Edited by: Mattias K. Sköld, Uppsala University, Sweden
Reviewed by: Rosa Margarita Gomez, Independent Researcher, Colombia; Hari S. Sharma, Uppsala University, Sweden; Mariella Pazzaglia, Sapienza Università di Roma, Italy
This article was submitted to Neurotrauma, a section of the journal Frontiers in Neurology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
A healthy lifestyle reduces the risk of cardio-vascular disease. As wheelchair-bound individuals with spinal cord injury (SCI) are challenged in their activities, promoting and coaching an active lifestyle is especially relevant. Although there are many commercial activity trackers available for the able-bodied population, including those providing feedback about energy expenditure (EE), activity trackers for the SCI population are largely lacking, or are limited to a small set of activities performed in controlled settings. The aims of the present study were to develop and validate an algorithm based on inertial measurement unit (IMU) data to continuously monitor EE in wheelchair-bound individuals with a SCI, and to establish reference activity values for a healthy lifestyle in this population. For this purpose, EE was measured in 30 subjects each wearing four IMUs during 12 different physical activities, randomly selected from a list of 24 activities of daily living. The proposed algorithm consists of three parts: resting EE estimation based on multi-linear regression, an activity classification using a k-nearest-neighbors algorithm, and EE estimation based on artificial neural networks (ANNs). The mean absolute estimation error for the ANN-based algorithm was 14.4% compared to indirect calorimeter measurements. Based on reference values from the literature and the data collected within this study, we recommend wheeling 3 km per day for a healthy lifestyle in wheelchair-bound SCI individuals. Combining the proposed algorithm with a recommendation for physical activity provides a powerful tool for the promotion of an active lifestyle in the SCI population, thereby reducing the risk for secondary diseases.
The life expectancy of individuals with a spinal cord injury (SCI) has increased significantly over the last decades (
Vanhees et al. showed that accelerometers and inertial measurement units (IMUs) can be used as objective assessment tools for physical activity and EE in the able-bodied population (
Although accelerometer-based EE estimation models were developed for subjects with SCI using a manual wheelchair (
The main aim of this study therefore was to develop and validate an EE estimation model for wheelchair-bound SCI individuals based on non-obstructive IMU recordings in a natural setting. We included a comprehensive set of 24 different physical activities, covering a broad range of activities of daily living. Furthermore, the collected data formed a basis for a recommendation that could promote a healthy lifestyle in the wheelchair-bound SCI population.
Thirty chronic SCI subjects (age 45.4 ± 11.4 years, 11 tetraplegics, 19 paraplegics) who rely on a wheelchair for daily ambulation were recruited. Inclusion criteria were an age over 18 years old and suffering from SCI for more than 6 months post injury. Subjects with all neurological levels of injury (NLI) according to the International Standard for Neurological Classification of Spinal Cord Injury (ISNCSCI), and ASIA Impairment Scale (AIS) grades (A, B, C, and D) were included (Table
Demographics and assessment scores of the included participants.
No of participants | 30 |
Age (years) | 45.4 ± 11.4 (27–74) |
Weight (kg) | 74.3 ± 17.1 (45.6–116.8) |
Height (m) | 1.76 ± 0.09 (1.54–2.03) |
Reported hours of sport/week | 2.5 ± 2.9 (0.0–10.0) |
Male | 27 |
Female | 3 |
C3–C8 | 11 |
T1–L1 | 19 |
A | 17 (14/3) |
B | 7 (4/3) |
C | 3 (1/2) |
D | 3 (0/3) |
An IMU (ReSense) developed by Leuenberger and Gassert was used for this study (Figures
EE was assessed using a portable metabolic cart (Oxycon mobile, Carefusion, Hoechberg, Germany; Figures
BIA was used to determine fat mass (FM) and fat-free mass (FFM) of each subject. Based on FM and FFM, an additional reference value for the resting energy expenditure (REE) was calculated. For BIA measurements, a signal electrode as well as a measurement electrode were attached at each hand and foot and connected to the BIA device (AKERN BIA 101 system, SMT medical, Würzburg, Germany). FM and FMM were calculated using the proprietary software (BodyComposition, MEDI CAL HealthCare GmbH, Karlsruhe, Germany).
