Summary
Introduction: This article is part of the Focus Theme of Methods of Information in Medicine on “Pervasive Intelligent Technologies for Health”.
Background: Energy Expenditure (EE) estimation algorithms using Heart Rate (HR) or a combination of accelerometer and HR data suffer from large error due to inter-person differences in the relation between HR and EE. We recently introduced a methodology to reduce inter-person differences by predicting a HR normalization parameter during low intensity Activities of Daily Living (ADLs). By using the HR normalization, EE estimation performance was improved, but conditions for performing the normalization automatically in daily life need further analysis. Sedentary lifestyle of many people in western societies urge for an in-depth analysis of the specific ADLs and HR features used to perform HR normalization, and their effects on EE estimation accuracy in participants with varying Physical Activity Levels (PALs).
Objectives: To determine 1) which low intensity ADLs and HR features are necessary to accurately determine HR normalization parameters, 2) whether HR variability (HRV) during ADLs can improve accuracy of the estimation of HR normalization parameters, 3) whether HR normalization parameter estimation from different ADLs and HR features is affected by the participants’ PAL, and 4) what is the impact of different ADLs and HR features used to predict HR normalization parameters on EE estimation accuracy.
Methods: We collected reference EE from indirect calorimetry, accelerometer and HR data using one single sensor placed on the chest from 36 participants while performing a wide set of activities. We derived HR normalization parameters from individual ADLs (lying, sedentary, walking at various speeds), as well as combinations of sedentary and walking activities. HR normalization parameters were used to normalized HR and estimate EE.
Results: From our analysis we derive that 1) HR normalization using resting activities alone does not reduce EE estimation error in participants with different reported PALs. 2) HRV features did not show any significant improvement in RMSE. 3) HR normalization parameter estimation was found to be biased in participants with different PALs when sedentary-only data was used for the estimation. 4) EE estimation error was not reduced when normalization was carried out using sedentary activities only. However, using data from walking at low speeds improved the results significantly (30–36%).
Conclusion: HR normalization parameters able to reduce EE estimation error can be accurately estimated from low intensity ADLs, such as sedentary activities and walking at low speeds (3 – 4 km/h), regardless of reported PALs. However, sedentary activities alone, even when HRV features are used, are insufficient to estimate HR normalization parameters accurately.
Keywords
Energy expenditure - heart rate - wearable sensors - personalization
