Readiness, recovery, and strain: an evaluation of composite health scores in consumer wearables

Study design & integrated health index selection

This study employed a secondary analysis of data extracted and synthesized as part of a previously published living umbrella review evaluating the accuracy of consumer wearable devices [43] – primary research studies identified as part of the living umbrella review were parsed for any reference to a CHS. This was supplemented by an industry report identifying major wearable device manufacturers [7], which ranked leading manufacturers of wearable devices based on global market share. From this list, the top 10 consumer wearable device manufacturers were identified, representing key stakeholders in the wearable technology sector with substantial user bases, and their product documentation was screened for any reference to a CHS as defined below.

Inclusion and exclusion criteria

Having first identified the most widely used consumer wearable devices in the research literature [43] and the companies with the largest market share [7], next, we collated a list of the different indices. For the purposes of this analysis, a CHS was defined as a composite measure derived from the integration of multiple biometric signals, such as heart rate variability, sleep metrics, physical activity, and body temperature, into a single score that reflects general health, recovery, or readiness for daily functioning.

Our inclusion criteria were as follows: 1) eligible CHS were required to combine multiple biometric signals, such as sleep metrics, heart rate variability, physical activity, and stress, into a composite score reflecting general health or readiness; 2) devices or platforms included in this study were required to be commercially available, with sufficient documentation detailing the components of the CHS; this documentation could include technical white papers, app screenshots, user manuals, or research publications; and 3) the CHS were designed for general health monitoring rather than sport-specific or performance-focused applications, ensuring relevance to a broader population.

Exclusion criteria were as follows: 1) single-construct metrics that only incorporated one biometric contributor (e.g., heart rate variability) or indices which focused solely on sleep or physical activity in isolation; 2) metrics explicitly tied to training load or other sport-specific performance measures; 3) devices or platforms lacking any publicly available information relating to CHS calculation or its contributing metrics; and 4) research-grade devices, prototypes or algorithms exclusively intended for clinical use and unavailable to the general public.

Procedures for evaluation

The evaluation of CHS was conducted through a systematic and structured approach designed to assess their contributors, methodologies, validation, and applicability. Data for the evaluation were collected from publicly available sources, including technical white papers, patents, user manuals, app screenshots and device interfaces, as well as research studies and validation reports published by manufacturers or independent investigators. Each index was analysed for its intended purpose, the metrics it integrates, its calculation methodology, and any available evidence supporting its validity. Missing or proprietary information was explicitly documented, and the implications of such gaps for scientific and clinical use were synthesized.

Each index was first contextualized with an overview, detailing its official name, the device or platform offering the index, and its primary application, such as readiness assessment, recovery monitoring, or general health evaluation. The contributors to the index were then assessed, focusing on the biometric signals integrated into the calculation, such as heart rate variability, sleep quality, physical activity, and stress levels. For each contributor, the evaluation considered its physiological relevance, the methods used for data acquisition and processing, and any thresholds or ranges associated with health outcomes, where available.

The calculation methodology of each index was then examined to understand how contributors were integrated into a composite score. This included an assessment of whether specific weightings or algorithms were disclosed and whether the indices incorporated short-term, long-term, or combined metrics. The scoring range and any interpretive categories, such as readiness levels or energy states, were documented and evaluated for their clarity and practical utility.

Whether each index had undergone any internal or independent validation was then assessed, provided this information was publicly available. Available evidence supporting the validity of the indices was synthesized, including any independent validation studies and their applicability across diverse populations or physiological conditions. Practical applications were considered, with an emphasis on the index’s utility for remote health monitoring, personalised health interventions, and general wellness management.

Finally, the transparency of the index – the extent to which it could be understood and replicated – was evaluated by reviewing the availability of technical documentation and the accessibility of detailed explanations regarding the construction and interpretation of each index. The degree to which manufacturers provided guidance to users, including recommendations based on the scores, was also examined. For indices with incomplete or proprietary data, the absence of key details, such as algorithmic weightings or validation data, was ascertained. The evaluation framework is summarised in Table 1.

Analysis and narrative synthesis

We used the evaluation framework to categorize the indices based on their primary purpose, the biometric contributors they integrated, and the methodologies employed for score calculation. A comparative approach was adopted to assess the consistency and variation in how indices synthesized physiological data into guidance or recommendations for the user. Particular focus was placed on understanding the weighting and integration of individual contributors, such as heart rate variability, sleep metrics, and physical activity, within the composite scores. The scoring scales and their interpretive categories were examined to evaluate their clarity and practical implications for users.

