Smartwatch estimates of physical activity energy expenditure differed by device, while higher body fat percentage was associated with greater absolute and absolute percentage error among Hispanic adults, according to a study published in PLOS One. Investigators did not identify a significant association between Fitzpatrick skin type and absolute percentage error.
“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,” wrote lead author Jason Kostrna of the Department of Teaching and Learning, Florida International University in Miami, Florida, and colleagues.
Investigators enrolled 58 Hispanic adults aged 18 to 50 years with Fitzpatrick skin types III to V. Participants had a mean age of 23 years, and 31 (53%) were female. Thirty-three had a body mass index below 30, whereas 25 had a body mass index of 30 or greater. Fitzpatrick types III, IV, and V represented 52%, 41%, and 7% of the cohort, respectively, with 4 participants classified as type V.
Participants simultaneously wore 5 wrist-worn devices. Four generated physical activity energy expenditure (PAEE) estimates: Apple Watch Series 8, Fitbit Sense 2, Samsung Galaxy Watch 5, and Garmin Forerunner 955. Device-derived PAEE was evaluated using the COSMED K5 metabolic system as the indirect calorimetry criterion. The protocol included 5 minutes of rest, three 2-minute moderate-intensity intervals at 64% to 76% of maximum heart rate, two 2-minute vigorous-intensity intervals at 77% to 95%, and 5 minutes of recovery.
Body fat percentage derived from bioelectrical impedance analysis was used in the primary analyses. Investigators used self-reported Fitzpatrick classification as the primary skin-tone measure because colorimetry and spatial frequency domain spectroscopy measurements were considered unreliable under the experimental conditions.
Investigators calculated directional bias, absolute error, and absolute percentage error and used mixed-effects models incorporating device, body fat percentage, Fitzpatrick skin type, and pairwise interactions. The models included participant-level random intercepts to account for repeated observations from the same participant.
Following data-quality filtering, analyses included 52 Apple, 51 Garmin, 50 Samsung, and 44 Fitbit participant-device pairings. Mean bias was 22 kcal for Apple, 69 kcal for Garmin, 57 kcal for Samsung, and 3 kcal for Fitbit. Mean absolute percentage error was 36% for Apple, 78% for Garmin, 66% for Samsung, and 32% for Fitbit. Garmin and Samsung produced higher energy expenditure estimates relative to indirect calorimetry, while the degree of overestimation was smaller with Apple.
Fitbit produced 7 implausible estimates exceeding 450% of the criterion value, which were excluded from the primary analysis. When those estimates were included, Fitbit mean bias increased to 129 kcal. Although outlier exclusion was planned a priori, the investigators reported that the numerical thresholds were finalized during data review rather than fully specified in advance.
Greater body fat percentage was associated with greater bias, absolute error, and absolute percentage error across devices, with device-specific differences in the magnitude of some associations. Investigators did not identify a significant association between Fitzpatrick skin type and absolute percentage error.
“Participants with higher body fat percentage demonstrated greater bias and reduced relative accuracy,” wrote the investigators.
Investigators noted that findings from the controlled cycling protocol may not extend to other activities or populations. The limited number of participants with Fitzpatrick type V skin also reduced the study’s ability to detect potential skin-tone effects. Simultaneous use of several devices may have influenced watch fit or sensor contact.
Disclosures: The study was supported by a National Science Foundation pilot grant awarded to Jason Kostrna through the Precise Advanced Technologies and Health Systems for Underserved Populations program. The researchers declared no financial or nonfinancial competing interests.
Source: PLOS One