Activity monitors provide an inexpensive and convenient way to measure sleep, yet relatively few studies have been conducted to validate the use of these devices in examining measures of sleep quality or sleep stages and if other measures, such as thermometry, could inform their accuracy. The purpose of this study was to compare one research-grade and four consumer-grade activity monitors on measures of sleep quality (sleep efficiency, sleep onset latency, and wake after sleep onset) and sleep stages (awake, sleep, light, deep, REM) against an electroencephalography criterion. The use of a skin temperature device was also explored to ascertain whether skin temperature monitoring may provide additional data to increase the accuracy of sleep determination. Twenty adults stayed overnight in a sleep laboratory during which sleep was assessed using electroencephalography and compared to data concurrently collected by five activity monitors (research-grade: ActiGraph GT9X Link; consumer-grade: Fitbit Charge HR, Fitbit Flex, Jawbone UP4, Misfit Flash) and a skin temperature sensor (iButton). The majority of the consumer-grade devices overestimated total sleep time and sleep efficiency while underestimating sleep onset latency, wake after sleep onset, and number of awakenings during the night, with similar results being seen in the research-grade device. The Jawbone UP4 performed better than both the consumer- and research-grade devices, having high levels of agreement overall and in epoch-by-epoch sleep stage data. Changes in temperature were moderately correlated with sleep stages, suggesting that addition of skin temperature could increase the validity of activity monitors in sleep measurement.
Natashia Swalve, Brianna Harfmann, John Mitrzyk and Alexander H. K. Montoye
Alexander H.K. Montoye, John Vusich, John Mitrzyk and Matt Wiersma
Background: Consumer-based activity monitors use accelerometers to estimate Calories (kcals), but it is unknown if monitors measuring heart rate (HR) use HR in kcal prediction. Purpose: Determine if there is a difference in kcal estimations in Fitbits measuring HR compared to those not measuring HR. Methods: Participants (n = 23) wore five Fitbits and performed nine activities for five minutes each, split into four groupings (G1: sitting, standing, cycling 50–150W; G2: level (0%) and inclined (10%) walking at 1.1 m/s; G3: level (0%) and inclined (10%) walking at 1.4 m/s; G4: level (0%) and inclined (3%) jogging at 2.2–4.5 m/s) in the laboratory. Three Fitbits (Blaze, Charge HR, Alta HR) assessed steps, HR, and kcals, and two Fitbits (Alta, Flex2) assessed steps and kcals. Steps, HR, and kcals data from the Fitbits were compared to criterion measures and between Fitbits measuring HR and Fitbits without HR. Results: Fitbits with HR had significantly higher kcal predictions (10.5–23.8% higher, p < .05) during inclined compared to level activities in G2–G4, whereas Fitbits without HR had similar kcal estimates between level and inclined activities. Mean absolute percent errors for kcal predictions were similar for Fitbits measuring HR (33.7–38.3%) and Fitbits without HR (32.4–36.6%). Conclusion: Fitbits measuring HR appear to use HR when predicting kcals. However, kcal prediction accuracies were similarly poor compared to Fitbits without HR compared to criterion measures.
Alexander H.K. Montoye, Jordana Dahmen, Nigel Campbell and Christopher P. Connolly
Purpose: This purpose of this study was to validate consumer-based and research-grade PA monitors for step counting and Calorie expenditure during treadmill walking. Methods: Participants (n = 40, 24 in second trimester and 16 in third trimester) completed five 2-minute walking activities (1.5–3.5 miles/hour in 0.5 mile/hour increments) while wearing five PA monitors (right hip: ActiGraph Link [AG]; left hip: Omron HJ-720 [OM]; left front pants pocket: New Lifestyles NL 2000 [NL]; non-dominant wrist: Fitbit Flex [FF]; right ankle: StepWatch [SW]). Mean absolute percent error (MAPE) was used to determine device accuracy for step counting (all monitors) and Calorie expenditure (AG with Freedson equations and FF) compared to criterion measures (hand tally for steps, indirect Calorimetry for Calories). Results: For step counting, the SW had MAPE ≤ 10% at all walking speeds, and the OM and NL had MAPE ≤ 10% for all speeds but 1.5 miles/hour. The AG had MAPE ≤ 10% for only 3.0–3.5 miles/hour speeds, and the FF had high MAPE for all speeds. For Calories, the FF and AG had MAPE > 10% for all speeds, with the FF overestimating Calories expended. Trimester did not affect PA monitor accuracy for step counting but did affect accuracy for Calorie expenditure. Conclusion: The ankle-worn SW and hip-worn OM had high accuracy for measuring step counts at all treadmill walking speeds, whereas the NL had high accuracy for speeds ≥2.0 miles/hour. Conversely, the monitors tested for Calorie expenditure have poor accuracy and should be interpreted cautiously for walking behavior.
Alexander H.K. Montoye, Kimberly A. Clevenger, Kelly A. Mackintosh, Melitta A. McNarry and Karin A. Pfeiffer
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.
