https://orcid.org/0000-0002-4838-7094
Figures
Abstract
Smart watches are commonly used to provide continuous feedback on activity and caloric expenditure and are leveraged for weight management, clinical decisions, and public health strategies. Most wrist-worn wearables combine photoplethysmography, accelerometry, and proprietary algorithms to estimate caloric expenditure. Prior research indicates significant errors, yet the roles of potential moderators, specifically skin tone and body fat percentage (BF%), remain insufficiently examined. Therefore, the primary objective of this study was to quantify the accuracy of smartwatch-derived physical activity energy expenditure (PAEE) estimates relative to indirect calorimetry and to examine whether error varies by device brand, body fat percentage, and skin tone. We tested whether brand, BF%, and Fitzpatrick skin type (III–V) predict caloric expenditure error versus indirect calorimetry. Hispanic adults (n = 58) completed a single 10-minute recumbent-cycle protocol with alternating 2-minute intervals at ~64–76% and ~77–95% HRmax (Tanaka formula), bracketed by 5-minute rest/recovery. Participants wore Apple Watch Series 8, Fitbit Sense 2, Samsung Galaxy Watch 5, and Garmin Forerunner 955; COSMED K5 metabolic system provided the criterion. After device-specific data quality filters, analyzable participant–device pairings were Apple = 52, Garmin = 51, Samsung = 50, Fitbit = 44. One-sample tests indicated significant mean bias for three of four devices, p < .05. Importantly, the non-significant Fitbit bias depended on device-specific outlier removal. Bias (M, SD) and 95% CI (kcal): Apple 21.60 (36.63), 11.59–31.60; Garmin 68.61 (55.86), 53.28–83.94; Samsung 56.76 (42.03), 45.11–68.41; Fitbit 3.14 (40.95), −8.96 to 15.24. Mixed-effects models showed a device main effect (p < .001), a BF% main effect (p < .01), and a device by BF% interaction (p = .02): Physical activity energy expenditure (PAEE) error increased with adiposity across all brands (p < .01). Common smart watches substantially misestimate PAEE relative to indirect calorimetry, with error magnitude increasing as BF% rises and varying by brand. Current consumer devices do not yet provide reliable caloric monitoring for individuals or for research; improving accuracy across body types is essential for clinical and public health applications.
Citation: Kostrna J, Oparina E, Palacios C, Rodriguez AJ, Pei J, Ajmal A, et al. (2026) Body fat, skin tone, and the accuracy of smartwatch caloric expenditure estimates. PLoS One 21(7): e0353261. https://doi.org/10.1371/journal.pone.0353261
Editor: Hidetaka Hamasaki, Japanese Academy of Health and Practice, JAPAN
Received: January 15, 2026; Accepted: June 14, 2026; Published: July 29, 2026
Copyright: © 2026 Kostrna et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The datasets generated and/or analyzed during the current study are available in the OSF repository, https://osf.io/e745u/overview.
Funding: This research was supported by a pilot grant awarded to J.K. from the National Science Foundation (NSF) via the Precise Advanced Technologies and Health Systems for Underserved Populations (PATHS-UP) program project number 800022921. The sponsors did not play any role in n the study design, data collection and analysis, decision to publish, or preparation of the manuscript. https://pathsup.org/.
Competing interests: All authors declare no financial or non-financial competing interests.
Introduction
Wearable fitness trackers have revolutionized personal health monitoring by enabling convenient and continuous estimates of physical activity energy expenditure (PAEE). With over a hundred million users adopting these technologies, their accuracy has become a paramount concern for both research and healthcare [1]. Accurate PAEE is central to understanding energy balance, guiding weight management, and informing exercise prescriptions [2]. Errors in these estimates can mislead individuals attempting to lose weight, distort clinical recommendations, undermine exercise prescriptions, provide inaccurate information for healthcare decisions, and compromise the validity of population-level research [3,4]. Given the global rise in the prevalence of obesity, with nearly half of the U.S. population projected to have overweight or obesity within the next decade, ensuring that wearables provide reliable PAEE estimates is a pressing scientific and public health concern [5].
