Preliminary Field Validity of Free-Living Energy Expenditure Estimation in Wheelchair Users with Spinal Cord Injury via Wearable Device-based Models

doi:10.21203/rs.3.rs-3794642/v1

Download PDF

Article

Preliminary Field Validity of Free-Living Energy Expenditure Estimation in Wheelchair Users with Spinal Cord Injury via Wearable Device-based Models

https://doi.org/10.21203/rs.3.rs-3794642/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Study Design

Cross-sectional validation study.

Objectives

To develop a raw acceleration signal-based random forest (RF) model for predicting total energy expenditure (TEE) in manual wheelchair users (MWUs) and evaluate the preliminary field validity of this new model along with four existing models published in prior literature using the Doubly Labeled Water (DLW) method.

Setting

General community and research institution in Pittsburgh, USA.

Methods

A total of 78 participants’ data from two previous studies were used to develop the new RF model. A seven-day cross-sectional study was conducted to collect participants’ free-living physical activity and TEE data, resting metabolic rate, demographics, and anthropometrics. Ten MWUs with spinal cord injury (SCI) completed the study, with seven participants having valid data for evaluating the preliminary field validity of the five models.

Results

The RF model achieved a mean absolute error (MAE) of 0.59 ± 0.60 kcal/min and a mean absolute percentage error (MAPE) of 23.6%±24.3% on the validation set. For preliminary field validation, the five assessed models yielded MAE from 136 kcal/day to 1141 kcal/day, and MAPE from 6.1–50.2%. The model developed by Nightingale et al. in 2015 achieved the best performance (MAE: 136 ± 96 kcal/day, MAPE: 6.1%±4.7%), while the RF model achieved comparable performance (MAE: 167 ± 99 kcal/day, MAPE: 7.4%±5.1%).

Conclusions:

Two existing models and our newly developed RF model showed good preliminary field validity for assessing TEE in MWUs with SCI and the potential to detect lifestyle change in this population. Future large-scale field validation study and model iteration is recommended.

Biological sciences/Physiology/Metabolism

Health sciences/Neurology/Neurological disorders/Spinal cord diseases

Habitual energy expenditure (EE) assessment is important for characterizing energy balance, and promoting healthy physical activity (PA) behaviors [1]. Positive energy balance is one of the factors associated with many chronic diseases, such as obesity and type 2 diabetes [2]. Body fat, weight gain, and weight regain negatively correlate with total daily energy expenditure (TDEE) [3]. Managing TDEE is especially important for manual wheelchair users (MWUs) with spinal cord injury (SCI), as the rates of chronic diseases are higher in this population [4]. Studies have revealed this population has lower TDEE than general population controls, making it difficult to maintain a healthy energy balance [5–7].

While energy intake can be estimated using the nutrition facts panel on food packages, the assessment of free-living EE can be challenging given the difficulty of consistently and accurately tracking all PA contributing to TDEE. Compared with the body of work in the ambulatory population, relevant work in MWUs has been limited. Several studies have reviewed the common methods available for assessing free-living EE and PA in MWUs, including Doubly Labelled Water (DLW), self-report questionnaires, basal metabolic rate (BMR)-based prediction models, and wearable devices [8–10].

The DLW technique requires the subject to drink a dose of water with stable isotopes of hydrogen and oxygen and provide urine or saliva samples during the monitoring period for analysis. Early validation studies of the doubly labeled water method showed good accuracy (Range of -2.5 to 3.3%) and precision (average 5.1%) in a range of subjects and conditions against respiratory gas exchange by four different laboratories [11]. Seale et al. validated it against the direct room calorimetry method during 5–7 days, achieving a mean percentage difference of 1.6% ± 4.5% [12]. However, the DLW technique is relatively expensive, require special equipment for analysis, and can only provide the total EE over a period of time instead of revealing underlying patterns. These limitations restrict its use in large-scale studies. The DLW technique is commonly used as a gold standard to validate other EE assessments [10, 13].

Traditional self-report questionnaires designed for people with SCI provide an inexpensive and easy-to-administer method to estimate EE [8]. Nonetheless, a study has shown these questionnaires have limited accuracy (i.e., 133 ± 305 kcal/day) against the criterion measurement of DLW [10]. These methods are also subject to recall bias, floor effects, participant over-reporting, and have limited accuracy in quantifying low-intensity PA [14, 15].

