The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 statement was followed in conducting this review (Checklist 1) [23]. This systematic review was not prospectively registered. A protocol was developed by the authors to guide the review process, but it was not published or made publicly available.

A comprehensive search was conducted in PubMed, Web of Science, and the Cochrane Central Register of Controlled Trials (CENTRAL) on April 3, 2024 to identify relevant studies. The search strategy was developed based on the PICO (population, intervention, comparison, outcome) framework [24] and combined terms related to “CVD,” “HR,” “wearable HR monitors,” and “PA” using Boolean operators. Search syntaxes were adapted for each database to account for differences in indexing and Boolean operator functionalities. Full search strings for each database are detailed in Table S1 in Multimedia Appendix 1. In addition to database searches, the reference lists of included articles were screened to identify additional relevant studies (backward citation search).

Studies were eligible for this review if they met the following inclusion criteria.

Duplicate records were identified and removed prior to screening, using a systematic deduplication method in EndNote, as described by Bramer et al [25] The screening process was conducted in multiple stages using Rayyan software (Rayyan Systems Inc.). Initially, 2 reviewers (PV and CB) independently screened article titles to exclude irrelevant studies. This was followed by a screening of abstracts. Articles deemed relevant based on abstract screening proceeded to full-text review, during which the reviewers assessed the complete articles to confirm their eligibility according to the predefined criteria. Studies where eligibility could not be determined from the abstract were included for full-text review. If discrepancies arose at any stage, they were resolved through discussion or, if necessary, a third reviewer (LD) was consulted to resolve them.

A structured data extraction form was developed in Microsoft Excel and pilot-tested on 2 randomly selected studies to ensure consistent, clear, and comprehensive data collection. Two independent reviewers (PV and CB) extracted data from the included studies and any discrepancies were resolved through consensus discussions between the 2 reviewers (PV and CB); and if necessary, a third reviewer (LD) was consulted.

The primary outcome of interest in this review was the methodological approach used to translate wearable HR data into PA-related metrics or physiological estimates. Specifically, we extracted and classified how HR data were operationalized (eg, threshold-based HR zone analysis, regression-based energy expenditure models, cardiorespiratory fitness estimation, change detection algorithms, or personalized activity scores).

In addition, extracted data included study characteristics (eg, title, authors, publication year, country, funding, and design), participant demographics, HR monitor specifications, and details on the intervention or observation period (eg, training program and duration). We also extracted study-specific reported outcomes to provide contextual interpretation of each methodological approach. The reported outcome parameters varied across studies and included measurement validity indicators (eg, comparison with CPET or ECG, mean absolute [percentage] error, intraclass correlation coefficient, coefficient of determination), adherence rates, physiological outcomes (eg, maximal oxygen uptake [VO_₂max]), feasibility metrics, and safety events. Given the heterogeneity in devices, populations, and study designs, these outcomes were synthesized descriptively and were not pooled quantitatively.

The methodological quality of the included studies was evaluated using the Crowe Critical Appraisal Tool (CCAT; Michael Crowe) [26]. This tool was selected due to its flexibility and applicability across diverse study designs, making it particularly suitable for this review, which included randomized controlled trials, observational studies, feasibility studies, methodological studies, and protocol descriptions.

The CCAT assessed studies across 8 domains, including preliminaries, introduction, design, sampling, data collection, ethics, results, and discussion. Each domain is scored on a scale from 0 to 5, where 5 indicates that the criterion is fully met and 0 indicates that it is not addressed. Intermediate scores reflect partial fulfillment of the domain criteria according to the CCAT user guide [27]. Domain scores were assigned using the guidance provided in the CCAT user guide, where item descriptors support reviewer judgment when determining the final domain score [27]. The maximum total score is 40.

For protocol descriptions, only applicable domains of the CCAT were scored. Domains that were not applicable due to the absence of empirical results (eg, results and discussion) were marked as “not applicable”. The total score for protocols was recalculated as a percentage of the maximum applicable score (eg, if 6 domains were scored, the total score was calculated out of 30 instead of 40). This approach ensured that protocol descriptions were fairly assessed for their methodological relevance without being penalized for the absence of empirical findings.