Most commercial wearables derive PAEE from heart rate (HR) measured via Photoplethysmography (PPG), the optical method underlying most wrist-worn wearables, often paired with accelerometry and proprietary demographic-based algorithms [6]. While PPG offers practical advantages, its reliance on light–tissue interaction makes it highly susceptible to error [4]. Skin tone and adiposity are two key factors that can bias PPG-based measurements [2]. Melanin in darker skin absorbs more of the green light used by common sensors, attenuating the pulsatile signal and contributing to higher error rates in HR detection [4,7]. Likewise, individuals with higher body mass index (BMI) often exhibit thicker, less perfused skin, which diminishes the intensity of light returning to the sensor [4,8]. Recent modeling studies suggest that obesity may have an even greater impact on PPG signal quality than skin pigmentation. Simulations show up to a 60% reduction in signal amplitude under conditions of severe obesity, compared to a 15% reduction attributable to increased melanin [7]. Consistent with these findings, empirical evaluations have shown that wearable HR and energy expenditure readings tend to be less accurate in individuals with darker skin tones or higher BMIs [2]. Because these same groups experience disproportionate burdens of obesity and cardiovascular disease, systematic inaccuracies in PAEE estimation have implications for health equity and measurement validity [9].
Physical exercise itself poses additional challenges to the accuracy of wearable devices. PPG measurements become notably less reliable during activity – for example, one study observed roughly 30% higher HR errors during exercise compared to rest [3]. Repetitive motion can induce motion artifacts, sometimes causing the device to mistake rhythmic movement for cardiac pulsations (a “signal crossover” effect). Since caloric expenditure rises steeply with intensity, such distortions can bias PAEE estimates precisely when they matter most. Proprietary differences in device algorithms and sensor design introduce another layer of variability, with comparative studies reporting an error of more than 20% in PAEE across widely used models [2]. These inconsistencies highlight the need for rigorous and systematic validation of wearables across a broad range of users and conditions.
To date, however, comprehensive evaluations of PPG-based energy expenditure accuracy across devices, skin pigmentation, body composition, and exercise intensities remain scarce. Here, we address this gap by assessing the PAEE estimation accuracy of four popular wrist-worn devices in a demographically varied cohort under controlled exercise conditions. We specifically test the hypothesis that darker skin tones and higher BMI are associated with greater errors in PAEE estimation, particularly during moderate to vigorous exercise, and that these errors vary across different wearable devices. This study aims to clearly delineate the physiological and technical limitations of current wearables and to inform the more equitable and reliable use of PPG technology in health monitoring.
Method
Participants
The study’s procedures were approved by the Florida International University Institutional Review Board (IRB-22–0471; approved October 12, 2022, recruitment starting date January 9, 2023, completion date September 23, 2025) and conducted in accordance with the Declaration of Helsinki [10]. Participants were recruited through flyers placed on campus, targeted departmental email invitations, and through snowball recruitment. To encourage participation, students enrolled in Physical Education: Sports and Fitness and Kinesiology courses were offered extra credit opportunities, with alternative assignments provided for students who chose not to participate. In addition, all participants received a $50 Target gift card as compensation for their time. Participation was voluntary, and alternative assignments were offered to minimize any perceived coercion. The recruitment strategy (flyers and departmental emails) was established a priori and was not altered during data collection.
Recruitment was stratified by body mass index (BMI, kg/m2), using World Health Organization cut‑point for obesity (30.0 kg/m2), and skin tone, using Fitzpatrick skin tone [11], to ensure adequate representation across participant groups. Participants were eligible if they were Hispanic between the ages of 18 and 50 years with Fitzpatrick skin types III to V [11]. Exclusion criteria included failure to meet these demographic requirements, medical contraindications for exercise as determined by the Physical Activity Readiness Questionnaire (PAR-Q+) [12], or the presence of forearm tattoos that could interfere with optical sensor measurements.