Prior to the experiment, three standard clinical assessments were conducted to gather information on the NLI, the severity of the lesion, and the independence of the SCI subjects. The ISNCSCI protocol was used to assess the NLI as well as the completeness of the lesion (
Each subject had to perform 12 different physical activities out of a set of 24 possible activities. These activities were divided into three activity classes based on measured EE, subjectively perceived exertion, and the amount of distance traveled. The “low-intensity” activity class included the following activities: rest (lying on a bed), watching TV, reading, doing crossword puzzles, playing cards, riding an elevator, playing with a tablet PC, writing, computer work, and passive wheeling (i.e., when the wheelchair was pushed by someone else). The “high intensity” class included the following activities: washing dishes, hanging out the laundry, using a handbike ergometer (30 W), playing table tennis, and weight lifting. The last class was called “wheeling” and included activities involving wheelchair self-propulsion. These activities included completing a wheelchair skill parcour (including a slalom with nine cones, four curbs of 3–8 cm height, and a ramp with an inclination of 8%), wheeling at different speeds (2, 3.5, 5, 6.5 km/h and self-chosen), wheeling uphill (inclination 2.6%), wheeling downhill (inclination 2.6%), and wheeling on a wheelchair ergometer.
Participants came to the Balgrist University Hospital for a single session of ~5 h in the morning after an overnight fast of at least 10 h. First, participants were informed about the experimental procedure, and the 12 pseudo-randomly selected tasks were explained in detail. Subsequently, body composition was assessed by use of BIA, height was measured while the subject was lying on the bed, and weight was measured with a wheelchair scale. Participants were equipped with one sensor module at each wrist, one module was fixed at the chest (approximately at the sternum) and an additional module was fixed to one wheel of the wheelchair. Participants used their own, individually adapted wheelchair for the entire duration of the study. Thereafter, the sensor modules were time-synchronized with the camera, the indirect calorimeter, and the HR monitor.
The first part of the experiment consisted of 20 min of rest, lying on a bed for assessment of REE, followed by a standardized breakfast equivalent to 30% of a participant's calculated daily EE. The second part of the experiment started at least 90 min after the end of breakfast, when EE had returned to baseline values. First, participants lay on a bed for 20 min. This was considered the REE measurement under the non-fasted condition (first task). Subsequently, participants performed eleven 8-min tasks, selected pseudo-randomly from the set of 24 tasks, with a minimum of 5 min between two consecutive tasks. The pseudo-random selection ensured that at least two tasks from each activity class were selected and that each task was performed approximately equally often across all subjects. The 8-min activities were ordered according to the expected intensity of the tasks, starting with the least intense task. After each task, subjects were asked to rate their perceived exertion on an 11-point numeric rating scale (0 = “no exertion,” 10 = “maximum exertion”).
Video recordings were taken during the entire experiment (GoPro Hero HD 2, Go Pro Inc., San Mateo, CA, USA) in order to verify all activities retrospectively. In case subjects were unable to start the experiment in the morning, a shortened version of the protocol was provided, i.e., subjects came to the clinic at least 2 h after the last food intake. After explanation of the study and signing the consent form, the experiment started with the assessment of body composition, height, and weight, followed by 20 min of REE measurement in the non-fasted condition. Afterwards, subjects followed the same protocol as described above.
The complete data processing, statistical analysis, as well as the training of the k-nearest neighbors (kNN) classifier and ANNs were performed using MATLAB 2014a (The MathWorks, Natick, MA, USA). All processing steps were conducted offline.
In total, four different algorithms were designed, evaluated (Figure
Flow chart summarizing the different evaluated algorithms. The first algorithm for estimating energy expenditure (EE) consists of a multiple linear regression (MLR) model and the second algorithm of an artificial neural network (ANN); both use features derived from IMU data, participant demographics and clinical assessment scores, as well as previously estimated resting EE (REE) as model input
In order to ensure that all IMU data consisted of the same number of samples and that they were temporally aligned, the recordings from the ReSense modules were resampled at 50 Hz using a cubic spline interpolation function. Afterwards, IMU recordings were synchronized with the OxyconMobile data and the video recordings, using time stamps, which were aligned at the beginning of the experiment. The acceleration signals were filtered using a 2nd order Butterworth high-pass filer with a cut-off frequency of 0.25 Hz in order to remove the static acceleration component due to gravity. Gyroscope data were filtered with the same high-pass filter. The altitude data was filtered using a 2nd order Butterworth low-pass filter with a cut-off frequency 0.2 Hz.
