Physical activity phenotypes from volume-based versus intensity-based wearable metrics: discordant classification and divergent metabolomic, body composition, and cardiac magnetic resonance associations
4 affiliations
- 1Yong Loo Lin School of Medicine, National University of Singapore, Singapore
- 2National Heart Centre Singapore, Singapore
- 3Duke-NUS Medical School, Singapore
- 4Nanyang Technological University, Singapore
Abstract
BackgroundConsumer wearables report direct volume-based metrics (steps, distance, floors climbed) alongside device-derived metrics generated by manufacturer algorithms (activity-intensity minutes, energy expenditure). Both are widely used to characterize individuals into activity phenotypes, yet whether these representations yield concordant activity phenotypes and biological associations is unclear.ObjectiveTo compare activity phenotypes derived from direct accelerometry with those derived from device-derived intensity metrics, and to evaluate each against metabolomic, body composition, and cardiac magnetic resonance (CMR) phenotypes.MethodsUsing the SingHEART cohort, two hierarchical cluster analyses using Ward linkage yielded five phenotypes each for volume-based and intensity-based metrics. Agreement was quantified by the adjusted Rand index. For analyses, the two high-variability phenotypes were combined, giving four groups: sedentary, average movers, regular exercisers, and high-variability movers. Serum ¹H-NMR metabolomics was analyzed by PERMANOVA with Holm adjustment across pairwise contrasts. Body and CMR associations were assessed using linear regression with false discovery rate correction.ResultsIn total, 984 adults (mean age 45.6 years, 48.3% male) wore accelerometers for a mean of 7.1 valid days. The two classifications agreed poorly: adjusted Rand index was 0.18, with only 46.2% assigned identically. Volume-based phenotypes differed in metabolomic profiles (P < .001). Active participants (high-variability movers plus regular exercisers) differed from sedentary participants across 106 of 250 biomarkers (FDR q < 0.05), including higher HDL, lower VLDL, LDL and lower GlycA. Conversely, although intensity-based phenotypes showed a nominally significant omnibus metabolomic association (P = .003), no metabolomic difference was observed between active and sedentary groups defined by intensity-based phenotyping (P = .29). Across 16 body composition and CMR outcomes, volume-based phenotypes were associated with 11 and intensity-based phenotypes with 6, in divergent directions: volume-based high-variability movers showed lower body fat, higher skeletal muscle, and larger ventricular volumes and mass, whereas intensity-based high-variability movers showed higher body mass index and waist circumference.ConclusionsVolume-based metrics and intensity-based metrics assign the same individuals to discordant activity phenotypes. Volume-based phenotypes showed more consistent associations across metabolomic, body composition, and cardiac imaging measures. These findings suggest that the choice of wearable-derived metrics substantially influences digital activity phenotyping and its biological interpretation.
Related
- Discordant muscle-adipose phenotypes and insulin resistance in prediabetes: sex-specific associations and predictive utility of TyG-WHtR combined with HDL-C
- Body Composition Phenotypes and Their Metabolic Associations in Treatment-Naïve Patients With Newly Diagnosed Type 2 Diabetes: A DXA-Based Cluster Analysis.
- Shared mediators, divergent networks: a tissue-response hypothesis for discordant phenotypes in chronic spontaneous urticaria and atopic dermatitis
- Proteomic and metabolomic correlates of sociodemographic factors and their associations with kidney and cardiovascular outcomes: findings from the Chronic Renal Insufficiency Cohort
- TyG-Biomarker discordance profiling identifies metabolically heterogeneous phenotypes with divergent cardiovascular and diabetes risk: metabolomic characterization from the UK biobank
BibTeX
@article{goh2026physical,
title = {{Physical activity phenotypes from volume-based versus intensity-based wearable metrics: discordant classification and divergent metabolomic, body composition, and cardiac magnetic resonance associations}},
author = {Sarah Shi Sien Goh and Khung Keong Yeo and Khung Keong Yeo and Jonathan Yap and Jonathan Yap and Siew Ching Kong and Weng Khong Lim and Mark Kei Fong Wong and Eddie Yin Kwee Ng and Xiayan Shen and Alex Weixian Tan and Weiting Huang},
journal = {Frontiers in Digital Health},
year = {2026},
month = oct,
doi = {10.3389/fdgth.2026.1982891},
url = {https://doi.org/10.3389/fdgth.2026.1982891},
note = {CC BY},
}
RIS
TY - JOUR
TI - Physical activity phenotypes from volume-based versus intensity-based wearable metrics: discordant classification and divergent metabolomic, body composition, and cardiac magnetic resonance associations
AU - Sarah Shi Sien Goh
AU - Khung Keong Yeo
AU - Khung Keong Yeo
AU - Jonathan Yap
AU - Jonathan Yap
AU - Siew Ching Kong
AU - Weng Khong Lim
AU - Mark Kei Fong Wong
AU - Eddie Yin Kwee Ng
AU - Xiayan Shen
AU - Alex Weixian Tan
AU - Weiting Huang
PY - 2026
DA - 2026/10/26
JO - Frontiers in Digital Health
DO - 10.3389/fdgth.2026.1982891
UR - https://doi.org/10.3389/fdgth.2026.1982891
AB - BackgroundConsumer wearables report direct volume-based metrics (steps, distance, floors climbed) alongside device-derived metrics generated by manufacturer algorithms (activity-intensity minutes, energy expenditure). Both are widely used to characterize individuals into activity phenotypes, yet whether these representations yield concordant activity phenotypes and biological associations is unclear.ObjectiveTo compare activity phenotypes derived from direct accelerometry with those derived from device-derived intensity metrics, and to evaluate each against metabolomic, body composition, and cardiac magnetic resonance (CMR) phenotypes.MethodsUsing the SingHEART cohort, two hierarchical cluster analyses using Ward linkage yielded five phenotypes each for volume-based and intensity-based metrics. Agreement was quantified by the adjusted Rand index. For analyses, the two high-variability phenotypes were combined, giving four groups: sedentary, average movers, regular exercisers, and high-variability movers. Serum ¹H-NMR metabolomics was analyzed by PERMANOVA with Holm adjustment across pairwise contrasts. Body and CMR associations were assessed using linear regression with false discovery rate correction.ResultsIn total, 984 adults (mean age 45.6 years, 48.3% male) wore accelerometers for a mean of 7.1 valid days. The two classifications agreed poorly: adjusted Rand index was 0.18, with only 46.2% assigned identically. Volume-based phenotypes differed in metabolomic profiles (P < .001). Active participants (high-variability movers plus regular exercisers) differed from sedentary participants across 106 of 250 biomarkers (FDR q < 0.05), including higher HDL, lower VLDL, LDL and lower GlycA. Conversely, although intensity-based phenotypes showed a nominally significant omnibus metabolomic association (P = .003), no metabolomic difference was observed between active and sedentary groups defined by intensity-based phenotyping (P = .29). Across 16 body composition and CMR outcomes, volume-based phenotypes were associated with 11 and intensity-based phenotypes with 6, in divergent directions: volume-based high-variability movers showed lower body fat, higher skeletal muscle, and larger ventricular volumes and mass, whereas intensity-based high-variability movers showed higher body mass index and waist circumference.ConclusionsVolume-based metrics and intensity-based metrics assign the same individuals to discordant activity phenotypes. Volume-based phenotypes showed more consistent associations across metabolomic, body composition, and cardiac imaging measures. These findings suggest that the choice of wearable-derived metrics substantially influences digital activity phenotyping and its biological interpretation.
SN - 2673-253X
ER -