A priori power analysis indicated that 52 participants were required to sufficiently power a multivariate analysis of variance (MANOVA) with two groups and two dependent variables (absolute error and directional error) at α = .05, β = .80, and an effect size of f²(v) =.20 [3]. To account for attrition and data loss arising from equipment failure or measurement invalidity, the target sample size was increased to 58 participants. Because early volunteers tended to have a BMI below 30, initial enrollment was weighted toward participants within this lower BMI range. Therefore, recruitment in this category was capped at 33 participants; thereafter, only individuals with a BMI of 30 or greater were recruited until the final target sample of 58 was reached.
Design
The study employed a mixed factorial design with a within‑subject factor of device (Apple, Garmin, Samsung, Fitbit) and between‑subject predictors of body fat percentage and Fitzpatrick skin type. Participants were randomly assigned to one of several wearable device placement conditions following a Latin‑square schedule to minimize potential interference. They completed one exercise bout on a recumbent cycle ergometer following a standardized incremental intensity protocol adapted from the American National Standards Institute Consumer Technology Association testing protocol [13].
Measures
Energy expenditure was assessed using a COSMED K5 portable metabolic analyzer (COSMED, Rome, Italy), which provides breath-by-breath gas exchange analysis [14] and is accompanied by a Polar H10 chest strap electrocardiogram (ECG; Polar Electro, Kempele, Finland). Participants simultaneously wore five commercial wrist-worn devices: Apple Watch Series 8 (Apple Inc., Cupertino, CA), Fitbit Sense 2 (Google, Mountain View, CA), Samsung Galaxy Watch 5 (Samsung Electronics, Seoul, South Korea), Garmin Forerunner 955 (Garmin, Olathe, KS), and Empatica E4 (Empatica Inc., Cambridge, MA). The Empatica E4 (Empatica Inc., Cambridge, MA) measured only heart rate and not PAEE and was therefore excluded from analysis. Device placement on the right and left wrists was randomized using a Latin‑square schedule, devices were separated by at least 3 cm to minimize interference.
Anthropometric measurements included height, assessed with a stadiometer, weight, measured with a calibrated digital scale, and body fat percentage (BF%), determined by both skinfold calipers [15] and handheld bioelectrical impedance analysis (BIA; Omron HBF-306C; Omron Healthcare, Kyoto, Japan). BF% derived from BIA was used in all primary analyses due to its standardized protocol and lower operator dependence; skinfold-derived BF% was collected as a secondary measure. Wrist and forearm circumferences were measured with a flexible tape. Skin tone was evaluated using three methods: self-report via the Fitzpatrick Skin Type classification [11], skin reflectance colorimetry with a Nix Pro 2 colorimeter (Nix Sensor Ltd., Hamilton, Ontario, Canada), and objective imaging with an experimental Spatial Frequency Domain Spectroscopy (SFDS) system [16]. Because the Nix colorimeter yielded inconsistent and physiologically implausible readings and SFDS measurements were affected by inconsistent phantom calibration, ambient light contamination, and variability in acquisition timing, these methods were deemed unreliable under our experimental conditions; therefore, participant-reported Fitzpatrick classification was used as the primary skin tone measure. Participants also completed a demographic and medical history questionnaire, along with the PAR-Q+ [12], to confirm eligibility and safety for exercise testing.
Procedure
Each laboratory session lasted approximately 60 minutes and was conducted in a controlled environment at Florida International University’s kinesiology lab. Upon arrival, participants provided written informed consent and completed demographic, medical history, and PAR-Q+ questionnaires. Anthropometric data, including height, weight, BMI, body fat percentage, and forearm dimensions, were then collected. Skin tone was subsequently assessed using Fitzpatrick classification, a colorimeter, and SFDS imaging.