Another widely used EE prediction method is prediction equations. A study by Desneves et al. evaluated the free-living EE estimated by six published demographics/anthropometrics measurements-based BMR prediction equations multiplied by a combined stress and activity factor of 1.3 against DLW on acute SCI patients [13]. The equations showed varied performance, with one requiring fat-free mass (FFM) having the best accuracy with a mean absolute percent error (MAPE) of 8.6%. However, the limit of Agreement (LoA) between the predicted and criterion TDEE was still quite wide (i.e., -1003 to 1601 kcal/day), indicating unreliable performance for individual patients.

Accelerometer-based wearable devices have become a popular approach to assess EE for their low cost and high acceptance [16, 17]. Despite the existence of multiple custom models developed for predicting EE in MWUs, the only model that was validated against the criterion measurement of DLW on TDEE prediction was the discontinued SenseWear Armband’s proprietary model [8, 10, 18]. The model yielded a wide LoA from − 450.5 to 805.7 kcal/day [18]. In a recent study by our group, we evaluated five sets of custom prediction models for MWUs, which were based on the ActiGraph devices (ActiGraph, LLC., Pensacola, FL, USA), the commonly used tri-axial accelerometer-based research-grade wearable devices [19]. The five models were developed using PA data collected in lab settings with relatively small sample sizes ranging from 15 to 49 and were validated on the same dataset used to develop the model. We evaluated their performance on an out-of-sample dataset collected in lab settings from 29 participants [19]. However, the field validity of these models remained unknown due to the lack of free-living data.

In this paper, we developed a custom model using machine learning techniques with raw accelerometer signal data, such as the Vector Magnitude (VM), from two previous studies. Its preliminary field validity was evaluated along with the aforementioned ActiGraph devices-based models against the criterion TDEE measurement of DLW in MWUs with SCI.

EE Prediction Models

Four out of five published EE prediction models based on ActiGraph devices were assessed, including three models based on proprietary Vector Magnitude Count (VMC) and one based on raw acceleration signals [9, 20–22]. Garcia et al.’s model was not included due to its computational complexity in generating the features [23]. The included models have varied outputs (i.e., physical activity EE, total EE, and oxygen consumption). All outputs were converted to total energy expenditure (TEE) for evaluation against the DLW measurement. Details of the conversion and the original model equations can be found in Table 1.

Table 1

Equations and TEE Conversion of Existing Models
Model (Author)	Equation (units)	TEE Conversion (kcal)
Model 1 (Nightingale, 2014) [20]	PAEE = (0.000929 × VMC) – 0.284818 (kJ∙min^− 1)	TEE (kal∙min^− 1) = (PAEE × 0.239^* + RMR)/0.9^†
Model 2 (Nightingale, 2015) [21]	PAEE = (0.000245 × VMC) + 0.132379 (kal∙min^− 1)	TEE (kal∙min^− 1) = (PAEE + RMR)/0.9^†
Model 3 (Tsang, 2018) [9]	TEE = -0.006197602220975 + 0.000000088463104 × BMR2 × VM + 0.000823693371782 × BMR1 + 0.000577607827818 × X (kal∙min^− 1)	N/A
Model 4 (Learmonth, 2016) [22]	Right wrist: VO2 = 0.0022 × VMC + 3.13 (ml/min^− 1kg^− 1) Left wrist: VO2 = 0.0021 × VMC + 3.14 (ml/min^− 1kg^− 1)	TEE (kal∙min^− 1) = (VO2 /1000^‡) × Weight (kg) × 5^§
Note. In model 3, BMR1 refers to the basal metabolic rate by World Health Organization [24]. BMR2 refers to the basal metabolic rate by Mifflin et al. (1990) [27].
PAEE – Physical Activity Energy Expenditure
RMR - Resting Metabolic Rate
^*Factor used to convert kJ to kcal
^†Factor used to adjust for the diet-induced thermogenesis, which is commonly estimated as 10% of the TEE[18, 25, 26]
^‡Factor used to convert ml to L
^§Calorie equivalent of VO2, 5 kcal/L