Scott A. Conger, Alexander H.K. Montoye, Olivia Anderson, Danielle E. Boss and Jeremy A. Steeves
Speed of movement has been shown to affect the validity of physical activity (PA) monitors during locomotion. Speed of movement may also affect the validity of accelerometer-based PA monitors during other types of exercise. Purpose: To assess the ability of the Atlas Wearables Wristband2 (a PA monitor developed specifically for resistance training [RT] exercise) to identify the individual RT exercise type and count repetitions during RT exercises at various movement speeds. Methods: 50 male and female participants completed seven sets of 10 repetitions for five different upper/lower body RT exercises while wearing a Wristband2 on the left wrist. The speed of each set was completed at different metronome-paced speeds ranging from a slow speed of 4 sec·rep−1 to a fast speed of 1 sec·rep−1. Repeated Measures ANOVAs were used to compare the actual exercise type/number of repetitions among the seven different speeds. Mean absolute percent error (MAPE) and bias were calculated for repetition counting. Results: For each exercise, there tended to be significant differences between the slower speeds and the fastest speed for activity type identification and repetition counting (p < .05). Across all exercises, the highest accuracy for activity type identification (91 ± 1.8% correct overall), repetition counting (8.77 ± 0.17 of 10 reps overall) and the lowest MAPE (14 ± 1.7% overall) and bias (−1.23 ± 0.17 reps overall) occurred during the 1.5 sec·rep−1 speed (the second fastest speed tested). Conclusions: The validity of the Atlas Wearables Wristband2 to identify exercise type and count repetitions varied based on the speed of movement during RT exercises.
Alexander H.K. Montoye, Scott A. Conger, Joe R. Mitrzyk, Colby Beach, Alecia K. Fox and Jeremy A. Steeves
Background: Resistance training (RT) is an integral component of physical activity guidelines, but methods for the objective assessment of RT have been limited. Recently, the Atlas Wearables Wristband2 has been marketed to measure RT, but its reliability is unknown. Purpose: To determine the reliability of the Wristband2 for measuring RT exercises. Methods: Participants (n = 62) aged 18–52 yrs. wore two Wristband2 monitors on the left wrist and performed 2 sets of 12 repetitions of 14 different resistance training exercises. Test-retest reliability was determined by calculating percent agreement for exercise type and for repetitions recorded by a single Wristband2 between sets 1 and 2 for each exercise, and inter-monitor reliability was determined by calculating percent agreement for exercise type and for repetitions recorded by both Wristband2 monitors in set 1 of each exercise. Results: Test-retest reliability for exercise type was 80.0 ± 1.0% (lowest: 69.4% for bench press; highest: 95.2% for biceps curls) and for repetition count was 47.9 ± 2.2% (lowest: 19.4% for calf raises; highest: 82.3% for lateral raises). Inter-monitor reliability for exercise type was 80.4 ± 1.3% (lowest: 66.1% for bench press; highest: 95.2% for biceps curls) and for repetition count was 59.6 ± 2.2% (lowest: 32.3% for calf raises; highest: 88.7% for lateral raises). Subgroup analyses by gender, RT experience, and participant height revealed minimal differences in reliability. Repetition agreement of ≤1 repetition increased test-retest reliability to 74.7% and inter-monitor reliability to 83.7%. Conclusion: The Wristband2 had acceptable test-retest and inter-monitor reliability for the majority of exercises tested and for counting repetitions to within 1 repetition/set.
Jeremy A. Steeves, Scott A. Conger, Joe R. Mitrzyk, Trevor A. Perry, Elise Flanagan, Alecia K. Fox, Trystan Weisinger and Alexander H.K. Montoye
Background: Devices for monitoring physical activity have focused mainly on measuring aerobic activity; however, the 2018 Physical Activity Guidelines for Americans also recommend muscle-resistance training two or more days per week. Recently, a wrist-worn activity monitor, the Atlas Wristband2, was developed to recognize resistance training exercises. Purpose: To assess the ability of the Wristband2 to identify the type and number of repetitions of resistance training exercises, when worn on the left wrist as directed by the manufacturer, and when worn on the right wrist. Methods: While wearing monitors on both wrists, 159 participants completed a circuit-style workout consisting of two sets of 12 repetitions of 14 different resistance training exercises. Data from the monitors were used to determine classification accuracies for identifying exercise type verses direct observation. The average repetitions and mean absolute error (MAE) for repetitions were calculated for each exercise. Results: The Wristband2 classification accuracy for exercise type was 78.4 ± 2.5%, ranging from 54.7 ± 3.4% (dumbbell [DB] bench press) to 97.5 ± 1.0% (DB biceps curls), when worn on the left wrist. An average of 11.0 ± 0.2 repetitions, ranging from 9.0 ± 0.3 repetitions (DB lunges) to 11.9 ± 0.1 repetitions (push-ups), were identified. For all exercises, MAE ranged from 0.0–4.6 repetitions. When worn on the right wrist, exercise type classification accuracy dropped to 24.2 ± 5.1%, and repetitions decreased to 8.1 ± 0.8 out of 12. Conclusions: The Wristband2, worn on the left wrist, had acceptable exercise classification and repetition counting capabilities for many of the 14 exercises used in this study, and may be a useful tool to objectively track resistance training.