Following initial assessments, participants were randomly assigned to one of the wearable device placement conditions. All wearable devices were fitted to the wrists, then the Polar H10 chest strap and COSMED K5 analyzer were secured. Participant demographic profiles (age, sex, height, and weight) were entered identically across all devices prior to testing. Participants then completed a cycling session on a recumbent ergometer. The bout began with a 5-minute rest period followed by a sequence of alternating intensity intervals. Cycling intensity alternated between moderate (64–76% of maximum heart rate) and vigorous (77–95% of maximum heart rate), with maximum heart rate estimated using the Tanaka formula (HRmax = 208–0.7 × age) [17]. Each intensity phase lasted two minutes, with the cadence maintained at 85 revolutions per minute, while the experimenter adjusted the resistance to achieve the target heart rate (Fig 1). At the end of the exercise, a 5-minute rest period was provided to allow the heart rate to return to baseline levels. Each watch’s relevant indoor cycling mode was started at the beginning of the rest phase and ran through the recovery period with total caloric expenditure recorded. At the conclusion of the exercise protocol, participants were debriefed, thanked for their time, and compensated. All procedures were preregistered on OSF: https://osf.io/e745u/overview.
Fig 1 presents the cycling protocol used during the exercise trial.
After a 5-minute rest period, participants alternated between 2-minute intervals of moderate-intensity cycling (64–76% of their maximum heart rate, HRmax) and vigorous-intensity cycling (77–95% of their maximum heart rate, HRmax). The sequence consisted of three moderate intervals and two vigorous intervals, followed by a 5-minute recovery period.
Data analysis
Device-derived physical activity energy expenditure (PAEE) was evaluated against indirect calorimetry (K5 used as criterion).
Three complementary error metrics were computed to quantify device accuracy:
- Bias = Device − K5
- Absolute Error (AE) = |Bias|
- Absolute Percentage Error (APE) = 100·AE/K5
Bias represents the directional error in kilocalories (kcal), AE indicates the magnitude of error in kcal, and APE expresses the percentage magnitude error scaled to the criterion measure.
Outlier exclusion was planned a priori, but the exact numerical boundaries were finalized during data review rather than fully specified in advance. In that review, values of ≤1 kcal or >450% of the K5 estimate were deemed clearly implausible for the present protocol and were treated as device errors. We applied these thresholds uniformly across devices and report sensitivity analyses with and without these exclusions to make their influence transparent (sensitivity analysis available on OSF at: https://osf.io/e745u/overview. After applying these filters, the number of analyzable participant–device pairings was Apple = 52, Garmin = 51, Samsung = 50, Fitbit = 44. No other participant-level exclusions were applied.
Descriptive statistics (mean, SD, median, interquartile range, and 95% confidence intervals) were computed for each device. Directional error was evaluated using one-sample t-tests to test whether mean device bias differed from zero. Mixed-effects models incorporated all pairwise interactions among device, z-scored body fat percentage, and Fitzpatrick type, as prespecified in the preregistration.
Inferential models tested predictors of error using linear mixed‑effects models with log(APE + 1), log(AE + 1), and bias (untransformed) as the outcomes, device (4 levels), body fat percentage (BF%; z‑scored), Fitzpatrick skin type (categorical), and all pairwise interactions as fixed effects; participant was modeled as a random intercept to account for within‑person dependence. Absolute error (AE) and absolute percentage error (APE) were log-transformed to reduce right skewness and improve conformity with the model’s normality assumptions. Type III tests used Satterthwaite approximations (lmerTest). Model fit was summarized with marginal/conditional R².
Model diagnostics included residual plots, formal tests for heteroscedasticity and normality, and influence checks. Diagnostics indicated homoscedastic error variance but non‑normal residuals. To safeguard inference, we supplemented conventional SEs with participant‑clustered robust standard errors, which did not alter the significance pattern of primary effects or interactions (clubSandwich CR2). Influence was examined via Cook’s distances (influence.ME) and leave‑one‑participant‑out refits, which did not result in changes to the significance outcomes of our models.
For interpretation, we obtained estimated marginal means (EMMs) and simple slopes (emmeans). EMMs were computed at BF% = mean (z = 0) and, when appropriate, collapsed across Fitzpatrick skin types using the observed sample distribution; for log‑models, predictions were back‑transformed (exp(pred) − 1) to APE (%) or AE (kcal). Multiple comparisons were controlled using the Benjamini–Hochberg false discovery rate. Visualization included violin and box plots for bias, AE, and APE, along with model-based curves illustrating BF% by device effects with 95% confidence intervals. Raw and cleaned data, along with relevant codes for cleaning and analysis, are available on OSF: https://osf.io/e745u/overview.