In addition, a random forest (RF) model using raw acceleration signals from ActiGraph activity monitors was developed by aggregating data from two published studies that shared similar inclusion/exclusion criteria and study protocols [27]. In both studies, participants were fitted with a K4b2 portable metabolic cart (COSMED USA Inc., Concord CA, USA) for criterion TEE measurement, and an ActiGraph activity monitor on the dominant wrist and performed a series of activities of daily living and exercises. The PA measured by the activity monitor and TEE measured by the metabolic cart were extracted and cleaned. A set of features, including features based on raw acceleration signals and demographic variables, was generated and then selected using Pearson’s r. We used the “grid search” approach to tune the hyperparameters for the RF model. The Leave-One-Out Cross-Validation was used to evaluate the model performance of each hyperparameter combination. The hyperparameter combination that achieved the best mean squared error (MSE) was used to train the final RF model using all participants’ data for the subsequent free-living field validation. More details about the model development can be found in Supplementary Appendix A.

Free-Living Validation via DLW

This study was conducted at the University of Pittsburgh Endocrinology and Metabolism Research Center at Montefiore Hospital, Pittsburgh. It was approved by the Institutional Review Boards of the VA Pittsburgh Healthcare System and the University of Pittsburgh. We included participants who were: 1) between 18 and 65 years, 2) having an SCI, 3) at least one-year post-injury, 4) using a manual wheelchair as a primary means of mobility (≥ 80% of weekly mobility), 5) medically stable, 6) living within 1 hour of the research site, 7) being able to tolerate sitting for three hours, and 8) being able to self-catheterize for the doubly labeled water activities. We excluded participants who were: 1) having any active pelvic or thigh wounds, 2) having a history of cardiovascular disease, 3) having high blood pressure, seizures, lung disease, fainting/dizziness, shortness of breath at rest or with usual activities, or unusual fatigue, and 4) pregnant. Convenience sampling was used for recruitment.

As the purpose of this paper was to validate EE prediction models, only EE-related instruments and data were reported.

Study Protocol

The study consisted of two lab visits with a one-week home trial in between. Participants’ criterion TDEE and upper extremity kinematic data during the study period were measured by the DLW and the ActiGraph activity monitor (GT9X Link), respectively. The DLW dose was approximately 3.5 oz (actual weight of the DLW was noted for each participant). After drinking the dose, the bottle was rinsed with 3 oz tap water, which the subject then drank. The flowchart (Fig. 1) shows the timeline. The anthropometric measurements of height were conducted in a supine position with a tape measure to the closest 1 cm and converted to inches. The participant’s weight was calculated by subtracting the wheelchair weight from their total weight measured by a wheelchair scale (Detecto,Webb City MO, USA) and rounded to the nearest 0.1 lb. Body composition was measured via the dual-energy X-ray absorptiometry scan (DEXA) using the GE Lunar iDXA (GE Healthcare, Wauwatosa WI, USA). Their resting metabolic rate (RMR) was measured in the supine position on a bed for 30 minutes in a dimly lit, quiet room using the TrueOne stationary metabolic cart (Parvo Medics, Salt Lake City UT, USA) following 12 hours fast. They were instructed not to talk and not to fall asleep during the measurement. More details about the study instruments can be found in Supplementary Appendix B.

All three urine samples from each participant were refrigerated and kept at the lab until the whole study concluded. They were then sent to the AdventHealth Orlando, Translational Research Institute (Orlando, FL.) for processing and analysis.

[Figure 1]

Data Collection and Preparation

The raw acceleration outputs and the minute-by-minute VMC were downloaded and extracted using ActiGraph’s proprietary software ActiLife (v6.11.9) from the GT9X Link. After data cleaning, the ActiGraph data was segmented into valid wear, non-wear, and sleep periods based on the self-report device wear and activity log, and the wear time validation function in the ActiLife software. The developed RF model, as well as the four existing models were then applied to the valid wear period, while lab-measured RMR was used to compensate for non-wear and sleep periods for estimating the TDEE. Data collection and preparation procedure flowchart is presented in Fig. 2, while more details can be found in Supplementary Appendix C.

