Training-induced changes in daily energy expenditure : Methodological evaluation using wrist-worn accelerometer, heart rate monitor, and doubly labeled water technique

This is a self-archived version of an original article. This version may differ from the original in pagination and typographic details. Author(s): Title: Year: Version: Copyright: Rights: Rights url: Please cite the original version: CC BY 4.0 https://creativecommons.org/licenses/by/4.0/ Training-induced changes in daily energy expenditure : Methodological evaluation using wrist-worn accelerometer, heart rate monitor, and doubly labeled water technique © 2019 Kinnunen et al. Published version Kinnunen, Hannu; Häkkinen, Keijo; Schumann, Moritz; Karavirta, Laura; Westerterp, Klaas R.; Kyröläinen, Heikki Kinnunen, H., Häkkinen, K., Schumann, M., Karavirta, L., Westerterp, K. R., & Kyröläinen, H. (2019). Training-induced changes in daily energy expenditure : Methodological evaluation using wrist-worn accelerometer, heart rate monitor, and doubly labeled water technique. PLoS ONE, 14(7), Article e0219563. https://doi.org/10.1371/journal.pone.0219563 2019 RESEARCH ARTICLE Training-induced changes in daily energy expenditure: Methodological evaluation using wrist-worn accelerometer, heart rate monitor, and doubly labeled water technique Hannu KinnunenID 1 *, Keijo Ha ¨kkinen 2 , Moritz Schumann 3 , Laura Karavirta 4,5 , Klaas R. Westerterp 6 , Heikki Kyro ¨la ¨inen 2 1Optoelectronics and Measurement Techniques Research Group, University of Oulu, Oulu, Finland, 2Biology of Physical Activity, Faculty of Sport and Health Sciences, University of Jyva ¨skyla ¨, Jyva ¨skyla ¨, Finland, 3Department of Molecular and Cellular Sports Medicine, German Sport University, Cologne, Germany, 4Gerontology Research Center, Faculty of Sport and Health Sciences, University of Jyva ¨skyla ¨, Jyva¨skyla¨, Finland, 5Polar Electro Oy, Kempele, Finland, 6School of Nutrition and Translational Research in Metabolism (NUTRIM), Maastricht University, Maastricht, The Netherlands *[email protected] Abstract Introduction Wrist-mounted motion sensors can quantify the volume and intensity of physical activities, but little is known about their long-term validity. Our aim was to validate a wrist motion sensor in estimating daily energy expenditure, including any change induced by long-term participation in endurance and strength training. Supplemental heart rate monitoring during weekly exercise was also investigated. Methods A 13-day doubly labeled water (DLW) measurement of total energy expenditure (TEE) was performed twice in healthy male subjects: during two last weeks of a 12-week Control period (n = 15) and during two last weeks of a 12-week combined strength and aerobic Training period (n = 13). Resting energy expenditure was estimated using two equations: one with body weight and age, and another one with fat-free mass. TEE and activity induced energy expenditure (AEE) were determined from motion sensor alone, and from motions sensor combined with heart rate monitor, the latter being worn during exercise only. Results When body weight and age were used in the calculation of resting energy expenditure, the motion sensor data alone explained 78% and 62% of the variation in TEE assessed by DLW at the end of Control and Training periods, respectively, with a bias of +1.75 (p <.001) and +1.19 MJ/day (p = .002). When exercise heart rate data was added to the model, the combined wearable device approach explained 85% and 70% of the variation in TEE assessed by DLW with a bias of +1.89 and +1.75 MJ/day (p <.001 for both). While significant increases PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 1 / 19 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Kinnunen H, Ha¨kkinen K, Schumann M, Karavirta L, Westerterp KR, Kyro¨la¨inen H (2019) Training-induced changes in daily energy expenditure: Methodological evaluation using wrist-worn accelerometer, heart rate monitor, and doubly labeled water technique. PLoS ONE 14(7): e0219563. https://doi.org/10.1371/journal. pone.0219563 Editor: Ioannis G. Fatouros, University of Thessaly, GREECE Received: January 3, 2019 Accepted: June 26, 2019 Published: July 10, 2019 Copyright: ©2019 Kinnunen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its supporting information files. Funding: This study was supported by grants from the Finnish Ministry of Education and Culture, and Polar Electro. At the time of data collection, authors HKi and LK were employed by Polar Electro. The funder provided support in the form of salaries for authors HKi and LK but did not have any additional in TEE and AEE were detected by all methods as a result of participation in regular training, motion sensor approach underestimated the change measured by DLW: +1.13±0.66 by DLW, +0.59±0.69 (p = .004) by motion sensor, and +0.98±0.70 MJ/day by combination of motion sensor and heart rate. Use of fat-free mass in the estimation of resting energy expenditure removed the biases between the wearable device estimations and the golden standard reference method of TEE and demonstrated a training-induced increase in resting energy expenditure by +0.18±0.13 MJ/day (p <.001). Conclusions Wrist motion sensor combined with a heart