gain significant insights for program improvement from qualitative interviews with both high engagers and low engagers alike [41].

Regarding objective 3, the results showed high participant submission rates for screeners focused on important health indicators, including ASCVD risk, medication adherence, and depressive symptoms. Specifically, 90% (440/489) of the participants achieved the a priori benchmark of submitting all 3 health screeners. This result indicates that participants were willing and able to complete these screeners using the app. This is a key result for the feasibility and acceptability of taking screeners in the Heart Health app because prior work has highlighted that it is important for patients in heart health prevention to show high levels of approval for risk surveys [34]. Moreover, these findings support the use of the Heart Health app as a user-friendly solution that enables patients to complete important health screeners outside of the primary care setting. This is an important use because there are significant barriers to screener completion during patient visits, such as short appointment times, workflow disruption, and a lack of patient preparedness [19,20,27]. Moreover, screener completion has been shown to have several benefits to patients; for instance, health screeners can help increase patient awareness and knowledge of their values on key CVD risk factors (eg, cholesterol level and blood pressure) [20].

The acceptability of screener submission in the app also has important implications for future delivery of the Heart Health program. In a commercial version of the program, participants would complete health screeners, such as the PHQ-2, MAQ, and ASCVD Risk Estimator, every 60 to 90 days. Longitudinal self-monitoring has many benefits for the self-management of CVD, such as enabling participants to track changes in their symptoms over time [44] and improve medication adherence [45]. Further integration of the Heart Health app with health partners would also enable participants’ health providers to track their patients’ changes over time and potentially support clinical decision-making [20,46].

Beyond the primary acceptability and feasibility study results, secondary analyses also revealed initial insights into weight loss during the program. Specifically, participants showed an average peak weight loss of 3.8% (SD 2.9%) during the study, and weight loss percentage was not correlated with baseline BMI. This percentage weight loss compares favorably to that in prior studies examining weight loss in digital health interventions [47]. These results should be considered preliminary because the pilot study was not developed with the intention of testing weight loss or other clinical outcomes. However, these results support the future consideration of weight loss as a clinical outcome for the Heart Health program.

We observed an 80% (407/509) retention rate in month 3 of the study. This retention rate is on par with target benchmarks from non-AI digital health studies that are higher touch than Heart Health [38]. In addition, the withdrawal rate for the study was low: only 20 (3.9%) of the 509 participants withdrew after enrollment. Withdrawal rates for digital health apps with frequent survey content range from 15% to 51% [48]. It is important to note that the retention rates reported here should be interpreted as research study retention rates; study retention rates are typically higher than retention rates in programs that are commercially released to real-world populations. Indeed, it is well established that retention is one of the biggest challenges with digital health solutions, and there are often high attrition rates with health programs delivered via mobile apps [41,42]. Improving retention rates in real-world populations is a major focus in the digital health space [49] and will be a focal point in a commercial version of the Heart Health program.

There are several limitations to the study results. First, this was a pilot study examining feasibility and acceptability as a first step in the evidence-generation process; the findings presented here are not meant to show the clinical efficacy of the Heart Health program and should not be interpreted as such. Future studies could examine the impact of the Heart Health program in a controlled trial; for example, a key next step in evidence generation is to examine clinical outcomes (eg, weight loss) between program participants and a control group receiving usual care or a control group that involves human coaching. As a second limitation, the sample used for this study contained individuals recruited to participate in a research study rather than real-world participants of a commercial version of the product. As such, an important future direction for generating evidence for the Heart Health program will be examining these objectives and other study outcomes in a commercial version of the product with a real-world sample. In addition, this study included opportunities for incentives throughout the Heart Health program that are similar to incentives provided in existing Lark programs but may not be identical to incentives provided in the commercial implementation of the Heart Health program. Importantly, only 1 incentive was directly tied to any type of app engagement: the Fitbit device incentive provided for completing the 90-day program. It is common for research studies to provide an incentive for study completion, and it is also common for Lark’s existing health care partners to provide members with a Fitbit device for reaching a major program milestone. As a result, we do not believe that this incentive had any undue impact on participant engagement that would differ significantly from the commercial implementation of the program. Differences in incentive structure may have an impact on participant engagement in future program implementation, and this will be dependent on health care partners.

A third limitation is that the sample in this study came primarily from a single health partner in California, and participants tended to be older White adults from this region. Future work should examine the impact of the program among diverse samples; indeed, this is a key limitation and a future direction in the broader digital health space [12,42]. Finally, a key next step in improving the program is to gain a deeper understanding of the user experience with the app; this could include qualitative or interview-based studies with individuals who show different engagement patterns (eg, low engagers, average engagers, and super users) to gather detailed feedback on features that could be improved.

These acceptability and feasibility results provide a key first step in the process of evidence generation for a new AI-powered digital program for heart health. These results indicate that participants in this study were satisfied with their experience using the Heart Health app, highly engaged with app features, and willing and able to complete health screening surveys in the app. While digital health solutions are increasingly common, many have not been tested using formalized evidence-generation strategies that are traditionally used for new therapeutics, such as pharmaceuticals [13]. As such, the results from this pilot study provide an example blueprint for early-stage evidence generation for new digital health offerings in the prevention and management of CVD.