Energy Expenditure Prediction Models Research Articles

Cross-Validation and Comparison of Energy Expenditure Prediction Models Using Count-Based and Raw Accelerometer Data in Youth

Background: Machine learning may improve energy expenditure (EE) prediction from body-worn accelerometers. However, machine learning models are rarely cross-validated in an independent sample, and the use of machine learning raises additional questions including the effect of accelerometer placement and data type (count vs. raw) for optimal EE prediction. Purpose: To assess the accuracy of artificial neural network (ANN) models for EE prediction in youth using count-based or raw data from accelerometers worn on the hip, wrist, or in combination, and compare these to count-based, EE regression equations. Methods: Data were collected in two settings; one (n = 27) to calibrate the EE prediction models, and the other (n = 34) for model cross-validation. Participants wore a portable metabolic analyzer (EE criterion) and accelerometers on the left wrist and right hip while completing 30 minutes of exergames (calibration, cross-validation) and a maximal exercise test (calibration only). Six ANNs were created from the calibration data, separately by accelerometer placement (hip, wrist, combination) and data format (count-based, raw) to predict EE (15-second epochs). Three count-based linear regression equations were also developed for comparison to the ANNs. Results: The count-based, hip ANN demonstrated lower error (RMSE: 1.2 METs) than all other ANNs (RMSE: 1.7–3.6 METs) and EE regression equations (RMSE: 1.5–3.2 METs). However, all models showed bias toward the mean. Conclusion: An ANN developed for hip-worn accelerometers had higher accuracy for EE prediction during an exergame session than wrist or combination ANNs, and ANNs developed using count-based data had higher accuracy than ANNs developed using raw data.

Validation of Accelerometer-Based Energy Expenditure Prediction Models in Structured and Simulated Free-Living Settings

ABSTRACTThis study compared accuracy of energy expenditure (EE) prediction models from accelerometer data collected in structured and simulated free-living settings. Twenty-four adults (mean age 45.8 years, 50% female) performed two sessions of 11 to 21 activities, wearing four ActiGraph GT9X Link activity monitors (right hip, ankle, both wrists) and a metabolic analyzer (EE criterion). Visit 1 (V1) involved structured, 5-min activities dictated by researchers; Visit 2 (V2) allowed participants activity choice and duration (simulated free-living). EE prediction models were developed incorporating data from one setting (V1/V2; V2/V2) or both settings (V1V2/V2). The V1V2/V2 method had the lowest root mean square error (RMSE) for EE prediction (1.04–1.23 vs. 1.10–1.34 METs for V1/V2, V2/V2), and the ankle-worn accelerometer had the lowest RMSE of all accelerometers (1.04–1.18 vs. 1.17–1.34 METs for other placements). The ankle-worn accelerometer and associated EE prediction models developed using data from both structured and simulated free-living settings should be considered for optimal EE prediction accuracy.

Validation Of Accelerometer-based Energy Expenditure Prediction Models In Structured And Semi-structured Settings

Validation Of Accelerometer-based Energy Expenditure Prediction Models In Structured And Semi-structured Settings

Comparison of linear and non-linear models for predicting energy expenditure from raw accelerometer data

This study had three purposes, all related to evaluating energy expenditure (EE) prediction accuracy from body-worn accelerometers: (1) compare linear regression to linear mixed models, (2) compare linear models to artificial neural network models, and (3) compare accuracy of accelerometers placed on the hip, thigh, and wrists. Forty individuals performed 13 activities in a 90 min semi-structured, laboratory-based protocol. Participants wore accelerometers on the right hip, right thigh, and both wrists and a portable metabolic analyzer (EE criterion). Four EE prediction models were developed for each accelerometer: linear regression, linear mixed, and two ANN models. EE prediction accuracy was assessed using correlations, root mean square error (RMSE), and bias and was compared across models and accelerometers using repeated-measures analysis of variance. For all accelerometer placements, there were no significant differences for correlations or RMSE between linear regression and linear mixed models (correlations: r = 0.71–0.88, RMSE: 1.11–1.61 METs; p > 0.05). For the thigh-worn accelerometer, there were no differences in correlations or RMSE between linear and ANN models (ANN—correlations: r = 0.89, RMSE: 1.07–1.08 METs. Linear models—correlations: r = 0.88, RMSE: 1.10–1.11 METs; p > 0.05). Conversely, one ANN had higher correlations and lower RMSE than both linear models for the hip (ANN—correlation: r = 0.88, RMSE: 1.12 METs. Linear models—correlations: r = 0.86, RMSE: 1.18–1.19 METs; p < 0.05), and both ANNs had higher correlations and lower RMSE than both linear models for the wrist-worn accelerometers (ANN—correlations: r = 0.82–0.84, RMSE: 1.26–1.32 METs. Linear models—correlations: r = 0.71–0.73, RMSE: 1.55–1.61 METs; p < 0.01). For studies using wrist-worn accelerometers, machine learning models offer a significant improvement in EE prediction accuracy over linear models. Conversely, linear models showed similar EE prediction accuracy to machine learning models for hip- and thigh-worn accelerometers and may be viable alternative modeling techniques for EE prediction for hip- or thigh-worn accelerometers.