Data were labeled using temporal markers from the OxyconMobile, the IMU modules, and from the video recordings. For the REE measurements, the mean of a 4-min window (min 14–18) was taken. For each of the individual activities, the last 4 min of each activity were segmented in windows of 1 min without overlap. Taking the last 4 min ensured that the EE had reached a steady state. After visual inspection of the data, 1,324 windows remained, which were later included in the development of the different models. As HR data was partly missing for some subjects, all analysis involving the HR was only based on 897 windows (67.8%).
Features were calculated from the processed acceleration signal containing the dynamic component, from the gyroscope data and the altitude signal for each window. All statistical features derived from the accelerometer and gyroscope data were calculated from the respective magnitudes in order to ensure that the orientation of the sensors, and therefore in which orientation the sensor is placed, had no influence on the final algorithm. Statistical features derived from the sensor data were based on previously used features in activity classification studies (
Five different MLR models were developed for the estimation of the REE. The different MLR models have all the following basic form:
Results of the REE estimation based on BIA measurement, models known from the literature (°), and MLR and ANN models developed in this study.
BIA |
23.3 ± 12.2 | 409.7 ± 255.7 | 23.5 ± 12.2 | 23.2 ± 12.5 | −22.5 ± 13.7 | −400.7 ± 270.2 | 48.3 | 956.1 |
Harris-Benedic° | 14.2 ± 7.9 | 229.8 ± 129.1 | 13.9 ± 7.9 | 14.5 ± 8.2 | 1.9 ± 16.4 | 0.5 ± 267.1 | 26.7 | 528.2 |
updated Harris-Benedict° | 14.2 ± 7.8 | 227.8 ± 123.1 | 13.5 ± 7.4 | 14.6 ± 8.2 | 2.3 ± 16.3 | 7.0 ± 262.4 | 25.9 | 511.4 |
Mifflin-St. Jeor° | 15.1 ± 11.8 | 256.1 ± 204.9 | 16.7 ± 12.8 | 14.1 ± 11.4 | −8.1 ± 17.5 | −169.5 ± 283.1 | 34.6 | 648.8 |
h, w, a, g | 15.7 ± 12.0 | 250.3 ± 182.6 | 15.0 ± 8.6 | 16.1 ± 13.8 | 1.3 ± 19.9 | 63.3 ± 306.7 | 48.7 | 810.5 |
h, w, a, g, SCIM III | 16.6 ± 15.5 | 262.4 ± 227.6 | 13.2 ± 8.0 | 18.7 ± 18.5 | −0.4 ± 22.9 | 37.7 ± 348.8 | 67.8 | 933.3 |
h, w, a, g, AIS | 16.3 ± 11.3 | 261.5 ± 175.1 | 16.2 ± 7.1 | 16.3 ± 13.5 | 1.2 ± 20.0 | 63.2 ± 312.0 | 48.5 | 807.3 |
h, w, a, g, NLI | 16.4 ± 11.5 | 264.2 ± 175.7 | 15.1 ± 8.4 | 17.2 ± 13.2 | 0.9 ± 20.3 | 57.9 ± 315.7 | 48.9 | 814.4 |
h, w, a, g, motor score | 16.6 ± 12.7 | 264.1 ± 190.8 | 16.0 ± 9.3 | 16.9 ± 14.6 | 1.2 ± 21.0 | 64.9 ± 322.9 | 49.2 | 818.2 |
h, w, a, g | 16.2 ± 12.2 | 248.9 ± 156.5 | 14.3 ± 8.3 | 17.4 ± 14.1 | 2.7 ± 20.4 | −6.6 ± 297.7 | 32.2 | 535.7 |
h, w, a, g, SCIM III | 14.2 ± 12.2 | 215.7 ± 160.0 | 10.0 ± 5.8 | 16.7 ± 14.4 | 2.8 ± 18.7 | 4.2 ± 271.5 | 39.7 | 660.4 |
h, w, a, g, AIS | 17.0 ± 11.0 | 263.0 ± 137.6 | 15.4 ± 7.4 | 18.0 ± 12.9 | 2.4 ± 20.4 | −12.6 ± 300.6 | 49.7 | 528.5 |
h, w, a, g, NLI | 16.6 ± 11.0 | 256.5 ± 136.5 | 14.6 ± 6.9 | 17.9 ± 12.9 | 2.7 ± 20.0 | −5.8 ± 294.5 | 33.3 | 554.3 |
h, w, a, g, motor score | 15.8 ± 12.3 | 240.1 ± 155.0 | 12.2 ± 8.4 | 17.9 ± 14.0 | 2.9 ± 20.0 | −1.2 ± 289.4 | 54.4 | 579.0 |
h, w, a, g, AIS, NLI | 17.4 ± 11.7 | 266.8 ± 140.5 | 14.4 ± 8.2 | 19.2 ± 13.2 | 2.9 ± 21.0 | −5.7 ± 305.6 | 52.3 | 557.0 |
In order to classify the different windows into one of the previously described activity categories “low-intensity,” “high-intensity,” and “wheeling,” a kNN classifier with