Where information was missing or proprietary, its absence was explicitly documented, and the potential consequences for usability and scientific credibility were analysed. For example, indices with undisclosed algorithmic methodologies were flagged as presenting challenges for independent validation and reproducibility. Similarly, gaps in validation data, particularly in diverse populations or under varying physiological conditions, were noted as potential limitations in the generalizability of the indices.

Evaluation of composite health scores in consumer wearables

Contributors to composite health scores

The primary biometric signals incorporated in the CHS were heart rate (86 % of CHS), heart rate variability (86 % of CHS), resting heart rate (79 % of CHS), activity (including, but not limited to, accelerometery derived motion, step counts and energy expenditure estimates; 71 % of CHS), and sleep quantity (71 % of CHS). These were followed by sleep quality/architecture (57 % of CHS), body temperature (29 % of CHS) and respiratory rate (14 % of CHS). Blood oxygen saturation, body morphology, blood pressure and heart rhythm (ECG) each contributed once to the various CHS.

Thus, the two sensing modalities underpinning the greatest number of biometric signals are PPG – which is used to capture heart rate, heart rate variability, respiratory rate and blood oxygen saturation – and accelerometery, which is used to capture movement, enabling assessment of activity and sleep.

The sensing modalities and biometric signals incorporated in the identified CHS are summarised in Table 2.

Comparative analysis

Overview of composite health scores and key contributors

This study sought to evaluate the design, contributors, and validity of consumer wearable CHS, aiming to clarify their construction and scientific basis. Our search of the scientific literature and of leading device manufacturers, including the available company documentation, identified 14 individual CHS: Coros’ Daily Stress [49], Fitbit’s Daily Readiness [44], Garmin’s Body Battery™ [45] and Training Readiness [50], Oura’s Readiness [46] and Resilience [51] Scores, Polar’s Nightly Recharge™ [52], Samsung’s Energy Score [53], Suunto’s Body Resources [54], Ultrahuman’s Dynamic Recovery [55], WHOOP’s Recovery [57], Strain [47] and Stress Monitor [56] and Withings’s Health Improvement Score [58]. Through a systematic analysis of the methodologies employed by these 10 manufacturers across their 14 indices, we identified consistent trends in how these tools integrate temporal biometric data streams – such as physical activity, cardiovascular function, and sleep – into a composite score.

Our findings reveal that CHS incorporate diverse physiological signals, including physical activity, body morphology, blood oxygen saturation, blood pressure, heart rhythm (ECG), heart rate, heart rate variability, resting heart rate, respiratory rate, sleep duration and quality, and body temperature. However, while all of the CHS ostensibly transform different variations of these complex physiological data into a discrete numerical value, the validity and robustness of their approaches is unclear. These tools are marketed as supporting health monitoring, optimising performance, and promoting recovery [44], [45], [46], [47, [49], [50], [51], [52], [53], [54], [55], [56], [57], [58] – and although specific metrics, such as heart rate variability for stress monitoring [60], or actigraphy for sleep onset detection [61], 62], are underpinned by robust empirical evidence, the composite indices themselves have undergone little to no validation in the peer-reviewed scientific literature [63], 64]. This is concerning given their popularity among users [48], and their intended applicability across diverse populations and physiological conditions. The reliance on opaque algorithms also presents significant barriers to independent scrutiny, hindering reproducibility and raising concerns about the robustness of the outputs [65], 66].

Physiological rationale for resting heart rate and heart rate variability

Findings from this analysis revealed that heart rate variability and resting heart rate were two of the most common contributors to the different CHS; 86 % and 79 % of indices incorporated resting heart rate and heart rate variability in their calculations, respectively. This prominence reflects their physiological significance; resting heart rate reflects the intrinsic pace-making activity of the sinoatrial node, which is modulated by the balance between sympathetic and parasympathetic inputs [67]. Under resting conditions, parasympathetic dominance reduces the sinoatrial node’s intrinsic firing rate (∼100 beats per minute) to typical values of 50–70 beats per minute in the general population [67]. Higher resting heart rate is associated with lower cardiorespiratory fitness and less parasympathetic tone, while lowered values often indicate reduced sympathetic activity or increased cardiovascular efficiency [68]. The inclusion of resting heart rate in CHS is therefore well-justified, as it provides a measure of baseline cardiovascular function [68], 69].