rate monitor during exercise sessions, showed high agreement with the golden standard measurement of daily TEE and its change induced by participation in a long-term training protocol. The positive findings concerning the validity, especially the ability to follow-up the change associated with a lifestyle modification, can be considered significant because they partially determine the feasibility of wearable devices as quantifiers of health-related behavior. Introduction Physical activity is one of the main determinants of individual energy expenditure, and the most important one to explain its changes [1]. Increasing daily total energy expenditure (TEE) supports the regulation of favorable body composition [2], and helps in the prevention and treatment of lifestyle diseases [3,4]. Various methods are commonly used to assess physical activity, including accelerometers and heart rate monitors. The most valid method to measure free-living energy expenditure, yet which is suitable for advanced research use only, is the doubly labeled water (DLW) method [5,6]. Accelerometers, that are essentially motion sensors, have become increasingly popular, both as a research tool [7] and as consumer activity trackers [8]. Wrist-worn motion sensors have been associated with higher wearing times and lower selection bias compared to waistworn sensors [9–11]. Higher wearing times may contribute to higher validity [12] in the long term. However, the validity of wrist-worn motion sensors has been studied much less extensively than other positions—waist, chest, leg or arm—particularly in a long-term setting using DLW-assessment [5]. In a short-term measurement, using indirect calorimeter as a reference, a wrist-worn motion sensor has generally been inferior to other wearing positions, yet at its best it has shown a comparable accuracy to waist-worn sensors [13]. Only two studies have compared wrist-worn motion sensors to DLW in a long-term measurement. The first one [14] was performed with wrist-mounted GENEA accelerometers in pregnant and non-pregnant women, where a vector magnitude based linear regression model explained 26% of the variation in physical activity energy expenditure in the non-pregnant group. In the second study, an early prototype of the wrist-worn Polar Active (Polar Electro, Kempele, Finland) was used [15]; Hand motion frequency, body height and weight combined explained 74% of the variation in DLW measured TEE in soldiers. Both studies used the same population in model development and model testing, and it has not been studied how these findings can be generalized in different populations. Polar Active (Polar Electro, Kempele, Finland) is a wristworn motion sensor that was mainly targeted for children. Good agreement was found Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 2 / 19 role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section. For remaining authors no conflicts of interest are declared. Competing interests: At the time of data collection, authors HKi and LK were employed by Polar Electro. HKi is also an inventor in several patents owned by the funder. Both HKi and LK no longer work for the funder. None of the mentioned interests alters our adherence to PLOS ONE policies on sharing data and materials. between Polar Active and two hip-worn actigraphy devices for the fulfillment of the daily recommendation for moderate to vigorous physical activity of 60 min in a study among 48-year old female and male subjects, even though Polar Active showed higher activity output than the hip-worn monitors [16]. Obviously, different algorithms are optimal for wristworn than waist-worn motion sensors. Heart rate monitors, when individually calibrated, provide an alternative for the objective assessment of exercise induced energy expenditure [17]. According to Plasqui 2013 [5] it is not clear if other physiological data improves the validity of accelerometer methods. When combined with chest mounted motion sensor, heart rate has been shown to improve the estimation of energy expenditure in certain activities, where motion of the waist is not closely related to energy expenditure [18]. Importantly, chest-worn motion sensor combined with heart rate monitor has been shown to increase the accuracy over motion sensor in a long-term DLW- assessment [19] as well. Continuous wearing of a chest-worn heart rate monitor over several days can be uncomfortable. Subsequently, the latest wearable products have incorporated optical measurement of heart rate into wrist-worn devices providing reasonable accuracy in several types of physical activities [20]. DLW is considered the golden standard for long-term assessment of daily energy expenditure. Overall validity of activity monitors can be judged with DLW-assessed activity induced energy expenditure (AEE) as a reference [1]. Observation for 1 or 2 weeks of free-living activity is likely to include the full range of physical activity behavior of an individual [2]. DLW does not separate the contribution of different intensities to TEE. High cost level, special instrumentation and the technical expertise that is required have limited the application of DLW in research. A benefit of motion sensor based assessment of physical