Energy expenditure estimation in beta-blocker-medicated cardiac patients by combining heart rate and body movement data.

Accurate assessment of energy expenditure provides an opportunity to monitor physical activity during cardiac rehabilitation. However, the available assessment methods, based on the combination of heart rate (HR) and body movement data, are not applicable for patients using beta-blocker medication. Therefore, we developed an energy expenditure prediction model for beta-blocker-medicated cardiac rehabilitation patients. Sixteen male cardiac rehabilitation patients (age: 55.8 ± 7.3 years, weight: 93.1 ± 11.8 kg) underwent a physical activity protocol with 11 low- to moderate-intensity common daily life activities. Energy expenditure was assessed using a portable indirect calorimeter. HR and body movement data were recorded during the protocol using unobtrusive wearable devices. In addition, patients underwent a symptom-limited exercise test and resting metabolic rate assessment. Energy expenditure estimation models were developed using multivariate regression analyses based on HR and body movement data and/or patient characteristics. In addition, a HR-flex model was developed. The model combining HR and body movement data and patient characteristics showed the highest correlation and lowest error (r2 = 0.84, root mean squared error = 0.834 kcal/minute) with total energy expenditure. The method based on individual calibration data (HR-flex) showed lower accuracy (i2 = 0.83, root mean squared error = 0.992 kcal/minute). Our results show that combining HR and body movement data improves the accuracy of energy expenditure prediction models in cardiac patients, similar to methods that have been developed for healthy subjects. The proposed methodology does not require individual calibration and is based on the data that are available in clinical practice.

Wrist-independent energy expenditure prediction models from raw accelerometer data

Purposes: (1) Develop artificial neural network (ANN) models for wrist accelerometer data which can predict energy expenditure (EE) using data collected from either wrist. (2) Develop ANNs for detecting the wrist on which the accelerometer was worn.Forty-four adults wore GENEActiv accelerometers on the left and right wrists and a portable metabolic analyzer while participating in a 90 min semi-structured activity protocol. Participants performed 14 sedentary, lifestyle, exercise, and ambulatory activities and were allowed to choose activity order, duration, and intensity. ANNs were created to predict EE and wrist detection using a leave-one-out cross-validation. In total, 12 combinations of feature sets (mean and variance of raw, vector magnitude, and absolute value data), training methods (left- and right- wrist), and testing methods (left- and right-wrist data) were used to develop EE prediction ANNs. Accuracy of the ANNs was evaluated using correlations, root mean square error (RMSE), and bias, using metabolic analyzer data as the criterion for EE.ANNs using raw data from the same wrist (e.g. EE predicted from right wrist ANNs using accelerometer data from right wrist) had the highest accuracy for EE prediction (r = 0.84, RMSE = 1.25–1.26 METs); conversely, opposite-wrist prediction accuracy (e.g. EE predicted from right wrist ANNs using accelerometer data from left wrist) was lower (r = 0.60–0.64, RMSE = 1.93–2.01 METs). Preprocessing into absolute values prior to ANN development allowed for, high EE prediction accuracy, with no difference in accuracy for same- versus opposite-wrist prediction (r = 0.80–0.83, RMSE = 1.30–1.49 METs). Wrist detection ANNs correctly determined wrist placement 100% of the time.Highly accurate, wrist-independent EE prediction ANNs were developed by computing absolute values of raw acceleration data prior to ANN development. This method provides a potential approach for advancing predictive accuracy of wrist-worn accelerometers.

Evaluation of custom energy expenditure models for SenseWear armband in manual wheelchair users.

