Frontiers in Digital Health · Nov 19, 2026

Improving IMU-based Energy Expenditure Estimation: The Importance of Including Age-Specific Training Data

By Werner L. Popp, Jan Stutz, Philipp A. Eichenberger et al.

Background: Accurate energy expenditure (EE) estimation models using inertial measurement units (IMU) during activities of daily living (ADL) in older adults are needed for research, tailored exercise interventions or early detection of health status decline. Still, validation studies have been performed mostly in younger adults. This study therefore aimed to determine if EE estimation in older adults is improved when models are trained with data from older or mixed cohorts rather than solely younger adults. Methods: 29 healthy, sedentary younger (age 18-35; n=14) and older men (age 50-80; n=15), were recruited. 34 defined ADLs, across four activity classes (sedentary, low-intensity, high-intensity, locomotion), were used to mimic real-world situations. Each participant performed 12 pseudo-randomly selected activities (8 min each, 5 min breaks). Activity metrics were recorded continuously with 10-degrees-of-freedom IMUs on the chest, hip, ankles, and wrists. EE was determined via indirect calorimetry using a portable ergospirometry device, as reference. Multiple linear regression (MLR) and k-nearest-neighbors (kNN) classifiers were used to develop and cross-validate EE estimation models. Results: Overall activity classification accuracy in older adults improved from 90.8% when trained on younger, to 95.6% when trained on older, and 95.5% on mixed-age adults. EE estimation (MLR-class-estimated model) in older adults improved from a mean absolute error (MAE) of 32.8% [95% CI 30.3, 35.9%] when trained on younger, to 14.2% [95% CI 11.7, 17.3%] when trained on older, and 13.9% [95% CI 12.3, 15.6%] when trained on mixed-age adults. When combining both age groups, the best EE estimation model (MAE 14.9% [95% CI 13.6, 16.4%]) included training data from both younger and older adults. Conclusion: This Phase I proof-of-concept study provides initial evidence to include age-matched or mixed-age training data for improved EE estimation in healthy, older men, highlighting that models trained exclusively on younger cohorts may have limited generalizability even to a healthy older cohort under laboratory conditions. Given the small, all-male sample and the exploratory nature of the laboratory validation, these results should not be viewed as definitive. Future studies are warranted to externally validate these models in free-living environments, across larger cohorts, and within more diverse populations, including women.

The paperJournal siteBiomarkers

AgingResearch.News