[Figure 2]

Data Analysis

RF model performance analysis

The performance of the final model was presented by the mean and SD of the mean absolute error (MAE) and MAPE across all participants. The result was segregated into three activity intensity levels based on the metabolic equivalent task (MET): 1) Sedentary (MET < 1.5), 2) Light (1.5 ≤ MET ≤ 3), 3) Moderate to Vigorous Physical Activity (MVPA) (MET > 3). The importance of each feature was estimated by averaging their importance across all decision trees in the model [28]. The Bland-Altman (BA) plot was used to further evaluate agreements and bias between estimated and criterion EE.

Field validation

Any participant’s data with non-wear time larger than 20% of the trial length was excluded. Due to the small sample size and preliminary nature of this study, descriptive analysis was provided to evaluate the field validity of the models. The final estimated TDEE was validated against the criterion TDEE measured by the DLW. The overall MAE and MAPE were assessed. All data analyses in this paper were performed by SPSS version 26 (IBM, Armonk NY, USA) and Microsoft Excel (Micosoft, Redmond WA, USA).

RF Model Development

A total of 7,640 minutes of data from 78 participants was used to develop the RF model [27]. Table 2 depicts the demographics of the 78 participants.

Table 2

Demographics (Model development)
Characteristics	Mean	SD
Age (years)	38.7	11.7
Height (cm)	170.9	14.7
Weight (kg)	76.0	19.1
Years Using Wheelchair	12.6	9.8
	Mean	%
Sex
Male	60	76.9
Female	18	23.1
Diagnosis
SCI	60	76.9
Spinal Bifida	9	11.5
Cerebral Palsy	2	2.6
Multiple Sclerosis	1	1.3
Other	6	7.7
Dominant hand
Right	66	84.6
Left	12	15.4

Overall, the RF model achieved an MAE of 0.59 ± 0.60 kcal/min (95% CI [0.58, 0.60]) and a MAPE of 23.6 ± 24.3% (95% CI [23.0, 24.1]). When segregated to different intensities, the RF model achieved MAE and MAPE as follows: 0.31 ± 0.39 kcal/min (95% CI [0.30, 0.33]) and 27.7 ± 34.5% (95% CI [26.1, 29.3]) for sedentary activity minutes, 0.52 ± 0.44 kcal/min (95% CI [0.51, 0.54]) and 25.5 ± 23.1% (95% CI [24.7, 26.4]) for light activity minutes, and 0.82 ± 0.75 kcal/min (95% CI [0.80, 0.85]) and 19.1 ± 15.7% (95% CI [18.5, 19.7]), respectively. The top three influential features are the VM mean absolute deviation (18.4% importance), the VM mean (11.8% importance), and the weight (9.2% importance).

Figure 3 shows the BA plot, with X-axis being the average of the predicted EE and criterion EE, and Y-axis being the difference between them. Additionally, a significant negative linear relationship between the EE difference and the mean EE was found, which had an estimated slope of -0.25 (p < .001) and an intercept of 0.68 (p < .001).

[Figure 3]

Free-Living Validation via DLW

A total of 10 MWUs with SCI consented and went through the DLW protocol with no dropouts. Table 3 depicts participant demographics.

Table 3

Participant Demographics and Data Availability
ID	Age (years)	Height (cm)	Weight (kg)	Sex	Injury	FFM (kg)	RMR (kcal/min)	Trial length (min)	Sleep time (min)	Non-wear time (min)
1	56	172.7	99.8	Male	T7(incomplete)	57.2	1.29	10,146	4,048	235
2	50	175.3	83.9	Male	T10(complete)	47.4	1.02	10,118	4,996	35
3	44	165.1	69.7	Male	T10(incomplete)	50.1	1.16	10,106	2,853	645
4	25	182.9	93.9	Male	T7(incomplete)	56.7	1.10	10,101	3,095	1,629
5	29	179.1	87.8	Male	T10(complete)	42.2	1.04	-	-	-
6	45	176.0	91.3	Male	C6(incomplete)	58.0	1.31	10,148	-	8,423
7	26	191.8	88.5	Male	T8(incomplete)	58.1	1.25	10,234	3,656	1,796
8	40	182.9	98.0	Male	T3(complete)	53.7	1.21	10,122	5,723	0
9	56	162.6	76.0	Female	C2(incomplete)	44.8	1.12	16,154	1,527	5939
10	37	182.9	101.6	Male	C7(incomplete)	61.3	1.58	11,925	4,029	315
Mean SD	40.8 11.5	177.1 8.8	89.0 10.3	- -	- -	52.9 6.5	1.21 0.16	- -	- -	- -