activity is the ability to track the amount and timing of activities at different intensity zones, such as inactivity, light-, moderate- and vigorous intensity. The present study focuses on the daily energy expenditure and its’ long-term changes and utilizes DLW as the reference method. In short-term measurements, indirect calorimetry has been widely used as the reference method for assessing the validity of accelerometers in various sports and non-sports activities [21]. Energy metabolism is estimated from respiratory gas exchange measurements utilizing ventilation volume, O 2 and CO 2 concentrations in exhaled air so that O 2 consumption and CO 2 production can be measured and transformed into estimates of energy expenditure via metabolic equations. It has achieved the golden standard status in the estimation of acute energy metabolism due to high accuracy and reproducibility [22]. Indirect calorimeter is the only method that can provide reliable calibration data in freeliving conditions across different intensities from resting levels to high intensity exercise. However, it is not suitable for long-term measurement. Aerobic and strength training have different characteristics in respect to their effects on human energy expenditure during exercise, between exercise sessions, and in the long-term training (Table 1). Aerobic exercise significantly increases energy expenditure during and a few hours after cessation of exercise: the excess post-exercise consumption accounts only 6–15% of the acute net effect [23]. On the other hand, energy expenditure during strength exercise is usually at a moderate level while it can increase REE for several days after a single bout of exercise [24]. Long-term participation in strength training also induces an increase in REE via increased lean body mass; Lean body mass explains up to 80% of the between subject variability in REE [25]. Subsequently, with motion sensor and heart rate monitor, it is quite straightforward to assess the increased energy expenditure of dynamic aerobic training. However, it is more challenging to assess the corresponding effect associated with strength training: acceleration sensor does not reflect static workload during strength exercises nor any changes in REE following exercise. The ability of heart rate monitor to quantify excess post exercise Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 3 / 19 energy expenditure is also limited because heart rate response after exercise varies for many reasons between individuals [26]. Wearable devices and mobile applications provide real time behavioral feedback to users which may increase common understanding of healthy activity levels. This target may not be reached unless the accuracy and reliability are good enough [30]. To be embedded for continuous use in wearable devices, the data processing and algorithms should not include an excess amount of computing or memory use. There is a subsequent need for robust methods to estimate the intensity of physical activity and its role in daily energy expenditure. The aims of the present study were: 1) To assess the validity of a wrist-worn motion sensor in estimating daily energy expenditure in non-athletic male subjects in their normal physical activity state, and after long-term lifestyle change of added participation in regular aerobic and resistance exercise. 2) To evaluate whether the combination of motion sensor and heart rate monitor improves the estimation of daily energy expenditure over the motion sensor alone, considering that the training program consisted of stationary aerobic and resistance exercises. 3) To assess the ability of the combination of motion sensor and heart rate monitor to detect the change in TEE and AEE induced by the lifestyle change of added physical training. There are no prior references on the DLW-assessed validity of wrist-worn motion sensor in non-athletic male population. The special novelty of the present study is the evaluation of wrist-worn motion sensor in detecting a change in energy expenditure induced by a lifestyle modification. This paper also introduces a robust method for combining continuous motion sensor and intermittent heart rate data into a minute-by-minute estimate of the intensity of physical activity. Materials and methods Subjects Fifteen males participated in the present study (Table 2). Subject selection was based on preferential randomization among a group of participants (n = 21) in a larger study that evaluated combined strength and aerobic training. All subjects of the larger study received detailed requirements of the sub-study and were asked to provide a written notice of interest to Table 1. Contribution of sedentary life, aerobic and strength training to daily energy expenditure during exercise, between exercise sessions, and after long-term participation in training. Activity type (Level�) During 45min exercise # Between exercise sessions Long-term effects Inactive life (1.5 MET) 0.4 MJ - Decline in fitness, increase of fat mass, decrements in global body functioning further contribute to an inactive lifestyle and decreased daily EE [27]. Strength training, such as moderate to vigorous circuit training (4.3–8.0 MET) 1.1–2.0 MJ Resting EE mildly