Results
The final sample consisted of 58 participants (31 females, 53.4%; 27 males, 46.6%), with a mean age of 23 years (SD = 5.91). The average BMI for the total sample was 29.73 (SD = 7.43). Among participants with BMI < 30 (n = 33), the mean BMI was 24.24 (SD = 3.41), whereas participants with BMI ≥ 30 (n = 25) had a mean BMI of 36.97 (SD = 4.43). The distribution of Fitzpatrick skin types was as follows: 51.7% type III (n = 30), 41.4% type IV (n = 24), and 6.9% type V (n = 4).
Device-level PAEE error
Across all participants, device estimates showed systematic error relative to indirect calorimetry (K5; COSMED, Rome, Italy). Descriptive analyses indicated that mean bias was significantly different from zero for three of the four devices (one-sample t-tests, all p < .05), confirming consistent directional error. Garmin and Samsung systematically overestimated energy expenditure, while Apple’s overestimations were noticeably less (Table 1). Fitbit was the only device without a significant mean bias; however, this reflected a misleading average. Fitbit produced seven implausible estimates (> 450% of K5 values), representing approximately 13% of Fitbit trials. When these outliers are included, Fitbit’s mean bias rises to 128.6 kcal, by far the largest of any device. Fitbit also failed to generate caloric estimates in one session (Garmin = 1, Samsung = 2, Apple = 0). The absolute error (AE) averaged between ~80–110 kcal, depending on the device, and the absolute percentage error (APE) was consistently high, with median values ranging from ~15–25%. Distributions of bias, AE, and APE are visualized in Fig 2, where violin and boxplots illustrate both magnitude and directionality of device-level errors. Device placement did not have a significant impact on any of the analyses (all p values > .05).
Fig 2 shows the distribution of error metrics in physical activity energy expenditure estimates (PAEE) across four consumer wearable devices. Panel (a) shows bias (mean signed error, kcal), panel (b) shows absolute error (AE, kcal), and panel (c) shows absolute percentage error (APE, %). Each violin plot displays the full distribution, boxplots represent interquartile ranges with medians, and dots represent individual participant values.
Effects of body fat percentage and Fitzpatrick skin tone on PAEE APE
Mixed-effects models of log-transformed APE revealed significant main effects of device (p < .001) and body fat percentage (BF%; p = .01), as well as a device × BF% interaction (p = .02). Specifically, higher BF% was associated with greater APE across all devices, but slopes differed in magnitude: errors increased more steeply for Fitbit and Garmin compared to Apple (Fig 3). Fitzpatrick skin type showed no robust main effect (p = .89). Full model statistics are presented in supplemental data (see PAEE_LMM_results.xlsx on OSF: https://osf.io/e745u/overview). Regressions of BF% by device by error are also provided (see PAEE_BF_Regression_Untransformed.xlsx on OSF: https://osf.io/e745u/overview).
Effects of body fat percentage and Fitzpatrick skin tone on PAEE AE
Mixed-effects models of log-transformed AE revealed significant main effects of device (p < .001) and body fat percentage (BF%; p = .01), with a significant device by BF% interaction (p = .02). At the mean BF%, back-transformed predicted AE was lowest for Apple (M = 25.15 kcal, SE = 4.59) and Fitbit (M = 18.82 kcal, SE = 3.56) and highest for Garmin (M = 47.86 kcal, SE = 8.60) and Samsung (M = 40.98 kcal, SE = 8.42). As BF% increased, AE rose across all devices, with steeper slopes for Garmin and Samsung relative to Apple (Fig 3). This suggests that the magnitude of error is systematically higher for some devices, particularly at higher adiposity levels. Complete model statistics are presented in supplemental data (see PAEE_AE_Bias_backtrans.xlsx on OSF: https://osf.io/e745u/overview).
