Author(s): HEYDENREICH, J., LUTZ, H., MANZKE, C., OESCHGER, R., GILGEN-AMMANN, R., Institution: LEIPZIG UNIVERSITY, FACULTY OF SPORT SCIENCE, Country: GERMANY, Abstract-ID: 1465

INTRODUCTION:
Accurate assessment of total energy expenditure (TEE) is a key parameter in sport science, health research, and clinical practice. Body-worn activity sensors offer a practical and consumer-friendly approach for estimating TEE under free-living conditions; however, their validity is often insufficiently examined against the gold-standard doubly labeled water (DLW) method. Therefore, this study aimed to evaluate the validity of four body-worn activity sensors, including consumer wearables and a research-grade accelerometer, for assessing habitual TEE in daily life.
METHODS:
Fifty-six healthy adults (66% female; 29.4 ± 7.2 years, 22.9 ± 2.6 kg·m-2, estimated VO2max 49.5 ± 9.8 ml·kg-1·min-1) wore four activity sensors simultaneously at predefined and standardized body locations (Polar Ignite 3 [IG], Polar Verity Sense [VS], Garmin vivoactive 4 [VA], GENEActiv [GA]) during waking hours for 10 consecutive days under free-living conditions. Participants with at least 8 out of 10 valid days (≥12 h sensor wear per day), determined using a wear-time log, were included, resulting in sensor-specific sample sizes (IG: 51, VS: 41, VA: 47, GA: 41). For the wrist-worn consumer devices (IG and VA), TEE was derived directly from the manufacturers’ proprietary algorithms. For VS and GA, manufacturer algorithms were used to estimate energy expenditure (EE) during wear time only; non-wear EE was estimated based on resting metabolic rate (Harris & Benedict, 1918) and added to wear-time EE to derive daily TEE. DLW-derived TEE (TEEDLW) was calculated as mean daily TEE across the measurement period and served as the criterion measure. Statistical analyses included descriptive analysis (Mean±SD, MAPE), Wilcoxon signed-rank tests, Bland-Altman analyses, and linear regression (TEEDLW as dependent, sensor-derived TEE as independent variable; R², SEE; α=.05).
RESULTS:
Mean wear-time did not differ between sensors (~15 h·day⁻¹; p>.05), indicating that differences in TEE estimates were more likely algorithmic rather than compliance-related. VA and GA significantly underestimated TEE compared with TEEDLW (VA: 2528±457 kcal·d-1, GA: 2709±464 kcal·d-1, TEEDLW: 2930±697 kcal·d-1; p<.05), whereas no significant differences were observed for IG (3080±592 kcal·d-1) and VS (3003±632 kcal·d-1; all p>.05). Systematic bias [limits of agreement] amounted to -418 [-1334/498] kcal·d-1 for VA, -264 [-1273/746] kcal·d-1 for GA, 111 [-803/1025] kcal·d-1 for IG, and -4 [-1172/1164] kcal·d-1 for VS. MAPE ranged from 13.6–16.1%, with 31.7–56.9% of participants exhibiting a MAPE ≤10% across sensors. Associations with TEEDLW were moderate to high (R2 = 0.40–0.66), with substantial prediction errors (SEE: 448–587 kcal·d-1).
CONCLUSION:
Wearable activity sensors demonstrated moderate to high agreement with DLW-derived TEE at the group level. Nevertheless, substantial inter-individual variability and large limits of agreement limit their applicability for individual-level TEE assessment.