Two independent reviewers (PV and CB) assessed methodological quality and resolved discrepancies through consensus discussion. CCAT scores did not influence study inclusion decisions and were used to contextualize interpretation rather than to weight or exclude findings within the narrative synthesis.

Although the CCAT is primarily designed as a methodological appraisal tool, several domains address aspects closely related to potential sources of bias (eg, study design, sampling strategy, data collection procedures, and discussion of bias and confounding). Therefore, CCAT scores provide an indirect indication of methodological robustness and susceptibility to bias.

Given the heterogeneity of study designs and outcomes, a narrative synthesis was conducted. The primary aim of the synthesis was to identify and classify methodological approaches used to translate HR data into PA-related metrics.

During data extraction, the reported HR-based PA estimation methods from each study were examined and compared. Through an inductive process, recurring methodological approaches were identified and grouped into conceptual categories based on their underlying estimation strategy. This resulted in 4 distinct categories, explained in section “Heart Rate–Based Physical Activity Estimation Methods” of the Results. The categorization was discussed between the 2 reviewers (PV and CB) and finalized through consensus.

After categorization, studies within each methodological group were compared descriptively with respect to contextual characteristics such as HR monitor type, study population, and reported outcomes. Similarities and differences between studies were identified through iterative comparison across these dimensions. Methodological gaps were defined as areas where approaches lacked transparency, validation in cardiovascular populations, or consistent methodological reporting.

The study selection process is summarized in a PRISMA 2020 flow diagram (Figure 1). The search for studies quantifying PA using HR monitors in patients with CVD yielded 828 results. After removal of duplicates, 673 publications were screened first on title and then on abstract. A total of 53 articles were screened on full text, 20 of which met the inclusion criteria.

Of the 20 included studies, 15 [28-42] were research articles and 5 [43-47] were protocol papers (Table 1, left side). The study designs included 8 [28,29,38,43-47] randomized controlled trials (RCTs), 5 [30-33,42] nonrandomized interventional studies (4 feasibility studies and 1 pre-post study), and 7 [34-37,39-41] observational studies that monitored PA patterns using HR monitors without aiming to induce behavioral change. Studies were conducted across diverse geographical locations, primarily in Europe and the United Kingdom (n=12) [29,31,32,35,36,38-40,43,45-47], followed by North America (n=5) [28,34,37,41,44], Asia (n=2) [30,33], and Australia (n=1) [42]. Sample sizes varied widely, with feasibility studies including the fewest participants (range: 5‐31), whereas other study designs included up to 300 participants.

An overview of the study populations is shown in Table 1 (right side). In terms of health characteristics, 8 studies [29,32,33,38,39,42,45,47] focused on patients eligible for CR following ACS, CAD, or revascularization procedures (eg, PCI or CABG). Other study populations included patients with heart failure (n=2) [28,46], post-stroke patients who have experienced a stroke (n=2) [34,44], individuals with cardiovascular risk factors (n=1) [40], congenital heart disease (n=1) [36], valve disease (n=1) [35], and patients postmyocardial infarction (n=1) [43]. Two studies [37,41] investigated inpatient rehabilitation patients with mixed conditions including cardiac disorders, while 2 feasibility studies [30,31] were conducted in healthy adults, assessing HR-based PA methods intended for populations with heart diseases.

CR studies had a predominantly male representation (82%‐100% [29,33,38,39,42]), whereas populations with heart failure (55% male [28]) and populations who have experienced strokes (54% male [34]) were more balanced. Age ranged from young healthy adults (mean 26.3, SD 4.3 years [30]) to older patients with heart failure (up to mean 65.6, SD 9.3 years [28]). Patients undergoing CR typically ranged between 55.8 (SD 7.3) years [39] and 59.2 (SD 8.5) years [38]. BMI values were generally between 26 and 31 kg/m² for CR populations [29,33,38,39,42], with the highest BMI observed in patients with heart failure (33.9‐36.4 kg/m² [28]). Beta-blocker use was the highest among patients with heart failure (98% [28]) and ranged between 17% and 91% in CR populations [29,33,38,39,42].