All 10 participants provided the three urine samples for DLW analysis and thus had valid criterion TDEE. Participant #5 was excluded due to the loss of ActiGraph data files, while participants #6 and #9 were excluded due to non-wear time larger than 20% of the total trial length (i.e., 83% and 37%, respectively). The data availability summary is also presented in Table 3.

The results of the estimated and criterion TDEE (Mean = 2,333 kcal, SD = 289 kcal) are shown in Table 4.

Table 4

Field Validation Result
ID	Criterion TDEE (kcal/day)	Difference (%) (kcal/day)
ID	Criterion TDEE (kcal/day)	RF Model	Model 1	Model 2	Model 3	Model 4
1	2430	-153 (-6.3%)	-312 (-12.8%)	-102 (-4.2%)	-1050 (-43.2%)	805 (33.1%)
2	2066	206 (10.0%)	-83 (-4.0%)	134 (6.5%)	-967 (-46.8%)	1674 (81.0%)
3	2495	-227 (-9.1%)	-312 (-12.5%)	-59 (-2.4%)	-1315 (-52.7%)	627 (25.1%)
4	2184	-56 (-2.6%)	-306 (-14.0%)	-121 (-5.5%)	-890 (-40.8%)	683 (31.3%)
7	2404	33 (1.4%)	-167 (-6.9%)	17 (0.7%)	-964 (-40.1%)	1011 (42.1%)
8	1951	321 (16.4%)	130 (6.6%)	298 (15.3%)	-583 (-29.9%)	1587 (81.4%)
10	2802	-171 (-6.1%)	-22 (-0.8%)	224 (8.0%)	-1404 (-50.1%)	1600 (57.1%)
MAE (MAPE)	-	167 (7.4%)	190 (8.2%)	136 (6.1%)	1025 (43.4%)	1141 (50.2%)
SD MAE (SD MAPE)	-	99 (5.1%)	120 (5.0%)	96 (4.7%)	274 (7.6%)	465 (23.5%)

In this paper, we evaluated the field validity of five wearable devices-based EE prediction models. Four of these models have been previously published in existing literature. We developed a new RF model using a relatively large dataset (N = 78) with a significant amount of activity minutes (7,640 minutes) across 28 different types of structured and unstructured activities. Our RF model achieved an overall MAE of 0.59 kcal/min and a MAPE of 23.6% on the training dataset, while the existing models achieved an overall MAE from 0.87–6.41 kcal and MAPE from 31–37% on a subset of the training dataset [19].

The BA plot (Fig. 3) revealed that the RF model had a trend of overestimation for EE at the lower end of the spectrum and underestimation for EE at the higher end of the spectrum. This trend was also confirmed by the activity intensity-specific model performance and the significant negative linear regression between the mean EE and the EE difference, indicating the existence of a systematic error. We particularly examined the data points that fell outside of the LoA in the BA plot and noticed that in some of these minutes, participants stopped or reduced their upper limb movements. Because of the carry-over effect, the EE remained steady while the ActiGraph got low or no signals, leading to the large discrepancy observed between the estimated and criterion EE. Nightingale et al. explored the use of heart rate signals to improve the EE prediction for wheelchair users, where the individually calibrated heart rate alone explained 55% of the variance in the measured EE [29]. Considering the RF model had large MAE and MAPE in sedentary activities, introducing additional features, such as the individually calibrated heart rate, could enhance the RF model’s accuracy. In terms of feature importance, the VM mean absolute deviation and the mean VM exhibited the highest contribution to the predicted EE. This indicated that the dispersion and magnitude of the VM hold the most valuable information regarding upper extremity movement in MWUs, which in turn aids in predicting EE.