elevated for days, such as +0.2–0.4 MJ/ day for 3 days in [6]. The post-exercise effect of intense bouts can be significant, and in the case of short, intense exercises, it may even exceed the acute effect. Increases lean body mass and elevates resting EE, preserves functional capacity [28]. Aerobic training, such as running 8–11 km/h (8.3–11.0 MET) 2.1–2.8 MJ Resting EE elevated for 3–12 hours, totals 6–15% of the acute effect [9]. Increases aerobic fitness and promotes active lifestyle. May decrease other habitual daily physical activity, e.g. in older adults in [29]. EE: Energy Expenditure. �) Example physical activity level estimated in metabolic equivalents (MET) according to [21] (codes 07022, 02035, 02040, 12030, and 12070). #) the corresponding acute accumulated EE for a 45-min exercise session for a subject with basal metabolic rate 8.0 MJ/day (a value typical for a 30-year old male person who weighs about 80 kg). https://doi.org/10.1371/journal.pone.0219563.t001 Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 4 / 19 participate in the sub-study. Out of these subjects, the total number of 15 was then randomly drawn. The larger study has been described earlier in detail [31]. Briefly, prior to inclusion into the study, the subjects were moderately physically active as characterized by irregular participation of walking, cycling or occasionally team sports for not more than 3 times per week. No-one systematically engaged in any endurance or strength training before inclusion into the study. Exclusion criteria included pronounced overweight (BMI �30 kg/m 2 ) as well as acute and chronic illness or diseases that would contraindicate intense physical exercise. A completed health questionnaire and resting ECG were reviewed by a cardiologist prior to commencement of the study. The study was conducted according to the Declaration of Helsinki and ethical approval was given by the Ethics Committee at the University of Jyva¨skyla¨. Information about the possible risks of all study procedures were provided both verbally and in writing before subjects gave their written informed consent. Study timeline and training protocol The subjects underwent a 12-week Control period followed by supervised training of 12 weeks (Training). During Control period, the subjects were asked to maintain their activities of daily living, but no prescribed physical exercise was conducted. During the training of 12 weeks, 4 weekly sessions of strength and aerobic (2 x strength and 2 x aerobic) exercise sessions were conducted on alternating days. The strength training program consisted of heavy resistance loading focusing on muscle growth and maximal strength development (mainly 2–5 sets for 8–10 repetitions at 80–85% of 1RM and 2–5 sets with 3–5 repetitions at 85–95% of 1RM) including typical exercises (6–8 exercises per session) for the lower and upper extremities and trunk. Exercises for the lower body consisted of bilateral dynamic leg press, as well as both bilateral (weeks 1–7) and unilateral (weeks 8–12) dynamic knee extension and flexion. Additional exercises for the upper body included vertical shoulder press and lateral pull down, as well as exercises commonly used to improve trunk stability. The overall duration of each strength protocol was 30–50 min. During weeks 1–2, all exercises were conducted with a circuit model using 2–4 sets of 15–20 repetitions at an intensity of 40–60% of 1RM. During the following 10 weeks of training, protocols aiming for muscle hypertrophy (2–5 ×8–10 repetitions at 80–85% of 1RM, 1.5–2 min Table 2. Descriptive statistics of the subjects, and their adherence to wearing the motion sensor. Parameter Control Training Change P Body mass (kg) 81.6±12.1 82.1±11.8 0.43±1.49 .148 BMI (kg/m 2 ) 25.1±3.4 25.2±3.3 0.14±0.47 .153 FFM (kg) 58.0±6.5 60.0±7.2 1.97±1.48 <.001 FM (kg) 23.5±6.6 22.3±6.0 -1.26±1.24 .002 Fat% 28.4±4.6 26.7±4.5 -1.68±1.15 <.001 _ VO2peakðml=kg=minÞ35.7±7.2 38.7±4.7 2.9±3.7 .009 1RM Leg Press (kg) a) 144.4±25.6 157.0±22.9 12.6±4.1 <.001 Wearing time (%) 93.2±14.1 87.3±19.8 -5.8±12.7 .213 Wearing or night (%) 98.4±2.7 97.4±4.0 -1.0±2.9 .125 Values are mean±SD; n = 13; P-value indicates significance of the Change (increase/decrease). BMI: Body Mass Index; FFM: Fat Free Mass, FM: Fat Mass, _ VO2peak: peak oxygen consumption tested on treadmill. 1RM: one repetition maximum. Wear time is percentage of the duration of the 13-day DLW-assessment. a) One subject did not attend the final maximal strength test, hence n = 12. https://doi.org/10.1371/journal.pone.0219563.t002 Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 5 / 19 rest between the sets) and maximal strength (2–5 ×3–5 repetitions at 85–95% of 1RM, 3–4 min rest between sets), as well as during the last 2-week period protocols targeting explosive strength (2 ×8–10 repetitions at 40% of 1RM, 3–4 min rest between the sets) were incorporated into the training program. The aerobic training program included both steady-state and interval exercise sessions on a bicycle ergometer. The training intensity was controlled by the heart rate monitor. The intensity was progressively increased from steady-state cycling below the individually determined aerobic threshold during weeks 1–7 