However, resting heart rate alone is less sensitive to acute physiological changes, such as an acute training ‘stress’, limiting its ability to capture dynamic autonomic fluctuations [70], 71]. In contrast, heart rate variability provides a more nuanced view of autonomic balance by quantifying fluctuations in the time intervals between successive heartbeats (RR intervals) [72]. Heart rate variability directly reflects rapid, millisecond-scale adjustments in parasympathetic tone, particularly vagal modulation during the respiratory cycle [71]. High heart rate variability at rest indicates robust parasympathetic activity and physiological adaptability, whereas low heart rate variability is associated with stress, fatigue, and poor health outcomes [70], 73] – although it should be noted that what represents “high” and “low” are unique to the individual [73]. Among the indices analysed, heart rate variability – more accurately termed pulse rate variability because it is derived from PPG in wearables [74] – was predominantly calculated using the root mean square of successive differences (rMSSD) between heartbeats, a standard method for capturing parasympathetic-driven variability [71]. This approach is widely recognized for its ecological validity over short- and ultra-short measurement intervals (1–5 min), although not all popular consumer wearables have adopted this approach [74].

The complementarity of resting heart rate and heart rate variability underpins their integration into composite indices. Resting heart rate captures slower, cumulative changes in autonomic tone, such as those resulting from chronic stress or shifts in fitness [68], [69], [70], while heart rate variability provides a dynamic indicator of acute autonomic responses [67], [71], [72], [73]. For example, heart rate variability typically recovers more rapidly than resting heart rate following intense exercise, reflecting transient physiological states [71]. This temporal distinction supports the use of heart rate variability as an indicator of readiness or recovery, with resting heart rate providing context for longer-term trends.

Variability in measurement protocols across manufacturers

The measurement protocols for these metrics, however, varied significantly across different companies and indices. Some companies sample their heart rate data during specific sleep periods to minimise confounding factors, but differences in timing, duration, and algorithmic approaches were identified. For example, Fitbit’s Daily Readiness Score calculates heart rate variability using the longest period of sleep that exceeds 3 h [44], while Garmin’s Training Readiness metric averages 5-minute heart rate variability windows throughout a full night of sleep [50]. Oura’s Readiness Score computes heart rate variability as the mean of all 5-minute samples across the entire sleep period [46], whereas Polar’s Nightly Recharge™ focuses on a 4-hour window beginning 30 min after sleep onset [52]. WHOOP’s Recovery and Stress Scores measure heart rate variability during the final slow-wave sleep period [47], 56], 57], however the classification accuracy of WHOOP’s actigraphy-derived sleep stage identification is unclear – it is widely acknowledged that the accuracy of consumer wearables for measuring sleep is poor [43]. In contrast, Samsung’s Energy Score is based on continuous heart rate variability monitoring [53], but details on its integration and calculation remain opaque; similarly, Ultrahuman’s Dynamic Recovery Score detects deviations from user-specific ‘baselines’ rather than absolute values [55] – but the specific formulae and sampling periods were unclear.

Challenges with continuous monitoring and signal validity

Compounding this issue is the fact that heart rate variability’s physiological link to parasympathetic activity is only valid under specific conditions, typically at rest, and artifacts introduced by daily activities such as swallowing, talking, or drinking water can “break” this relationship when heart rate variability is measured continuously [75], 76]. This makes composite metrics derived from 24/7 heart rate variability data, such as stress and recovery scores, scientifically questionable without proper contextualisation or validation.

The approach to calculating resting heart rate among the different manufacturers was similarly inconsistent. Fitbit derives resting heart rate from data collected during sleep or periods of inactivity throughout the day [44]. Garmin calculates resting heart rate as the lowest 30-min average within a 24-h period [50], while Oura combines nightly averages with the lowest resting heart rate sampled every 10 min [46]. Polar’s Nightly Recharge™ incorporates resting heart rate from a 4-hour window after sleep onset which is similar to their approach to measuring heart rate variability [52]. WHOOP, again calculates resting heart rate during the same actigraphy-derived slow-wave sleep stage as their heart rate variability calculation [47], 56], 57]. Samsung measures resting heart rate continuously during periods of inactivity but does not disclose thresholds or protocols [53], while Suunto allows users to input resting heart rate manually or rely on the lowest value recorded during sleep [54]. Taken together, the variability in algorithms, for measuring heart rate variability and resting heart rate bely any meaningful comparison between their outputs; differences in sampling protocols, timing, and integration strategies across platforms limit direct comparability between the composite indices.