In terms of the TDEE estimation field validation, models #1–2 and the RF model achieved comparable performance, with their MAE ranging from 136–190 kcal, and the MAPE ranging from 6.1–8.2%. Models #3 and #4, with elevated error of MAPE ranging from 43.4–50.2%, are not viable for practical use. One of the factors that may contribute to the large error in Model #3 was the utilization of X-axis acceleration signals, which could affect TEE estimation in sedentary activities with different hand gestures. In terms of model #4, the lack of a population-specific factor that converts VO2 to TEE could contribute to the high MAPE. A previous study by Tanhoffer et al. found that the transition from a sedentary lifestyle to an exercise routine increased the mean TEE by approximately 367 kcal/day in eight participants with SCI, as measured by DLW [18]. As the MAE ± SD of the RF Model and Model # 1–2 are all lower than this value, these three models demonstrate the potential to detect lifestyle change in future clinical research and daily use.

Models #1 and #2 were originally developed by Nightingale et al. using ActiGraph’s proprietary VMC to predict physical activity energy expenditure (PAEE) in people with SCI. Supplemented with RMR measured by indirect calorimetry, these two models achieved the best field validity among all models. This result indicates these two models have high accuracy in predicting PAEE and can provide valid TEE estimation with objectively measured RMR. However, the use of proprietary VMC makes them not directly applicable to other commercially available wearable devices. Although ActiGraph recently published the proprietary VMC conversion process, it requires complex signal processing, and its feasibility and validity on other wearable devices haven’t been investigated [30]. Hence, it is recommended to use these two models in studies where ActiGraph sensors are utilized, and indirect calorimetry-measured RMR is available for their simplicity and high accuracy.

On the other hand, the RF model developed in this study using simple statistical features derived from raw acceleration signals also achieved comparable field validity as models #1 and #2. As the RF model is developed to predict TEE without RMR measurement and doesn’t involve any proprietary variable, this model can be used in studies where ActiGraph sensors are not used, or criterion RMR is not readily available. Integration of this RF model into commercial wearable devices for EE self-monitoring in MWUs with SCI merits further investigation. It is worth noting that accurate RMR prediction is still essential to compensate for non-wear and sleep periods where the RF model is not applicable. Popolation-specific RMR prediction equations and adjustment factors could be viable approaches to be used in conjunction with our RF model to better predict TDEE in MWUs with SCI.

Study Limitations

Future research could utilize the multi-sensor approach to incorporate the heart rate signals into the EE prediction model and explore the use of hidden Markov model for addressing the EE carry-over effect. Our model development dataset included participants with a wide range of diagnoses, with 76.9% being people with SCI and several cases of other diagnoses. However, it was not clear how diagnosis may affect the model performance as the model development dataset and the DLW testing dataset lacked sufficient representation of other diagnoses.

The field validation in this study had limitations due to a relatively small sample size (N = 7), which restricted the applicability of advanced statistical analyses such as the equivalence test and BA plot. Additionally, the absence of female participants in the final dataset hinders the evaluation of model performance for female users. Participant #8, whose RMR accounted for 89.6% of TDEE, was identified as an outlier based on the interquartile rule. Although this proportion exceeded the average RMR proportion of 80% TDEE in people with SCI, we didn’t find any evidence of device malfunction or random coding errors [6]. Hence, data from participant #8 was included in the analysis but should be interpreted with caution.

In this paper, we evaluated the preliminary field validity of five wearable devices-based EE prediction models. The three best models showed the potential to support clinical and daily use and detect lifestyle change, with their MAE ranging from 136–190 kcal and MAPE ranging from 6.1–8.2%. The two VMC-based models developed by Nightingale et al. to predict PAEE may be used with objectively measured RMR to estimate TEE in future studies that use ActiGraph devices for their simplicity, while the raw acceleration signals-based RF model developed in this study may be applied with RMR prediction equations to other studies that use commercially available wearable devices or lack of objectively measured RMR. The integration of the developed RF model to other commercial wearable devices for EE self-monitoring among MWUs with SCI is worth exploring. Future validation studies with larger sample sizes and more diverse demographics, such as sex and level of injury, should be considered.

DATA AVAILABILITY

The datasets supporting this study and the developed total energy expenditure prediction model are available upon reasonable request from the corresponding author.