Physical activity monitors are increasingly used to help the general population lead a healthy lifestyle by keeping track of their daily physical activity (PA) and energy expenditure (EE). However, none of the commercially available activity monitors can accurately estimate PA and EE in people who use wheelchairs as their primary means of mobility. Researchers have recently developed custom EE prediction models for manual wheelchair users (MWUs) with spinal cord injuries (SCIs) based on a commercial activity monitor--the SenseWear armband. This study evaluated the performance of two custom EE prediction models, including a general model and a set of activity-specific models among 45 MWUs with SCI. The estimated EE was obtained by using the two custom models and the default manufacturer's model, and it was compared with the gold standard measured by the K4b2 portable metabolic cart. The general, activity-specific, and default models had a mean signed percent error (mean +/- standard deviation) of -2.8 +/- 26.1%, -4.8 +/- 25.4%, and -39.6 +/- 37.8%, respectively. The intraclass correlation coefficient was 0.86 (95% confidence interval [CI] = 0.82 to 0.89) for the general model, 0.83 (95% CI = 0.79 to 0.87) for the activity-specific model, and 0.62 (95% CI = 0.16 to 0.81) for the default model. The custom models for the SenseWear armband significantly improved the EE estimation accuracy for MWUs with SCI.

Corrections to “Predicting Free-Living Energy Expenditure Using a Miniaturized Ear-Worn Sensor: An Evaluation Against Doubly Labeled Water” [Feb 14 566-575

Accurate estimation of daily total energy expenditure (EE) is a prerequisite for assisted weight management and assessing certain health conditions. The use of wearable sensors for predicting free-living EE is challenged by consistent sensor placement, user compliance, and estimation methods used. This paper examines whether a single ear-worn accelerometer can be used for EE estimation under free-living conditions. An EE prediction model as first derived and validated in a controlled setting using healthy subjects involving different physical activities. Ten different activities were assessed showing a tenfold cross validation error of 0.24. Furthermore, the EE prediction model shows a mean absolute deviation (MAD) below 1.2 metabolic equivalent of tasks. The same model was applied to a free-living setting with a different population for further validation. The results were compared against those derived from doubly labeled water. In free-living settings, the predicted daily EE has a correlation of 0.74, p 0.008, and a MAD of 272 kcal day. These results demonstrate that laboratory-derived prediction models can be used to predict EE under free-living conditions.

Predicting Free-Living Energy Expenditure Using a Miniaturized Ear-Worn Sensor: An Evaluation Against Doubly Labeled Water

Accurate estimation of daily total energy expenditure (EE)is a prerequisite for assisted weight management and assessing certain health conditions. The use of wearable sensors for predicting free-living EE is challenged by consistent sensor placement, user compliance, and estimation methods used. This paper examines whether a single ear-worn accelerometer can be used for EE estimation under free-living conditions.An EE prediction model as first derived and validated in a controlled setting using healthy subjects involving different physical activities. Ten different activities were assessed showing a tenfold cross validation error of 0.24. Furthermore, the EE prediction model shows a mean absolute deviation(MAD) below 1.2 metabolic equivalent of tasks. The same model was applied to a free-living setting with a different population for further validation. The results were compared against those derived from doubly labeled water. In free-living settings, the predicted daily EE has a correlation of 0.74, p 0.008, and a MAD of 272 kcal day. These results demonstrate that laboratory-derived prediction models can be used to predict EE under free-living conditions [corrected].

Physical activity classification utilizing SenseWear activity monitor in manual wheelchair users with spinal cord injury

Validation. The primary aim of this study was to develop and evaluate activity classification algorithms for a multisensor-based SenseWear (SW) activity monitor that can recognize wheelchair-related activities performed by manual wheelchair users (MWUs) with spinal cord injury (SCI). The secondary aim was to evaluate how the accuracy in activity classification affects the estimation of energy expenditure (EE) in MWUs with SCI. University-based laboratory. Forty-five MWUs with SCI wore a SW on their upper arm and participated in resting, wheelchair propulsion, arm-ergometery and deskwork activities. The investigators annotated the start and end of each activity trial while the SW collected multisensor data and a portable metabolic cart collected criterion EE. Three methods including linear discriminant analysis, quadratic discriminant analysis (QDA), and Naïve Bayes (NB) were used to develop classification algorithms for four activities based on the training data set from 36 subjects. The classification accuracy was 96.3% for QDA and 94.8% for NB when the classification algorithms were tested on the validation data set from nine subjects. The average EE estimation errors using the activity-specific EE prediction model were 5.3±21.5% and 4.6±22.8% when the QDA and NB classification algorithms were applied, respectively, as opposed to 4.9±20.7% when 100% classification accuracy was assumed. The high classification accuracy and low EE estimation errors suggest that the SW can be used by researchers and clinicians to classify and estimate the EE for the four activities tested in this study among MWUs with SCI.