Effects of body fat percentage and Fitzpatrick skin tone on PAEE bias
Bias models showed significant main effects of device (p < .001) and BF% (p < .01). At the mean body fat percentage back-transformed bias, Apple displayed a modest positive bias (M = 22.77 kcal, SE = 7.78), while Garmin (M = 37.62 kcal, SE = 8.37) and Samsung (M = 35.68 kcal, SE = 8.46) significantly overestimated energy expenditure. In contrast, Fitbit demonstrated a negative bias (M = −19.60 kcal, SE = 8.68), indicating underestimation relative to the criterion after outlier removal. Importantly, BF% was positively associated with bias across devices, suggesting that higher adiposity systematically increased the tendency for devices to deviate from the gold standard (Fig 3). Complete model statistics are presented in supplemental data (see PAEE_AE_Bias_backtrans.xlsx on OSF: https://osf.io/e745u/overview).
Fig 3 shows each scatterplot displays regression lines with 95% confidence intervals (shaded areas) for body fat percentage predicting bias (top row), absolute error (middle row), and absolute percentage error (bottom row). Dots represent individual participant values. Regression equations, p-values, and R² statistics are provided within each panel. Complete descriptive and regression outputs are available in the supplemental material (see PAEE_AE_Bias_backtrans.xlsx on OSF: https://osf.io/e745u/overview).
Discussion
This study examined the accuracy of physical activity energy expenditure (PAEE) estimates from four popular wrist-worn wearables (Apple, Garmin, Samsung, and Fitbit) against indirect calorimetry during structured cycling. Three core findings emerged. First, all devices exhibited substantial PAEE error. This result is consistent with prior studies, which report that calorie estimates are inaccurate, with mean absolute percentage errors often exceeding 20% in controlled exercise tasks [2,18,19]. Furthermore, device-specific differences were evident: Apple produced the smallest bias, whereas Garmin and Samsung systematically overestimated PAEE. Fitbit exhibited the largest variability, with extreme outliers and occasional missing data, which raises concerns about its reliability for absolute PAEE estimation under the present protocol. Second, adiposity consistently moderates error. Participants with higher body fat percentage (BF%) demonstrated greater bias and reduced relative accuracy. These findings align with recent evidence showing that algorithms mapping heart rate and accelerometry to energy expenditure perform less effectively in individuals with elevated adiposity, in part because most models are calibrated on leaner populations and do not account for altered hemodynamics, body composition, or mechanical efficiency in individuals with higher BF% [20,21]. Increased soft tissue motion at the wrist and calibration ranges that underrepresent individuals with high body fat percentage likely compound these errors [7,22,23]. Third, we found no measurable effect of Fitzpatrick skin tone (III–V) on PAEE error. Although optical heart rate sensors are sensitive to pigmentation [24], the present data align with evidence suggesting that, when tested under standardized conditions, skin tone may exert a smaller or inconsistent influence on PAEE than adiposity [3,25,26]. This null finding should be interpreted with caution, given the limited sample size, especially regarding Fitzpatrick V, and does not dismiss potential disparities in free-living contexts.
From a user perspective, these results highlight practical risks. Large PAEE errors may mislead individuals attempting to manage their weight or energy balance, with these errors disproportionately affecting those with a higher body fat percentage, a population that is already at high cardiometabolic risk. Clinically and in public health research, uncorrected PAEE bias and outliers can obscure intervention effects and misclassify activity exposures, undermining the validity of epidemiological surveillance and applied studies that rely on wearable-derived metrics. These limitations have already prompted calls for phenotype-specific algorithms and greater transparency in commercial devices [21].