Concerns regarding scientific robustness and real-world alignment

These discrepancies underscore broader uncertainties regarding the validity, interpretability, and real-world impact of each CHS. While marketed as objective indicators of readiness, recovery, or stress, the inconsistencies in their calculation methodologies – particularly in how and when physiological signals such as heart rate variability, resting heart rate, and activity levels are integrated – raise concerns about their scientific robustness. The research in this field is sparse, but the available studies highlight potential disconnects between CHS outputs and real-world physiological or psychological states.

For example, a 2023 study of elite swimmers investigated the relationship between wearable-derived physiological metrics and metabolic and psychological stress markers [63]. The study found a moderate positive correlation between heart rate variability and relative resting metabolic rate (r=0.448, p=0.032), suggesting that increased heart rate variability may reflect some aspects of metabolic function. However, no correlation was found between heart rate variability and thyroid hormone levels or the resting metabolic rate ratio, indicating that this wearable-derived measure does not consistently align with broader markers of energy availability. Furthermore, WHOOP’s recovery score showed no correlation with metabolic suppression when analyzing all participants together, and its relationship with the resting metabolic rate ratio was only significant in male swimmers (r=0.653, p=0.041), but not in females. The same study also explored the relationship between wearable-derived CHS metrics and self-reported stress levels. Heart rate variability was negatively correlated with sport-specific stress (r= −0.462, p=0.026) and total stress (r= −0.459, p=0.028), supporting its role as a sensitive marker of autonomic stress. However, WHOOP’s recovery score showed no significant correlation with self-reported stress or recovery measures in either sex, raising concerns about whether such CHS outputs provide meaningful insights into psychological stress or autonomic recovery.

Similarly, a 2025 study on Dutch police officers assessed the relationship between wearable-derived stress scores (including Garmin’s “Body Battery” metric) and subjective stress perception [64]. The study found that wearable-based stress metrics did not consistently align with users’ self-reported experiences of stress and recovery. While using a CHS-equipped wearable led to small but statistically significant improvements in stress awareness, self-efficacy, and well-being (Hedges’ g=0.25–0.46, p<0.05), many participants struggled to interpret their wearable-derived stress scores. Some perceived their stress levels as high despite their CHS metrics suggesting otherwise, and vice versa. Additionally, while CHS use increased awareness of physiological trends, this did not necessarily translate to significant behavioural changes, reinforcing concerns that CHS alone may be insufficient for guiding meaningful adaptations in stress regulation or recovery behaviours.

Multicollinearity and redundancy in CHS calculations

Taken together, these findings emphasize the need for independent validation of CHS to determine whether they provide physiologically and psychologically relevant insights or whether their outputs may mislead users due to misalignment with subjective experiences and established physiological markers. Indeed, when considered in isolation, a primary limitation of the CHS concept lies in the inherent interdependence of the core metrics that contribute to their overall scores. Many CHS exhibit an inherent bias due to multicollinearity among their input metrics, such as sleep duration and heart rate variability, which are often interdependent. For instance, a poor night’s sleep typically results in reduced heart rate variability, yet penalizing ‘readiness’ twice – once for low sleep quality and again for reduced heart rate variability – introduces redundancy and amplifies the negative score disproportionately. Conversely, if heart rate variability remains within normal ranges despite poor sleep, it suggests the user’s body has effectively mitigated the physiological impact of sleep disruption, making an additional penalty for poor sleep unnecessary. Such over-penalization creates a false perception of precision but often obscures the actual physiological response to stressors. This limitation highlights a deeper issue in the design of CHS: the tendency to conflate behavioural data with physiological outcomes, resulting in outputs that are not wholly reflective of the body’s actual state. Seventy-one percent of the indices integrated physical activity and ‘stress’ (inferred based on fluctuations in baseline heart rate taken throughout the day), and 71 % integrated some construct of sleep – often without adequately accounting for how an individual’s physiology responds to these factors. For instance, if a user undertakes a high-intensity workout, their physiological readiness for subsequent activity should ideally be assessed through a single measure like heart rate variability, which reflects how well the body has assimilated the stressor [71], 73]. Penalizing readiness solely on the basis of high activity data from the previous day, regardless of physiological recovery, undermines the purpose of measuring these signals. By relying on behavioural data to infer physiological states, CHS may fail to capture the nuanced variability in individual responses, limiting their utility for personalised health recommendations. Granted, several of the indices integrate ‘baseline’ measurements to contextualize short-term fluctuations – Fitbit, Garmin, Oura, Polar, Ultrahuman and WHOOP all incorporate user-specific baselines into their calculations – but the lack of consistency in defining and calculating baselines across platforms undermines our ability to compare indices or draw meaningful conclusions about their utility.

