Validation of the ActiGraph two-regression model for predicting energy expenditure

Area of Science:

• Exercise Physiology
• Biomedical Engineering
• Wearable Technology

Background:

• Accurate measurement of energy expenditure (EE) is crucial for understanding human metabolism and physical activity.
• Wearable accelerometers, like the ActiGraph GT1M, offer a practical method for monitoring physical activity and estimating EE in free-living conditions.
• Validation against criterion methods such as whole-room indirect calorimetry and doubly labeled water (DLW) is essential for establishing the reliability of these devices.

Purpose of the Study:

• To validate a two-regression model for predicting EE from ActiGraph GT1M accelerometer data.
• To assess the impact of a low-pass filter (LPF) on the accuracy of minute-to-minute EE predictions.
• To compare model-predicted EE with measurements from a whole-room indirect calorimeter and the DLW technique.

Main Methods:

• Thirty-four healthy volunteers participated in a 24-hour whole-room calorimetry study involving structured and self-selected activities.
• ActiGraph GT1M monitors were worn to collect activity counts.
• A subset of 22 volunteers underwent a 14-day free-living DLW protocol.

Main Results:

• The two-regression model overpredicted EE by 10.2% compared to the room calorimeter.
• The LPF significantly reduced errors in EE prediction but did not improve the accuracy of predicting time spent in different physical activity intensities.
• Predicted EE from both filtered and unfiltered models showed no significant difference compared to DLW-measured EE.

Conclusions:

• The validated two-regression model, particularly with the LPF, demonstrates good agreement for total EE estimation against criterion measures (room calorimeter and DLW).
• Despite good overall EE prediction, significant individual variability persists in accurately assessing time spent in sedentary and light-to-moderate physical activity intensities.