This study offers several strengths, including the use of a structured cycling protocol spanning moderate to vigorous intensities, a gold-standard metabolic criterion (COSMED K5) with documented validity, and evaluation of four major wearable brands. Nonetheless, several limitations should be acknowledged. First, the restriction to a single exercise modality may limit generalizability to free-living movement patterns. In more active modalities, inertial measurement unit (IMU) data can provide additional information for calculating PAEE. Second, the sample consisted of Hispanic adults with Fitzpatrick skin tones III–V, and the small number of participants in the Type V phenotypic subgroup constrained statistical power and may have masked potential effects. Third, because participants only completed a single exercise bout, we were unable to account for potential algorithmic learning effects, whereby devices adapt or recalibrate to users over repeated use, if such adaptations are included in proprietary estimations. Fourth, running the watches’ activity mode throughout the protocol, including the 5-minute initial rest and 5-minute recovery, may have introduced errors because participants were not actively cycling during those rest periods. In particular, APE may be affected by this difference as caloric expenditures will be lower than during active periods. Fifth, because participants wore multiple devices simultaneously, wrist fit, strap tension, and sensor contact may have been affected. Device placement was counterbalanced across wrists using a Latin-square design to reduce systematic bias, but some influence on signal quality cannot be ruled out. Finally, the proprietary nature of device algorithms prevents mechanistic attribution of error patterns and underscores the need for greater transparency from manufacturers. Therefore, these findings should be interpreted in the context of a controlled cycling protocol and a sample of Hispanic adults with Fitzpatrick skin types III–V; accuracy may differ across other activities and populations. At the same time, limited absolute caloric accuracy does not preclude potential utility for tracking relative or within-person changes over time, which warrants further study under free-living conditions.
Given the consistently high PAEE error, particularly among individuals with higher body fat percentages, improvements to the device or algorithm are needed. Future research should include approaches such as training models on datasets stratified by adiposity, sex, and age; enhancing motion-type detection to ensure activity-specific energy expenditure modeling; and integrating hybrid sensing modalities (e.g., multi-wavelength PPG, temperature) to improve accuracy.
Acknowledgments
The authors acknowledge the invaluable contributions of the undergraduate research team Sabrina Urban, Olivia Roeder, Brooke McCann, Albert Sirven, Alexis Smith, Christina Varela, Danay Romero, Lorena Velazquez Diaz, Ashley Khouly, Mariana Espinal, and Fabiana Marin Gallucci for their assistance with participant recruitment and data collection. Their enthusiasm, professionalism, and attention to detail were instrumental to the successful completion of this study.
References
- 1.
Statista. Wearables Statista. https://www.statista.com/topics/1550/wearables/. 2024.
- 2. Shcherbina A, Mattsson CM, Waggott D, Salisbury H, Christle JW, Hastie T, et al. Accuracy in wrist-worn, sensor-based measurements of heart rate and energy expenditure in a diverse cohort. J Pers Med. 2017;7(2):3. pmid:28538708
- 3. Bent B, Goldstein BA, Kibbe WA, Dunn JP. Investigating sources of inaccuracy in wearable optical heart rate sensors. NPJ Digit Med. 2020;3:18. pmid:32047863
- 4. Fine J, Branan KL, Rodriguez AJ, Boonya-Ananta T, Ramella-Roman JC, et al. Sources of inaccuracy in photoplethysmography for continuous cardiovascular monitoring. Biosensors (Basel). 2021;11(4):126. pmid:33923469
- 5. Ward ZJ, Bleich SN, Cradock AL, Barrett JL, Giles CM, Flax C, et al. Projected U.S. state-level prevalence of adult obesity and severe obesity. N Engl J Med. 2019;381(25):2440–50. pmid:31851800
- 6. Ray D, Collins T, Woolley S, Ponnapalli P. A review of wearable multi-wavelength photoplethysmography. IEEE Rev Biomed Eng. 2023;16:136–51. pmid:34669577
- 7. Boonya-Ananta T, Rodriguez AJ, Du Le VN, Ramella-Roman JC. Monte Carlo analysis of optical heart rate sensors in commercial wearables: The effect of skin tone and obesity on the photoplethysmography (PPG) signal. Biomed Opt Express. 2021;12(12):7445–57. pmid:35003845
- 8. Boonya-Ananta T, Rodriguez AJ, Ajmal A, Du Le VN, Hansen AK, Hutcheson JD, et al. Synthetic photoplethysmography (PPG) of the radial artery through parallelized Monte Carlo and its correlation to body mass index (BMI). Sci Rep. 2021;11(1):2570. pmid:33510428
- 9. Guha A, Wang X, Harris RA, Nelson A-G, Stepp D, Klaassen Z, et al. Obesity and the bidirectional risk of cancer and cardiovascular diseases in African Americans: Disparity vs. Ancestry. Front Cardiovasc Med. 2021;8:761488. pmid:34733899
- 10. World Medical Association. World Medical Association Declaration of Helsinki: Ethical principles for medical research involving human subjects. JAMA. 2013;310(20):2191–4. pmid:24141714
- 11. Fitzpatrick TB. The validity and practicality of sun-reactive skin types I through VI. Arch Dermatol. 1988;124(6):869–71. pmid:3377516
- 12. Warburton DE, Jamnik VK, Bredin SS, Gledhill N. The physical activity readiness questionnaire for everyone (PAR-Q) and electronic physical activity readiness medical examination (ePARmed-X). Health Fit J Can. 2011;4(2):3–17.