A total of 12 out of 20 studies [29-32,34,35,37,41-44,47] used a wrist-worn HR monitor (60%), 7 [33,36,38-40,45,46] used a chest-strap (35%), and 1 [28] did not specify the type (5%) (Table 2). Polar devices were the most commonly used (6 studies) [28,29,32,43,45,46], followed by Fitbit (4 studies) [34,37,41,44] and Garmin (2 studies) [38,39]. Other brands included Samsung, Philips, Firstbeat Technologies, and custom-designed wearables.

Four distinct approaches to HR-based PA estimation were identified: (1) HR zone analysis, (2) physiological estimation using HR-based models, (3) PA change detection, and (4) personalized activity scoring. These methods vary in complexity, ranging from simple threshold-based evaluations to algorithmic pattern detection. Each approach is summarized in Figure 2 and further elaborated in the subsections below, while detailed study specifications are presented in Table 2 (PA estimation approach) and in Table S2 in Multimedia Appendix 1.

The CCAT quality ratings ranged from 50% to 87%, with a mean of 71% (Table 3). Four studies scored above 75%, indicating high methodological quality, while 2 [26,27] studies scored below 60%, reflecting some limitations. The remaining 14 studies fell within the moderate range of 60%-75%. Most studies performed well in the domains of introduction, data collection, and discussion, with the majority scoring between 4 and 5 in these areas and no study scoring below 3. Fifteen studies [28-30,32-37,39-43,47] reached a moderate score of 3 in the design domain. Sampling and ethics were the weakest domains; 5 studies [30-32,39,45] scored below 3 for sampling, and 3 studies [30-32] scored below 3 for ethics, highlighting issues with sample representativeness and ethical reporting.

The methodological quality assessment was used to contextualize the interpretation of findings but did not determine study inclusion. Importantly, studies with lower scores were distributed across the different methodological categories identified in the narrative synthesis, and the methodological approaches used to derive HR-based PA metrics were largely consistent across studies irrespective of CCAT score. Overall, the quality of the included studies was moderate to high.

This review provides the first structured overview of HR-based methodologies for PA estimation in patients with CVD, highlighting their clinical utility, strengths, and the implementation gaps. We identified 4 distinct estimation approaches: HR zone analysis, physiological estimation using HR-based models, PA change detection, and personal activity intelligence score calculation. Most studies (14/20) used HR zone thresholds to evaluate PA intensity or adherence, often within CR contexts. Other methods incorporated algorithmic modeling or machine learning to derive metrics such as energy expenditure, CRF, or behavioral change. While these approaches varied in complexity and purpose, each demonstrated the potential of HR data to provide physiologically meaningful insight into daily PA patterns.

HR zone analysis emerged as the most commonly applied and straightforward method of HR-based PA estimation in this review. It is straightforward because it allows for a direct translation of international PA guidelines into quantifiable metrics, specifically, the time spent in moderate-to-vigorous intensity HR zones, and it aligns with the widely adopted the frequency, intensity, time, and type framework [1,4]. As such, it provides clinicians and researchers with an accessible method to both prescribe and evaluate PA intensity in line with global recommendations.

Both the World Health Organization (WHO) and the European Society of Cardiology (ESC) advise that adults should engage in at least 150 minutes of moderate or 75 minutes of vigorous intensity aerobic activity per week, or an equivalent combination of both, with intensity defined via several physiological indicators, including %VO₂max, %HRmax, %HRR, or ratings of perceived exertion [1,4]. Among the reviewed studies, %HRmax and %HRR were used most often to define intensity zones. ESC recommends that maximal HR values should ideally be based on exercise testing rather than estimated from age-predicted formulas, especially in clinical populations and in patients on rate-limiting medications such as beta-blockers [1]. Despite this, most studies in our review relied on simple prediction equations for HRmax, which may result in less precise zone definitions. Notably, none of the included studies explicitly reported whether vigorous-intensity minutes were counted double when evaluating weekly MVPA totals. While international guidelines do allow for this (eg, 75 min of vigorous activity or an equivalent combination with moderate activity) [1,4], the absence of clear reporting may limit interpretability and cross-study comparability. Including this distinction in future reporting could help ensure more accurate and standardized assessments of adherence.