ACKNOWLEDGEMENT

We would like to acknowledge and thank Steven J Anthony for his assistance in conducting the field validation study protocol and data collection.

AUTHOR CONTRIBUTIONS

ZH was responsible for conducting the study protocol, screening potential participants, performing data collection and analysis, obtaining and interpreting results, updating reference lists, and developing this paper in its entirety. ALV was responsible for designing the protocol, screening potential participants, performing data collection and analysis, obtaining and interpreting results, and editing the paper. JPD was responsible for performing data analysis, obtaining the results, and editing the paper. DD provided guidance throughout the study, including formulating the paper plan, supervising data collection, data analysis, and result interpretation, as well as editing the paper.

FUNDING

This study was funded by the Veterans Affairs (VA) Rehabilitation Research & Development under Grant # 1I01RX000971-01A2 conducted at the Human Engineering Research Laboratories (Pittsburgh, PA) and the James J. Peters VA Medical Center (Bronx, NY). This study used data from another study conducted by our team and funded by the National Institute of Disability, Independent Living and Rehabilitation Research, Rehabilitation Engineering Research Center on Recreational Technologies Benefiting Individuals with Disabilities (#H133E120005), conducted at the Human Engineering Research Laboratories (Pittsburgh, PA) and the Lakeshore Foundation (Birmingham, AL). The content is solely the responsibility of the authors and does not represent the official views of the Department of Veterans Affairs or the United States Government.

ETHICAL APPROVAL

This study was approved by the Institutional Review Boards of the VA Pittsburgh Healthcare System and the University of Pittsburgh. We certify that all applicable institutional and governmental regulations concerning the ethical use of human volunteers were followed during the course of this research.

COMPETING INTERESTS

The authors declare no competing interests.