and included interval sessions above the lactate threshold during weeks 8–12. Similarly, the duration of endurance cycling progressively increased throughout the 12 weeks of training from 30 to 50 minutes. Subjects were instructed to maintain a constant pedaling frequency at approximately 70 rpm during each exercise session, while the magnetic resistance of the ergometer was adjusted to achieve the required heart rate response. The duration of cycling within each training session was 30–50 min. The validation of the wrist worn motion sensor by the DLW method was performed over 2-week periods during the final weeks of the Control and Training periods. All 15 selected subjects completed the Control period, however 2 subjects dropped out during the Training period due to personal reasons. _ VO2peak and muscle strength measurement protocols Peak oxygen consumption ( _ VO2peak) was determined at the start and at the end of the 12-week training period via an incremental cycling test on a bicycle ergometer (Ergometrics 800, Ergoline, Bitz, Germany). The protocol begun at 50 W and increased by 25 W every 2 minutes. Subjects were asked to maintain a pedaling frequency of 70 rpm throughout the test. The test was stopped when the subjects failed to maintain the required cadence for more than 15 seconds. Oxygen uptake was determined breath-by-breath using a gas analyzer (Oxycon Pro, Jaeger, Hoechberg, Germany). On each testing day, air flow calibration was performed using a manual flow calibrator. Before each test, automatic air flow calibration was performed, and the gas analyzer was calibrated using a certified gas mixture of 16% O 2 and 4% CO 2 ._ VO2peak was calculated as the highest _ VO2value averaged over 60 seconds, and HR max was the average HR during the last 60 seconds of the test. One repetition maximum (1RM) of leg extensors was determined at the start and at the end of the 12-week training period using a dynamic horizontal bilateral leg press device (David 210, David Health Solutions, Helsinki, Finland). Following a warm up, a maximum of 5 trials were allowed to obtain a true 1RM. The greatest load that the subject could lift to full knee extension at an accuracy of 1.25 kg was accepted as 1RM. One subject was unable to perform the leg press test at the end of the training period. Energy expenditure and body composition assessed by doubly labeled water DLW based determination of average daily TEE and Total Body Water were measured according to the Maastricht protocol [32]. Briefly, following the collection of a baseline urine sample (day 0), the subjects ingested a weighed amount of 2 H 218 O, resulting in an initial excess body water enrichment of 150 ppm for deuterium and 300 ppm for oxygen-18. Subsequent urine samples were collected in the morning of days 1, 8 and 14 and in the evening of days 1, 8, and 13 of the DLW measurement periods. Body composition parameters (fat-free mass and fat mass) were calculated based on the 0.73 ratio of total body water to fat-free mass [33]. Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 6 / 19 Resting energy expenditure (REE) was calculated using both the Wang [34] and Schofield [35] equations. The selected Wang equation utilizes fat-free mass (REE FFM (MJ/day) = 0.0908 � FFM(kg) + 1.565), and the selected Schofield equation utilizes body weight and age (REE BW , Eqs 1and 2). The FFM based estimation was selected because it also adjusts to body composition change that can be expected as a response to long-term participation in strength training. Additionally, the body weight and age-based estimation was selected since required data in more easily accessible and the corresponding equations are more widely used in wearable devices. REEBW MJ day � �¼0:0632 �BW kgð Þþ2:895;if age 18 30 years ð1Þ REEBW MJ day � �¼0:0481 �BW kgð Þþ3:653;if age 18 30 years ð2Þ Activity induced energy expenditure (AEE) was calculated as 0.9 x TEE–REE; the equation assumes the thermic effect of nutrients to be 10% of TEE, which is based on a normal mixed diet [2]. Energy expenditure assessed by wrist motion sensor and heart rate monitor The subjects were instructed to wear a motion sensor (Polar Active, Polar Electro, Kempele, Finland) on their wrist throughout the 24-week study protocol, and a heart rate monitor (Polar RS800CX) during both supervised exercise sessions and any voluntary physical activity when they expected that the intensity exceeded that of walking. The subjects were also instructed to wear the heart rate monitor when biking. In this study, the motion sensor and heart rate monitor outputs were selected over the two-week periods corresponding with the timing of the DLW assessments. The motion sensor included a capacitive 1-D accelerometer (VTI Technologies, Vantaa, Finland; currently Murata Electronics). The measurement principle of the motion sensor has been presented earlier [15]. Briefly, hand motion frequency and regularity are determined for every 30 seconds, and they are transformed to the Physical Activity Level (in MET units) via an adjustment for body height, with higher body height corresponding to higher activity level at similar motion frequency. The motion sensor units selected for this investigation were tested for consistency before the study period to avoid dropouts due to technical problems. Combining motion sensor data with heart rate data into a single estimate of energy expenditure. The Physical Activity Level (in MET units) time series, which are stored in 30 second epochs by the motion sensor, were exported from the Polar gofit.com web service for further analysis. TEE derived from motion sensor was calculated by multiplying the Physical Activity Level with both estimates of REE. TEE derived from motion sensor and exercise heart rate data was calculated by replacing the motion sensor derived energy expenditure with the corresponding estimation from heart rate monitor whenever the latter was available and indicated moderate to vigorous physical activity (>4 MET). 