HRrest, another component needed for calculating HRR, was often defined through study-specific approaches, such as averaging nighttime HR values or identifying low-HR periods during inactivity. While literature suggests that early morning supine HR, measured immediately after waking but before getting out of bed, may offer the most reliable estimate, there is no standardized or transparent method for deriving HRrest from wearable data at the time of this writing [52]. Most commercial devices provide HRrest estimates through proprietary algorithms, leaving researchers with limited insight into how these values are calculated.

In terms of adherence assessment, some studies focused on guideline-based targets (eg, ≥150 min/week of MVPA), while others evaluated compliance with structured training programs based on the frequency, intensity, time, and type principles. Importantly, HR monitoring enabled a distinction between session attendance (frequency) and session quality (intensity and duration), offering a more nuanced understanding of adherence by distinguishing between mere participation and actual achievement of intensity and duration targets [53].

Finally, several studies integrated HR zone data into feedback mechanisms, either in real time (eg, via smartwatches with LED zone indicators) or post-session through digital dashboards and clinician-guided counseling. Effective feedback has been shown to be a key driver of behavior change in PA interventions [54]. When combined with motivational strategies such as self-monitoring and goal setting, these feedback loops may support greater behavioral change and patient engagement toward sustained PA adherence [43,44,54].

Several studies in this review combined HR monitoring with accelerometer-based methods to quantify PA adherence and estimate physiological parameters such as energy expenditure and VO₂max. These multimodal approaches aimed to improve accuracy and/or cross-validate findings by combining the complementary strengths of each method [55]. HR monitoring provides a direct physiological measure of exercise intensity and cardiovascular load, making it a more personalized and clinically relevant indicator of effort than mechanical metrics such as step count or accelerometer counts. Moreover, the physical demand associated with a given accelerometer value can vary substantially between individuals depending on their fitness, age, or health status [56]. For example, walking at the same pace may represent light intensity for one person and moderate intensity for another [36].

Despite these limitations, accelerometer and step count data could play a valuable role in assessing sedentary time, an increasingly recognized factor in cardiovascular health [57]. Sedentary behavior, defined as waking behavior in a sitting, reclining, or lying posture with low energy expenditure, has been associated with cardiometabolic risk, even in individuals who meet weekly MVPA guidelines [58,59]. This risk profile may reflect distinct physiological pathways, underscoring the importance of addressing sedentary behavior alongside PA adherence [60]. In the PACD study included in this review, a reduction in sedentary time (defined as time with fewer than 5 steps per minute) emerged as a key predictor of meaningful PA behavior change [41]. When paired with HR data to reflect oxygen demand and PA intensity, step-based indicators of inactivity may offer a low-burden, scalable method for tracking both sedentary time and physiological load. However, it is important to assess PA and SB as independent behavioral domains, each with distinct health effects and potentially requiring their own intervention strategies [60].

Although most studies focused on measuring adherence to PA guidelines (eg, time in zones), a smaller subset advanced toward estimating physiological outcomes from HR data, such as energy expenditure (PAL) or cardiorespiratory fitness (VO₂max). This transition from behavioral metrics to physiological modeling represents a valuable next step, especially in clinical populations where improving aerobic capacity and reducing cardiovascular risk are primary goals.

For example, three studies estimated PAL, defined as the ratio of total energy expenditure to resting metabolic rate. These models typically combined HR above rest with accelerometer data and participant characteristics [38,39,47]. PAL is widely used in nutrition science to estimate daily energy needs, but in cardiovascular care its clinical application remains limited [61]. As a descriptive metric, it lacks behavioral thresholds, is rarely linked to feedback mechanisms, and often relies on regression models with undisclosed coefficients, limiting transparency and reproducibility in research and practice. A more advanced example is the proprietary Firstbeat algorithm, which estimated VO₂max using HR, HR variability, movement data, and participant characteristics during a 30-minute walk [40]. While generalizability to cardiac populations remains uncertain and requires further validation, the Firstbeat algorithm showed acceptable agreement with CPET-derived values in athletic cohorts or adults with cardiovascular risk factors and is now embedded in commercial wearables from Garmin, following their acquisition of Firstbeat Technologies Oy in 2020 [40,62,63]. While this integration supports scalability, the proprietary nature of the algorithm limits its adaptability and full validation in diverse patient groups.