Lack of standardization and terminological ambiguity

Ultimately, our synthesis cannot shed light on many of these unknowns – and this reflects the primary limitation of this work, and the field in general: the lack of transparency and standardization among the CHS we evaluated. Despite their widespread adoption and growing role in consumer health monitoring [48], 65], the manufacturers provided limited details about the algorithms underpinning these tools, how often they change these algorithms, the rationale behind their contributors, or the validity of their outputs in diverse populations. Much of the data available for analysis was derived from publicly accessible resources, such as user manuals, white papers, and app interfaces, none of which offered the level of detail necessary to fully evaluate the scientific integrity of these indices.

A symptom of this lack of standardization is the ambiguity surrounding the terminology used in the different indices. For example, while multiple manufacturers – including Garmin, Fitbit, and Oura – use variations of the term “readiness”, their calculation methodologies and contributing metrics differ significantly. For instance, Fitbit’s Daily Readiness Score integrates heart rate variablity, resting heart rate, sleep quantity, and sleep architecture, whereas Garmin’s Training Readiness includes activity levels and recent training load, and Oura’s Readiness Score extends further to incorporate body temperature. Despite their terminological similarity, these indices are not methodologically equivalent, making direct comparisons difficult. Users may reasonably assume that readiness scores from different devices reflect comparable physiological constructs, yet differences in data sources, weightings, and interpretation frameworks introduce inconsistencies that challenge their scientific and practical applicability.

The net result of the lack of transparency it that key questions about how the different contributors are weighted, how composite scores are calculated, whether any given score can be considered superior in terms of validity, transparency, or scientific rigor, and whether these scores align with physiological realities remain unanswered. This lack of transparency has significant implications for the scientific reproducibility and clinical applicability of CHS. Without access to proprietary algorithms or validation datasets, researchers and clinicians cannot independently assess whether the scores produced by CHS are meaningful, reproducible, or generalizable. This is particularly concerning given the inherent variability in physiological responses among individuals and across populations. For example, metrics like heart rate variability are highly sensitive to factors such as age, sex, fitness level, and circadian rhythms [73], yet no evidence was found to suggest that these factors are consistently accounted for in the indices we reviewed. Similarly, the reliance on fixed baselines for assessing deviations in metrics like resting heart rate or heart rate variability may not adequately capture the dynamic, context-dependent nature of physiological responses, further limiting the utility of these tools.

The absence of rigorous validation in diverse populations also raises concerns about the equity of CHS. While some indices may perform well in the demographic groups most commonly represented in validation studies – typically young, healthy, and predominantly male participants [77], 78] – there is little evidence to suggest that these tools are reliable or accurate in populations with different characteristics. For instance, older adults, individuals with chronic conditions, and those from diverse racial or ethnic backgrounds may exhibit physiological patterns that differ from the normative baselines assumed by these algorithms. Without targeted validation efforts, CHS risk perpetuating biases that could undermine their utility for underserved populations, further exacerbating health inequities in digital health.

Implications for clinical, personal, and public health use

These limitations have significant implications for the clinical, telehealth, and personal utility of CHS. In remote health monitoring and telemedicine, where CHS could serve as potential digital biomarkers for assessing patient status, their lack of validation across diverse populations introduces risks of misclassification and inappropriate decision-making. Clinicians may be reluctant to rely on these scores for decision-making if the underlying methodologies are unclear or if the indices have not been validated across a sufficiently broad population. Moreover, without robust evidence to support the accuracy and reliability of CHS, their use in telehealth-based personalized health interventions or disease prevention programs may lead to inconsistent or inaccurate insights, potentially undermining trust among users and healthcare providers alike. From a personal health perspective, CHS have the potential to empower individuals by providing insights into their recovery, readiness, and overall health status. However, the utility of these tools is compromised when the outputs are overly complex, insufficiently validated, or fail to align with the user’s unique physiological profile. For example, indices that rely heavily on behavioural data may penalize individuals who deviate from normative patterns (e.g., athletes or shift workers), despite maintaining adequate physiological recovery. To maximize their utility, CHS must evolve to incorporate individualized baselines and adaptive algorithms that account for user-specific responses to stress, recovery, and activity. Future advancements in CHS may benefit from independent validation frameworks, industry-wide certification standards, or structured regulatory oversight to ensure their reliability and applicability without compromising proprietary algorithms.