Development and Validation of Energy Expenditure Prediction Models Based on GT3X Accelerometer Data in 5- to 9-Year-Old Children

Accelerometry has been established as an objective method that can be used to assess physical activity behavior in large groups. The purpose of the current study was to provide a validated equation to translate accelerometer counts of the triaxial GT3X into energy expenditure in young children. Thirty-two children aged 5-9 years performed locomotor and play activities that are typical for their age group. Children wore a GT3X accelerometer and their energy expenditure was measured with indirect calorimetry. Twenty-one children were randomly selected to serve as development group. A cubic 2-regression model involving separate equations for locomotor and play activities was developed on the basis of model fit. It was then validated using data of the remaining children and compared with a linear 2-regression model and a linear 1-regression model. All 3 regression models produced strong correlations between predicted and measured MET values. Agreement was acceptable for the cubic model and good for both linear regression approaches. The current linear 1-regression model provides valid estimates of energy expenditure for ActiGraph GT3X data for 5- to 9-year-old children and shows equal or better predictive validity than a cubic or a linear 2-regression model.

Predicting Energy Expenditure of Manual Wheelchair Users With Spinal Cord Injury Using a Multisensor-Based Activity Monitor

Hiremath SV, Ding D, Farringdon J, Cooper RA. Predicting energy expenditure of manual wheelchair users with spinal cord injury using a multisensor-based activity monitor. ObjectiveTo develop and evaluate new energy expenditure (EE) prediction models for manual wheelchair users (MWUs) with spinal cord injury (SCI) based on a commercially available multisensor-based activity monitor. DesignCross-sectional. SettingLaboratory. ParticipantsVolunteer sample of MWUs with SCI (N=45). InterventionSubjects were asked to perform 4 activities including resting, wheelchair propulsion, arm-ergometer exercise, and deskwork. Criterion EE using a metabolic cart and raw sensor data from a multisensor activity monitor was collected during each of these activities. Main Outcome MeasuresTwo new EE prediction models including a general model and an activity-specific model were developed using enhanced all-possible regressions on 36 MWUs and tested on the remaining 9 MWUs. ResultsThe activity-specific and general EE prediction models estimated the EE significantly better than the manufacturer's model. The average EE estimation error using the manufacturer's model and the new general and activity-specific models for all activities combined was –55.31% (overestimation), 2.30% (underestimation), and 4.85%, respectively. The average EE estimation error using the manufacturer's model, the new general model, and activity-specific models for various activities varied from –19.10% to –89.85%, –18.13% to 25.13%, and –4.31% to 9.93%, respectively. ConclusionsThe predictors for the new models were based on accelerometer and demographic variables, indicating that movement and subject parameters were necessary in estimating the EE. The results indicate that the multisensor activity monitor with new prediction models can be used to estimate EE in MWUs with SCI during wheelchair-related activities mentioned in this study.

Accurate Prediction of Energy Expenditure Using a Shoe-Based Activity Monitor

The aim of this study was to develop and validate a method for predicting energy expenditure (EE) using a footwear-based system with integrated accelerometer and pressure sensors. We developed a footwear-based device with an embedded accelerometer and insole pressure sensors for the prediction of EE. The data from the device can be used to perform accurate recognition of major postures and activities and to estimate EE using the acceleration, pressure, and posture/activity classification information in a branched algorithm without the need for individual calibration. We measured EE via indirect calorimetry as 16 adults (body mass index=19-39 kg·m) performed various low- to moderate-intensity activities and compared measured versus predicted EE using several models based on the acceleration and pressure signals. Inclusion of pressure data resulted in better accuracy of EE prediction during static postures such as sitting and standing. The activity-based branched model that included predictors from accelerometer and pressure sensors (BACC-PS) achieved the lowest error (e.g., root mean squared error (RMSE)=0.69 METs) compared with the accelerometer-only-based branched model BACC (RMSE=0.77 METs) and nonbranched model (RMSE=0.94-0.99 METs). Comparison of EE prediction models using data from both legs versus models using data from a single leg indicates that only one shoe needs to be equipped with sensors. These results suggest that foot acceleration combined with insole pressure measurement, when used in an activity-specific branched model, can accurately estimate the EE associated with common daily postures and activities. The accuracy and unobtrusiveness of a footwear-based device may make it an effective physical activity monitoring tool.