In total four different estimation models were designed for the activity dependent EE not including the HR. An MLR model and an ANN model were designed where the activity-dependent EE was estimated using IMU data and the estimated REE as predictors. In order to see if a prior classification into different activity classes increases the estimation accuracy, additional MLR and ANN based models were designed. Thereby each activity class had a separate MLR or ANN estimation model. In order to see how the classification accuracy of the previously mentioned kNN classifier influences the final EE estimation, the MLR and ANN models with prior activity classification were evaluated (i) assuming 100% correct classification and (ii) with the classes estimated by the kNN classifier. Similar to the REE estimation using MLR, the MLRs for the activity-depended EE estimation were computed using the sum of squared relative errors (
For a healthy lifestyle, different reference values exist for the able-bodied population. The most well-known reference value for a healthy lifestyle is probably the 10,000 steps a day reference (equivalent 300 kcal/day) (
The U.S. Department of Health and Human Services for example, suggests 150 min of moderate physical activity per week (e.g., 5 × 30 min) or 75 min of vigorous physical activity per week (
The performance of the REE and EE estimation models were analyzed in terms of MAE in percent and mean signed error (MSE) in percent. The performance of the kNN classifier was analyzed using overall classification accuracy in percent and in addition the sensitivity of the different classes was computed (
An overview of all REE estimation models is presented in Table
The overall classification accuracy of the kNN classifier was 97.9%. An overview of the classification accuracy of each individual activity class can be found in Figure
3D scatter plot of the different activity classes for the three features used
ANN and MLR models were based on a total of 1,324 windows. An overview of the EE estimation accuracy of all models can be found in Table
Evaluation of the different models developed as a part of this study.
MLR general | 15.3 ± 4.8 | 592.7 ± 270.9 | 16.4 ± 4.7 | 14.6 ± 4.9 | −2.9 ± 11.1 | −266.2 ± 444.4 | 27.3 | 1180.1 |
MLR class known | 15.0 ± 4.7 | 579.2 ± 255.1 | 15.9 ± 4.2 | 14.5 ± 5.1 | −2.5 ± 10.5 | −233.0 ± 416.7 | 25.7 | 1103.8 |
MLR class estimated | 15.2 ± 4.7 | 584.0 ± 253.6 | 16.3 ± 4.0 | 14.5 ± 5.0 | −2.5 ± 10.5 | −235.0 ± 415.4 | 25.8 | 1108.9 |
ANN general | 17.3 ± 6.8 | 606.5 ± 225.1 | 18.2 ± 6.6 | 16.8 ± 7.1 | 5.4 ± 13.4 | 20.3 ± 489.9 | 26.8 | 1118.0 |
ANN class known | 14.1 ± 5.4 | 513.6 ± 201.3 | 14.9 ± 4.7 | 13.6 ± 5.8 | 3.3 ± 9.8 | 2.9 ± 379.8 | 23.9 | 980.0 |
ANN class estimated | 14.4 ± 5.3 | 524.7 ± 205.2 | 15.5 ± 4.7 | 13.8 ± 5.7 | 3.5 ± 9.9 | −1.3 ± 384.4 | 24.4 | 1011.6 |
MLR class estimated | 16.0 ± 6.2 | 576.9 ± 281.6 | 16.9 ± 4.7 | 15.6 ± 7.0 | 2.9 ± 12.6 | −221.8 ± 437.9 | 35.3 | 1406.9 |
MLR class estimated with HR | 14.9 ± 4.8 | 513.9 ± 193.5 | 16.4 ± 4.5 | 14.1 ± 4.8 | −1.8 ± 12.3 | −154.2 ± 426.2 | 24.6 | 961.0 |
ANN class estimated | 12.9 ± 4.7 | 419 ± 141.4 | 12.8 ± 3.1 | 12.9 ± 5.4 | 3.4 ± 8.2 | 35.1 ± 284.6 | 19.4 | 659.7 |
ANN class estimated with HR | 13.6 ± 5.4 | 445.8 ± 160.7 | 14.5 ± 5.5 | 13.0 ± 5.5 | 3.0 ± 10.2 | 14.6 ± 336.4 | 23.0 | 816.0 |
Mean absolute error (MAE) for the EE estimation using the ANN model with prior activity classification. The overall MAE was 14.4 ± 5.3%. The MAE in percent for the single activities and classes is presented in dark colors and the MAE in kcal is presented in bright colors.