4 MET corresponds with walking 5 km/h, and our guidance for subjects was to wear a heart rate monitor during voluntary training when they expected the intensity to exceed that of walking. The heart rate monitor derives the energy expenditure estimation using individually measured _ VO2peak and maximal heart rate [17,36]. It was visually verified that each subject had worn the heart rate monitor during each supervised exercise session without obvious measurement errors. Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 7 / 19 Detection and substitution of non-wear time data during night and day. Initially, time in bed was estimated from MET data if the majority of 30 second epochs in a moving 10-min- ute window indicated full rest–Polar Active records 0.875 MET as the sleeping metabolic level after 10 minutes of full rest. Final daily time-to-bed and time-out-of-bed were determined visually as shown in Fig 1. Data corresponding to time in bed were set to the sleeping metabolic level. Non-wear was detected as a consequent absence of movements for more than 90 minutes, a time window that is comparable to that used previously [37]. If non-wear was detected in connection to night-time (between 11.30pm and 11am), the physical activity level was set to the sleeping metabolic level for the corresponding period of time. Occasional daily non-wear data (Table 2) was replaced with individual average physical activity level of daily wear time. Calculated training-induced change in TEE To support statistical power analysis, we calculated the change in TEE that could be expected as a result of our training program. Since training-induced energy expenditure is only a minor part of TEE, and participant behavior may change outside the exercise sessions over the course of the 24-week study, this calculation was targeted to provide a reference value at the group level only. The expected metabolic requirements of aerobic and strength exercise sessions were calculated based on Compendium of Physical Activities [21] with the assumption that exercise [4 x strength (6.0 MET) and 4 x aerobic exercise sessions (8.0 MET) within 13 days, 50 min per session] would replace sedentary (1.5 MET) behavior. To keep this approximation simple, REE was estimated from body weight for each subject. Fig 1. Seven days of physical activity intensity measured by the wearable sensors from one subject. Solid black line shows data measured by the motion sensor 24/7, and solid red line is data from heart rate monitor during four weekly bicycle ergometer and strength exercise sessions. Dashed green line marks time-out-of-bed and time-in-bed. https://doi.org/10.1371/journal.pone.0219563.g001 Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 8 / 19 in Polar Active, an ideal solution would also include a REE estimation that takes body composition—or related parameters—more strongly into account. The widely used REE estimation equations that utilize body dimensions, age and gender as proxies for body composition, were developed decades ago. The relation between these factors and body composition has changed over the recent decades [25]. TEE assessed by DLW includes resting metabolism, but it cannot be quantified separately. Subsequently, we also derived AEE from DLW assessed total energy expenditure by adjusting for estimated REE, the largest component of TEE. We acknowledge that lack of individual REE assessment by indirect calorimetry (respiratory gas analysis) is a limitation in the present study. However, REE was estimated from lean body mass as measured with isotope dilution. Thus, training effects on REE were detected trough changes in lean body mass, the main determinant of REE [25]. It was shown that the use of fat-free mass, instead of body weight and age in the estimation of REE, removed significant biases in the estimation of mean daily TEE and AEE. Long-term positive training adaptations include improved body composition, as was observed in this study. The increase in REE that arose from increased fat-free mass, accounted for +0.18 MJ/day that equates to 16% increase in TEE induced by the 12-week training period. These findings support the use of body composition in the estimation of REE. However, the measurement or estimation of body composition requires special analysis methods and skilled personnel, and it may not be easily applicable in consumer wearable devices. In research settings, bioimpedance measurement might provide the most