Keim, N.L., C.A. Blanton, and M.J. Kretsch, America’s obesity epidemic: measuring physical activity to promote an active lifestyle. Journal of the American Dietetic Association, 2004. 104(9): p. 1398–1409.
Buchholz, A.C., C.F. McGillivray, and P.B. Pencharz, Physical activity levels are low in free-living adults with chronic paraplegia. Obes Res, 2003. 11(4): p. 563–70.
Ball, E.J., et al., Total energy expenditure, body fatness, and physical activity in children aged 6–9 y. Am J Clin Nutr, 2001. 74(4): p. 524–8.
Froehlich-Grobe, K. and D. Lollar, Obesity and disability: time to act. Am J Prev Med, 2011. 41(5): p. 541–5.
Hall, K.D., et al., Energy balance and its components: implications for body weight regulation. Am J Clin Nutr, 2012. 95(4): p. 989–94.
Monroe, M.B., et al., Lower daily energy expenditure as measured by a respiratory chamber in subjects with spinal cord injury compared with control subjects. The American journal of clinical nutrition, 1998. 68(6): p. 1223–1227.
Polfuss, M., et al., Total energy expenditure and body composition of children with developmental disabilities. Disabil Health J, 2018. 11(3): p. 442–446.
Nightingale, T.E., et al., Measurement of Physical Activity and Energy Expenditure in Wheelchair Users: Methods, Considerations and Future Directions. Sports Med Open, 2017. 3(1): p. 10.
Tsang, K., Using Wearable Sensors for Physical Activity Measurement and Promotion in Manual Wheelchair Users. 2018, University of Pittsburgh, 2018. p. 216 pages.
Tanhoffer, R.A., et al., Comparison of methods to assess energy expenditure and physical activity in people with spinal cord injury. J Spinal Cord Med, 2012. 35(1): p. 35–45.
Schoeller, D.A., Measurement of energy expenditure in free-living humans by using doubly labeled water. (0022–3166 (Print)).
Seale, J.L., J.M. Conway, and J.J. Canary, Seven-day validation of doubly labeled water method using indirect room calorimetry. Journal of applied physiology, 1993. 74(1): p. 402–409.
Desneves, K.J., et al., Comparison of estimated energy requirements using predictive equations with total energy expenditure measured by the doubly labelled water method in acute spinal cord injury. Spinal cord, 2019. 57(7): p. 562–570.
Tudor-Locke, C.E. and A.M. Myers, Challenges and opportunities for measuring physical activity in sedentary adults. Sports Med, 2001. 31(2): p. 91–100.
Sallis, J.F. and B.E. Saelens, Assessment of physical activity by self-report: status, limitations, and future directions. Res Q Exerc Sport, 2000. 71 Suppl 2: p. 1–14.
Ridgers, N.D., M.A. McNarry, and K.A. Mackintosh, Feasibility and effectiveness of using wearable activity trackers in youth: a systematic review. JMIR mHealth and uHealth, 2016. 4(4): p. e6540.
Hawthorne, G., et al., Usability of Wearable Multiparameter Technology to Continuously Monitor Free-Living Vital Signs in People Living With Chronic Obstructive Pulmonary Disease: Prospective Observational Study. JMIR Hum Factors, 2022. 9(1): p. e30091.
Tanhoffer, R.A., et al., Energy expenditure in individuals with spinal cord injury quantified by doubly labeled water and a multi-sensor armband. Journal of Physical Activity and Health, 2015. 12(2): p. 163–170.
Shwetar, Y.J., et al., Comparative validity of energy expenditure prediction algorithms using wearable devices for people with spinal cord injury. Spinal Cord, 2020: p. 1–10.
Nightingale, T.E., et al., Predicting physical activity energy expenditure in manual wheelchair users. Med Sci Sports Exerc, 2014. 46(9): p. 1849–1858.
Nightingale, T.E., et al., Influence of accelerometer type and placement on physical activity energy expenditure prediction in manual wheelchair users. PLoS One, 2015. 10(5): p. e0126086.
Learmonth, Y.C., et al., Accelerometer output and its association with energy expenditure during manual wheelchair propulsion. Spinal cord, 2016. 54(2): p. 110–114.
García-Massó, X., et al., Validation of the use of Actigraph GT3X accelerometers to estimate energy expenditure in full time manual wheelchair users with spinal cord injury. Spinal cord, 2013. 51(12): p. 898–903.
Alfonzo-González, G., et al., Estimation of daily energy needs with the FAO/WHO/UNU 1985 procedures in adults: comparison to whole-body indirect calorimetry measurements. European journal of clinical nutrition, 2004. 58(8): p. 1125–1131.
Westerterp, K.R. and G. Plasqui, Physical activity and human energy expenditure. Current Opinion in Clinical Nutrition & Metabolic Care, 2004. 7(6): p. 607–613.
Nightingale, T.E., et al., Energy balance components in persons with paraplegia: daily variation and appropriate measurement duration. Int J Behav Nutr Phys Act, 2017. 14(1): p. 132.
Shwetar, Y., et al., Predicting physical activity intensity using raw accelerometer signals in manual wheelchair users with spinal cord injury. (1476–5624 (Electronic)).
Breiman, L., Random forests. Machine learning, 2001. 45(1): p. 5–32.
Nightingale, T., et al., Predicting physical activity energy expenditure in wheelchair users with a multisensor device. BMJ open sport & exercise medicine, 2015. 1(1): p. bmjsem-2015-000008.
Neishabouri, A., et al., Quantification of acceleration as activity counts in ActiGraph wearable. (2045–2322 (Electronic)).

There is no duality of interest

SupplementaryAppendices.pdf

Download PDF

Editorial decision: revise
24 Mar, 2024
Review #1 received at journal
19 Mar, 2024
Review #2 received at journal
18 Mar, 2024
Reviewer #2 agreed at journal
27 Feb, 2024
Reviewer #1 agreed at journal
23 Feb, 2024
Reviewers invited by journal
23 Feb, 2024
Submission checks completed at journal
25 Jan, 2024
First submitted to journal
22 Dec, 2023
Editor assigned by journal
22 Dec, 2023

You are reading this latest preprint version

Preliminary Field Validity of Free-Living Energy Expenditure Estimation in Wheelchair Users with Spinal Cord Injury via Wearable Device-based Models

Status:

Version 1

Abstract

Study Design

Objectives

Setting

Methods

Results

Conclusions:

Figures

INTRODUCTION

METHODS

Data Analysis

RF model performance analysis

Field validation

RESULTS

DISCUSSION

CONCLUSIONS

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1