The evaluation of the two 2.5 h-measurements to validate the algorithm under real-world conditions is shown in Figure
Pre-validation of two subjects using the class-dependent ANN algorithm. The two top plots show the measured EE from the indirect calorimetry and the estimate based on the IMU data, and the bottom plots show the cumulative value of the estimated and the measured EE value. At the end of the measurement period, the real EE was underestimated by 6.1% for subject #1 and overestimated by 4.6% for subject #2.
The metabolic cost of each single activity and the three activity classes is presented in Figure
SCI MET presented for all activities and classes. The gray dots to the right of each bar represent the values for the tetraplegic subjects and the black dots the values for the paraplegic subjects. The bars represent the mean of the pooled data including paraplegic and tetraplegic subjects.
Correlation between perceived exertion, assessed with a numeric rating scale, and measured SCI MET. The correlation coefficient was
The conversion of the recommendations for a healthy lifestyle in the able-bodied population to daily goals in the wheelchair-bound SCI population can be found in Table
Recommendations for the able-bodied population converted to EE during moderate activity, and to distance to travel for the wheelchair-bound SCI population.
10,000 steps of able-bodied | 50, 51 | 300 | 2.0 | 1,650 | 3,430 | 1,780 | ||
150 kcal of additional EE | 56 | 150 | 2.0 | 1,650 | 3,430 | 1,780 | ||
200 kcal of additional EE | 56 | 200 | 2.0 | 1,650 | 3,430 | 1,780 | ||
30 min of moderate activity | 55 | 3.5 | 1,644 | 4,255 | 2,611 | 30 | ||
55 | 5.0 | 1,559 | 4,502 | 2,943 | 30 | |||
60 min of moderate activity | 56 | 3.5 | 1,644 | 4,255 | 2,611 | 60 | ||
56 | 5.0 | 1,559 | 4,502 | 2,943 | 60 | |||
Recommendations based on this study (following ACSM recommendation) | ||||||||
Additional recommendations based on this study | 1,559 | 4,502 | 2,943 | |||||
1,644 | 4,255 | 2,611 | ||||||
1,650 | 3,430 | 1,780 |
The main aim of this study was to develop and validate an algorithm for estimating EE from IMU data applicable in a real-world situation in individuals with SCI. We present a highly accurate method to estimate ADL-dependent EE from IMU recordings. Most importantly, we provide reference values for wheelchair-bound SCI subjects to promote and coach a healthy lifestyle, which could be beneficial for reducing the risk of cardiovascular diseases.
REE reflects the energy used to maintain vital functions at room temperature. To account for its large contribution [~65% in the able-bodied population (
Here, we also investigated whether and how REE estimation models could be improved by considering clinical scores. However, the inclusion of the AIS score did not improve the models and is likely too unspecific in describing the extent of impairment. Also, including the level of injury or the motor scores of the ISNCSCI did not improve the estimation accuracy significantly. The only clinical score improving the MAE of the REE estimation was the SCIM III total score, although it only improved the ANN-based model. In general, all REE estimation models that were developed in this study were based on the data of only 30 subjects. This number is clearly too small to build a general model for this heterogeneous population of SCI subjects. For this reason, subsequent analyses were performed using the updated Harris-Benedict equation for REE estimation.