feasible tool to provide better estimates for REE in the future. Even though the use of fat-free mass in the calculation of REE removed the biases between the wearable devices and DLW when estimating TEE and AEE, it did not narrow the limits of agreement with respect to DLW. The main limitation of the present study is its relatively small sample size (n = 13). However, due to high validity of the DLW analyses and very low number of DLW validations in wrist-worn devices it is reasonable to run validation studies even with relatively small samples. Another limitation is that the intensity of a strength training cannot be controlled and analyzed using heart rate like during endurance exercise, since strength training intensity is essentially defined by neuromuscular rather than cardiovascular load. A typical endurance training session consisted of intensities slightly below the aerobic threshold for a duration of 30–40 minutes and included bouts at and above the lactate threshold. Finally, all subjects in this study were healthy young adult males, which warrants additional studies in different populations. Validating a new method and technology using data collected from a different population from the original development population can reveal certain limitations. The present devices did not include complex modeling; instead, they consisted of piecewise linear estimates of intensity of PA as a function of hand motion frequency [15], and that of HR [17]. The robustness of the algorithms may support their generalizability in different populations—yet the systematic overestimation observed in this study demonstrates a need for a systematic adjustment. Conclusions As a result of participating in the 12-week combined aerobic and strength training intervention, an increased total energy expenditure was observed together with favorable changes in cardiovascular fitness, maximal muscle strength and body composition among moderately active, healthy men. Daily energy expenditure estimates derived via a wrist-mounted motion sensor was strongly associated with corresponding two-week DLW-assessment both during a Control period, and when participating in regular training. As hypothesized, improved agreement with DLW-assessed total and activity induced energy expenditure was achieved by Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 15 / 19 complementing motion sensor with a heart rate monitor during weekly exercise sessions. A combination of wrist motion sensor (Polar Active) and heart rate measurement during exercise (Polar RS800CX) allowed a valid estimate of the change in daily total and activity induced energy expenditure resulting from participation in a training program. While the present results revealed a significant overestimation in the studied wearable device outcomes, positive findings that confirm their validity, especially their ability to detect a change associated with a lifestyle modification, can be considered significant because this helps determine the feasibility of wearable devices as quantifiers of health-related behavior. Supporting information S1 Table. Individual daily energy expenditure from Wearables. Data are averaged separately for control days, non-exercise days and exercise days. (PDF) S2 Table. Reported daily intake on exercise and non-exercise days. Data from control period, and separately on non-exercise and exercise days from training period. (PDF) S1 Dataset. Data necessary to replicate the present study findings. (XLSX) Acknowledgments We thank Polar Electro for providing the wearable devices and technical support for this study. We also thank our study participants for adhering to the study design and extensive testing protocols. Thanks to Jonathan McPhail for proofreading the manuscript. Author Contributions Conceptualization: Hannu Kinnunen, Keijo Ha¨kkinen, Moritz Schumann, Laura Karavirta. Data curation: Hannu Kinnunen, Moritz Schumann. Formal analysis: Hannu Kinnunen, Moritz Schumann, Klaas R. Westerterp. Funding acquisition: Keijo Ha¨kkinen. Investigation: Hannu Kinnunen, Moritz Schumann, Laura Karavirta, Klaas R. Westerterp. Methodology: Hannu Kinnunen, Keijo Ha¨kkinen, Moritz Schumann, Laura Karavirta, Klaas R. Westerterp. Project administration: Keijo Ha¨kkinen, Moritz Schumann. Resources: Keijo Ha¨kkinen. Software: Hannu Kinnunen. Supervision: Keijo Ha¨kkinen, Heikki Kyro¨la¨inen. Validation: Hannu Kinnunen, Moritz Schumann, Laura Karavirta, Heikki Kyro¨la¨inen. Visualization: Hannu Kinnunen. Writing – original draft: Hannu Kinnunen, Moritz Schumann. Writing – review & editing: Hannu Kinnunen, Keijo Ha¨kkinen, Moritz Schumann, Laura Karavirta, Klaas R. Westerterp, Heikki Kyro¨la¨inen. Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 16 / 19 References 1. Westerterp KR. Reliable assessment of physical activity in disease: an update on activity monitors. Curr Opin Clin Nutr Metab Care. 2014 Sep; 17(5):401–6. https://doi.org/10.1097/MCO.0000000000000080 PMID: 25105926 2. Westerterp KR. Physical activity and physical activity induced energy expenditure in humans: measurement, determinants, and effects. Front Physiol. 