- 13.
American National Standards Institute/Consumer Technology Association. Standard for testing and labeling wearable consumer devices: CTA-2065. CTA. 2018.
- 14. McLaughlin JE, King GA, Howley ET, Bassett DR, Ainsworth BE. Validation of the COSMED K4 b2 portable metabolic system. Int J Sports Med. 2001;22(4):280–4. pmid:11414671
- 15. Jackson AS, Pollock ML. Generalized equations for predicting body density of men. Br J Nutr. 1978;40(3):497–504. pmid:718832
- 16. Cuccia DJ, Bevilacqua F, Durkin AJ, Ayers FR, Tromberg BJ. Quantitation and mapping of tissue optical properties using modulated imaging. J Biomed Opt. 2009;14(2):024012. pmid:19405742
- 17. Tanaka H, Monahan KD, Seals DR. Age-predicted maximal heart rate revisited. J Am Coll Cardiol. 2001;37(1):153–6. pmid:11153730
- 18. Le S, Wang X, Zhang T, Lei SM, Cheng S, Yao W, et al. Validity of three smartwatches in estimating energy expenditure during outdoor walking and running. Front Physiol. 2022;13:995575. pmid:36225296
- 19. Reddy RK, Pooni R, Zaharieva DP, Senf B, El Youssef J, Dassau E, et al. Accuracy of wrist-worn activity monitors during common daily physical activities and types of structured exercise: Evaluation study. JMIR Mhealth Uhealth. 2018;6(12):e10338. pmid:30530451
- 20. Jamieson A, Chico TJA, Jones S, Chaturvedi N, Hughes AD, Orini M. A guide to consumer-grade wearables in cardiovascular clinical care and population health for non-experts. NPJ Cardiovasc Health. 2025;2(1):44. pmid:40909206
- 21. Wei B, Romano C, Pedram M, Nolan B, Morelli WA, Alshurafa N. Developing and comparing a new BMI inclusive energy expenditure algorithm on wrist-worn wearables. Sci Rep. 2025;15(1):20060. pmid:40537505
- 22. Charlton PH, Marozas V, Mejía-Mejía E, Kyriacou PA, Mant J. Determinants of photoplethysmography signal quality at the wrist. PLOS Digit Health. 2025;4(6):e0000585. pmid:40577339
- 23. Freedson P, Bowles HR, Troiano R, Haskell W. Assessment of physical activity using wearable monitors: Recommendations for monitor calibration and use in the field. Med Sci Sports Exerc. 2012;44(1 Suppl 1):S1-4. pmid:22157769
- 24. Fallow BA, Tarumi T, Tanaka H. Influence of skin type and wavelength on light wave reflectance. J Clin Monit Comput. 2013;27(3):313–7. pmid:23397431
- 25. Icenhower A, Murphy C, Brooks AK, Irby M, N’dah K, Robison J, et al. Investigating the accuracy of Garmin PPG sensors on differing skin types based on the Fitzpatrick scale: Cross-sectional comparison study. Front Digit Health. 2025;7:1553565. pmid:40212900
- 26. Kostrna J, Oparina E, Palacios C, Rodriguez AJ, Pei J, Ajmal A, et al. PPG-based heart rate accuracy in hispanic adults with fitzpatrick III-V Skin Tones: An evaluation of body composition and skin-tone effects. Sensors (Basel). 2026;26(10):2922. pmid:42197732