In the able-bodied population, the type of activity performed by a subject was shown to be of great importance when establishing models to estimate EE (
The MAE of the activity-dependent EE estimation ranged from 14.1% up to 17.3%, when the HR information was not included. However, we have to take into account that the model reaching a MAE of 14.1% assumed a perfect classification into the different activity classes. In general, the overall MAE is in the range of other accelerometer or IMU based models developed for the SCI population although those studies included fewer activities (
The inclusion of the HR showed a slight improvement for the MLR-based model. Thereby, the overall estimation improvement comes mainly from the improvement in the “high intensity” class. This might be due to the fact that the addition of the HR can, to a certain extent, improve the EE estimate of weight loading activities. The validity of combining accelerometer and HR measurements in the SCI population to estimate EE by using linear models has already been shown by Nightingale et al. (
Based on the insights from this study and existing literature for the able-bodied population, we seeked to propose activity-related recommendations for a healthy lifestyle in the SCI population. For subjects with SCI, activity recommendations were translated into daily distance traveled in a manual wheelchair. Since translations from able-bodied to SCI are based on the EE recordings of the present study, we assured that the different SCI MET values in the literature matched the SCI MET values of this study. Collins and co-workers investigated the metabolic cost of 27 different physical activities in 170 adults with SCI (
The most commonly used recommendation for an active lifestyle in the able-bodied population is the ACSM recommendation suggesting 30min of moderate to vigorous activity per day on at least 5 days per week. For the wheelchair-bound SCI population, the results of our study found wheeling at 3.5 or 5 km/h to represent an activity of moderate intensity. In order to choose one of the two speeds for the translation of the ACSM recommendation, we further examined the EE at these two wheeling speeds. According to Wilson et al. (
There exist, however, also other recommendations for the able-bodied population. For the translation of the 10,000 steps/day (roughly 300 kcal/day) to a distance to travel per day in the wheelchair, we selected an average wheeling speed of 2 km/h. This value was based on daily averages obtained from long-term recordings in the SCI population (
Therefore, based on data of the present study we recommend to travel for at least 3 km at 5 km/h on 5 days a week in order to achieve a health-promoting additional daily EE. This recommendation is in line with the recommendations of the U.S. Department of Health and Human Services and the ACSM (
While our recommendation aims at minimizing the risk for cardiovascular diseases, Blair and co-workers stated in their work that 60 min instead of 30 min of moderate to vigorous physical activity per day is beneficial for different health outcomes such as for example, maintaining a lean body mass or improving muscular strength and endurance (
We would like to acknowledge some limitations of this study. Firstly, the number of women included in this study (i.e., 3) may be considered too small (although it reflects a typical distribution in traumatic SCI) and, therefore, the sample measured may not be representative for the entire SCI population. Secondly, no individual HR calibration such as the method presented by Spurr and coworkers was used, which might have influenced the models including the HR (
The models presented in this study accurately estimate EE in an unprecedented pool of 24 activities and in 3 h of continuous measurements in wheelchair-bound SCI individuals, making it a powerful tool to be used during continuous and non-obstructive recordings in real-world situations. IMU-based EE estimation is a promising methodology that may be used, together with the proposed wheeling reference value of 3 km per day, to promote a healthy lifestyle in SCI individuals at later stages of and/or after rehabilitation. The use of such recordings and recommendations may help to increase physical activity of SCI individuals to an extent allowing to decrease the prevalence of cardiovascular disease and increase quality of life in the long run.
WP, MB, BW, CS, AC, MS, and RG designed the study, WP, LR, and MB collected the experimental data, WP and LR performed the data analysis, WP, LR, MB, BW, CS, AC, MS, and RG interpreted the results, revised the manuscript, and approved the final version.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer HS and handling Editor declared their shared affiliation.
The authors would like to thank Thi Dao Nguyen, Fanny Leimgruber, and Sophie Schneider for the help in data collection, Dr. Kaspar Leuenberger and Dr. Mike D. Rinderknecht for the help in data analysis and critical review of this manuscript, Larissa Angst for the help in preparing and testing the study protocol, and all subjects who volunteered for this study.