Prior automatic posture and activity identification improves physical activity energy expenditure prediction from...

Area of Science:

• Exercise Physiology
• Biomedical Engineering
• Wearable Technology

Background:

• Accelerometry is a common tool for measuring physical activity energy expenditure (PAEE).
• Traditional linear regression models for accelerometry have limitations in accurately predicting PAEE due to not accounting for diverse activities.
• Accurate PAEE measurement is crucial for understanding the health impacts of physical activity.

Purpose of the Study:

• To evaluate an enhanced accelerometry model that couples an automatic activity recognition (AAR) algorithm with activity-specific count-based models.
• To compare the predictive accuracy of this AAR model against simple linear models and existing device software models for PAEE and total energy expenditure (TEE).
• To assess the model's performance in free-living conditions using hip-worn triaxial accelerometers.

Main Methods:

• Developed and validated an activity-specific model integrated with an AAR algorithm using laboratory data from 61 subjects.
• Tested the model in two free-living validation groups: 20 subjects on a 3-h urban circuit and 56 subjects over a 14-day period.
• Compared PAEE and TEE predictions from the AAR model against simple linear models (SLM), Freedson, and Actiheart models using reference measures like heart-rate monitoring and doubly-labeled water.

Main Results:

• The AAR model demonstrated closer agreement with reference measures for PAEE during the urban circuit compared to SLM and Freedson models (lower RMSE).
• In the 14-day trial, the AAR model achieved performance comparable to the Actiheart model for both PAEE and TEE predictions (similar RMSE).
• The AAR model explained 43% more variance in daily PAEE predictions compared to models relying solely on the relationship between accelerometry counts and energy expenditure.

Conclusions:

• The developed AAR model significantly improves the accuracy and precision of PAEE and TEE estimations from triaxial accelerometry in free-living settings.
• This activity-specific approach overcomes limitations of simpler models, offering a more robust method for assessing physical activity's impact on health.
• The enhanced model provides a valuable tool for researchers and health professionals utilizing wearable accelerometers for objective physical activity monitoring.