Together, these examples illustrate a growing shift from pure activity tracking toward predictive modeling of physiological capacity. By combining HR data with individual and contextual factors, such models could move beyond surface-level monitoring to support personalized CR planning and long-term health surveillance. Such approaches may help clinicians and researchers estimate meaningful outcomes, such as changes in cardiorespiratory fitness, based on daily-life activity patterns, ultimately linking daily PA behavior to cardiovascular impact.

The PAI system stands out within this review as the only approach that translates continuous HR data into a personalized, interpretable single-number feedback metric intended to motivate behavioral change over time. PAI reflects accumulated cardiovascular effort across a rolling 7-day window. In the one included PAI study in this review, short-term improvements in PAI accumulation were observed after CR participants gained access to their PAI score via a smartphone application, although evidence remains exploratory [42].

As a caveat, the commonly cited ≥100 PAI/week threshold, proposed to reduce the risk of CVD and mortality, originates from retrospective analyses of large population studies (eg, the HUNT Study in Norway) [48,51,64-68]. During the screening process of this review, eight such validation studies were identified but excluded, as none used HR monitors to calculate PAI [48,51,64-68]. Instead, PAI scores were estimated post hoc from self-reported PA questionnaires, relying on participants’ recall of activity frequency, intensity, and duration [48,51,64-68]. This introduces substantial subjectivity and deviates from the intended physiological, HR-based calculation of PAI, potentially reducing both accuracy and internal validity.

Moreover, 7 of the 8 validation studies were conducted in healthy adult populations, frequently excluding participants with cardiovascular conditions. The reported mean ages were relatively young, ranging from 42.6 [64] to 55.1 [68] years. One study included participants with self-reported angina pectoris, myocardial infarction, or stroke, with a higher mean [SD] age of 67.6 [10.3] years[66]. However, activity in that group was still assessed via questionnaires, and the population was not clinically verified. To date, no validation studies have assessed the accuracy or clinical utility of PAI when calculated from real HR data in older adults or patients with CVD, populations for whom accumulating ≥100 PAI points per week may be less achievable but important. Also, the PAI algorithm rewards sustained vigorous activity more than moderate or lower-intensity efforts, limiting its applicability for clinical populations where activities such as walking or cycling are more prevalent [64]. A summary table of the characteristics of the 8 excluded PAI validation studies is available in Table S3 in Multimedia Appendix 1.

Although the algorithm is largely reproducible based on published formulas, certain implementation details remain proprietary [18]. Researchers with technical expertise can replicate the model in principle, but public tools for doing so are limited. The MIA Health app, currently the only available platform for real-time PAI calculation, expanded its availability in January 2025 from a few Nordic countries to most of Europe, North America, Australia, New Zealand, Taiwan, and Brazil, potentially increasing both clinical and consumer access [69].

Recently, a follow-up metric known as the activity quotient (AQ) was introduced by the same research group as a refinement of the PAI. Some retrospective studies have validated AQ using the same questionnaire-based methodology as PAI and report stronger associations with VO₂max and mortality [70]. However, none of these studies used HR monitors, and prospective AQ studies with wearable devices are still forthcoming [69]. In parallel to these developments, our research group is developing the Antwerp Activity Index and smart Antwerp Activity Index models [71,72] (P Vermunicht, PhD, unpublished data, 2026). These tools aim to provide HR-based, personalized activity scores by integrating daily-life HR data with individual characteristics, and are designed to predict cardiorespiratory fitness changes over time. While preliminary development has been promising, full clinical validation and prospective outcome studies are currently underway. Such emerging models may further advance the translation of HR-derived activity monitoring into actionable, personalized health guidance in cardiac populations.

Together, these developments reflect a growing interest in HR-based PA quantification and behavioral feedback. However, their clinical relevance for cardiac and older adults will depend on prospective validation in target populations, improved algorithmic transparency, and appropriate feedback tailoring for diverse activity profiles.

The reviewed studies illustrate a diverse range of HR-based methods, from guideline-driven time-in-zone metrics to more advanced models estimating physiological outcomes and personal activity scores. This variety reflects a growing interest in using HR data not only to monitor activity but also to estimate its cardiovascular implications. Future research should prioritize validation of these methods in real-world populations with CVD, with attention to feasibility, clinical interpretability, and predictive value. Tools that go beyond simple time-in-zone tracking and instead estimate outcomes like energy expenditure or changes in cardiorespiratory fitness may offer greater clinical value. As illustrated by systems such as PAI or the Firstbeat VO₂max model, combining HR data with individual characteristics can generate personalized feedback with the potential to guide long-term health behavior. Future work should strive for a balance between algorithmic sophistication, transparency, and accessibility.

Although not a focus of this review, a key translational consideration for HR-based PA estimation is that the validity of any derived PA metric depends on the accuracy of the underlying HR data [73,74]. Although ECG-based chest straps provide highly accurate HR measurements, they are less practical for daily monitoring. Wrist-worn photoplethysmography devices are more convenient, but their accuracy can be affected by artifacts and they should be validated against an ECG before clinical use [75,76]. Recent work from our group has therefore focused on artifact detection and cleaning procedures that operate directly on device-generated photoplethysmography HR data from commercial wearables, aiming to remove unreliable data while preserving clinically meaningful activity-related HR fluctuations [71,76-78]. Such artifact removal strategies are highly relevant when implementing the HR-based PA quantification methods described in this review, particularly in real-world cardiac populations where pathophysiological factors such as peripheral edema, poor tissue perfusion, and a higher incidence of arrhythmias may affect HR reliability [76,79].

Beyond methodological development, successful implementation of HR-based PA estimation also depends on how these metrics are embedded within broader digital care strategies. Mobile health tools that support patient education, self-management, and structured follow-up may enhance engagement and facilitate behavior change, thereby increasing the clinical relevance of HR-derived activity metrics [80,81].

This review has several limitations. It was not prospectively registered in a public database (eg, PROSPERO [International Prospective Register of Systematic Reviews]), which may limit transparency regarding protocol deviations. Although a protocol was developed to guide the review process, it was not publicly available. As a narrative synthesis, it does not provide pooled effect sizes or meta-analytic comparisons. The included studies varied widely in design, sample size, and how HR-based metrics were operationalized. While this limits direct comparability, the primary aim of this review was to map the diversity of methodological approaches rather than pool results. In addition, studies using techniques such as PACD or personalized scoring systems remain scarce, restricting broader generalization. Continued development and validation of such tools, especially in older and cardiac populations, will be essential to support their integration into clinical practice. Finally, due to the rapid evolution of wearable technologies and increasing research interest in HR-based activity estimation, new studies may have been published after the final search date. For instance, an eligible outcome paper from the OptimEx-Clin trial was published shortly after the search period [82]. While this study was not included for methodological consistency, its design was captured through the previously published protocol paper, which is part of this review. Future updates of this review could consider incorporating such newly available outcome data.

82

This review provides the first structured synthesis of HR-based methodologies for PA estimation in patients with CVD. Most studies relied on HR zone analysis to evaluate adherence to international PA guidelines, while others explored more advanced modeling approaches to estimate physiological outcomes such as energy expenditure or cardiorespiratory fitness.

Our findings underscore that HR monitoring offers a physiologically meaningful and scalable way to estimate PA, especially in clinical populations where traditional step-based or questionnaire-based methods may fall short. Personalized HR-based feedback systems may offer intuitive activity targets for patients. However, variability in how HR metrics are defined, validated, interpreted, and applied, currently limits comparability and implementation. Future research should focus on validating HR-based methods in real-world populations with CVD, improving methodological transparency, and tailoring feedback to individual needs. Ensuring reliable HR measurements remains a prerequisite for meaningful HR-based PA estimation in clinical practice. As wearable technologies continue to advance, HR-derived metrics hold strong potential to support more personalized and effective PA guidance in both clinical care and daily life.