2013; 4:90. https://doi.org/10.3389/fphys.2013.00090 PMID: 23637685 3. Myers J, McAuley P, Lavie CJ, Despres J-P, Arena R, Kokkinos P. Physical Activity and Cardiorespiratory Fitness as Major Markers of Cardiovascular Risk: Their Independent and Interwoven Importance to Health Status. Prog Cardiovasc Dis. 2015 Jan 1; 57(4):306–14. https://doi.org/10.1016/j.pcad.2014.09. 011 PMID: 25269064 4. Pescatello LS, Franklin BA, Fagard R, Farquhar WB, Kelley GA, Ray CA. Exercise and Hypertension: Med Sci Sports Exerc. 2004 Mar; 36(3):533–53. PMID: 15076798 5. Plasqui G, Bonomi AG, Westerterp KR. Daily physical activity assessment with accelerometers: new insights and validation studies: Accelerometer validity. Obes Rev. 2013 Jun; 14(6):451–62. 6. Tanskanen MM, Westerterp KR, Uusitalo AL, Atalay M, Hakkinen K, Kinnunen HO, et al. Effects of easy-to-use protein-rich energy bar on energy balance, physical activity and performance during 8 days of sustained physical exertion. PloS One. 2012; 7(10):e47771. https://doi.org/10.1371/journal.pone. 0047771 PMID: 23094083 7. Troiano RP, McClain JJ, Brychta RJ, Chen KY. Evolution of accelerometer methods for physical activity research. Br J Sports Med. 2014; 48(13). 8. Evenson KR, Goto MM, Furberg RD. Systematic review of the validity and reliability of consumer-wear- able activity trackers. Int J Behav Nutr Phys Act. 2015 Dec 18; 12:159. https://doi.org/10.1186/s12966- 015-0314-1 PMID: 26684758 9. Fairclough SJ, Noonan R, Rowlands AV, Van Hees V, Knowles Z, Boddy LM. Wear Compliance and Activity in Children Wearing Wrist- and Hip-Mounted Accelerometers: Med Sci Sports Exerc. 2016 Feb; 48(2):245–53. https://doi.org/10.1249/MSS.0000000000000771 PMID: 26375253 10. Hassani M, Kivimaki M, Elbaz A, Shipley M, Singh-Manoux A, Sabia S. Non-Consent to a Wrist-Worn Accelerometer in Older Adults: The Role of Socio-Demographic, Behavioural and Health Factors. Plos One. 2014; 9(10):e110816–e110816. https://doi.org/10.1371/journal.pone.0110816 PMID: 25343453 11. Roth MA, Mindell JS. Who provides accelerometry data? Correlates of adherence to wearing an accelerometry motion sensor: the 2008 Health Survey for England. J Phys Act Health. 2013; 10(1):70–8. PMID: 22398686 12. Herrmann SD, Barreira TV, Kang M, Ainsworth BE. Impact of accelerometer wear time on physical activity data: a NHANES semisimulation data approach. Br J Sports Med. 2014 Feb; 48(3):278–82. https://doi.org/10.1136/bjsports-2012-091410 PMID: 22936409 13. Welch WA, Bassett DR, Thompson DL, Freedson PS, Staudenmayer JW, John D, et al. Classification Accuracy of the Wrist-Worn Gravity Estimator of Normal Everyday Activity Accelerometer. Med Sci Sports Exerc. 2013; 45(10):2012–9. https://doi.org/10.1249/MSS.0b013e3182965249 PMID: 23584403 14. van Hees VT, Renstro¨m F, Wright A, Gradmark A, Catt M, Chen KY, et al. Estimation of Daily Energy Expenditure in Pregnant and Non-Pregnant Women Using a Wrist-Worn Tri-Axial Accelerometer. Ruiz JR, editor. PLoS ONE. 2011 Jul 29; 6(7):e22922. https://doi.org/10.1371/journal.pone.0022922 PMID: 21829556 15. Kinnunen H, Tanskanen M, Kyrolainen H, Westerterp KR. Wrist-worn accelerometers in assessment of energy expenditure during intensive training. Physiol Meas. 2012; 33(11):1841–54. https://doi.org/10. 1088/0967-3334/33/11/1841 PMID: 23110981 16. Leinonen A-M, Ahola R, Kulmala J, Hakonen H, Va ¨ha ¨-Ypya ¨H, Herzig K-H, et al. Measuring Physical Activity in Free-Living Conditions—Comparison of Three Accelerometry-Based Methods. Front Physiol [Internet]. 2017 Jan 10; 7. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5222829/ 17. Crouter S, Albright C, Bassett D. Accuracy of polar S410 heart rate monitor to estimate energy cost of exercise. Med Sci Sports Exerc. 2004; 36(8):1433–9. PMID: 15292754 18. Spierer DK, Hagins M, Rundle A, Pappas E. A comparison of energy expenditure estimates from the Actiheart and Actical physical activity monitors during low intensity activities, walking, and jogging. Eur J Appl Physiol. 2011; 111(4):659–67. https://doi.org/10.1007/s00421-010-1672-7 PMID: 20953878 19. Villars C, Bergouignan A, Dugas J, Antoun E, Schoeller DA, Roth H, et al. Validity of combining heart rate and uniaxial acceleration to measure free-living physical activity energy expenditure in young men. J Appl Physiol. 2012; 113(11):1763–71. https://doi.org/10.1152/japplphysiol.01413.2011 PMID: 23019315 Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 17 / 19 20. Horton JF, Stergiou P, Fung TS, Katz L. Comparison of Polar M600 Optical Heart Rate and Ecg Heart Rate during Exercise. Med Sci Sports Exerc. 2017 Dec 1; 49(12):2600–7. https://doi.org/10.1249/MSS. 0000000000001388 PMID: 29135785 21. Ainsworth BE, Haskell WL, Herrmann SD, Meckes N, Bassett DR, Tudor-Locke C, et al. 2011 Compendium of Physical Activities: a second update of codes and MET values. Med Sci Sports Exerc. 2011 Aug; 43(8):1575–81. https://doi.org/10.1249/MSS.0b013e31821ece12 PMID: 21681120 22. Haugen HA, Chan L-N, Li F. Indirect calorimetry: a practical guide for clinicians. Nutr Clin Pract Off Publ Am Soc Parenter Enter Nutr. 2007 Aug; 22(4):377–88. 23. Laforgia J, Withers RT, Gore CJ. Effects of exercise intensity and duration on the excess post-exercise oxygen consumption. J Sports Sci. 2006 Dec 1; 24(12):1247–64. https://doi.org/10.1080/ 02640410600552064 PMID: 17101527 24. Heden T, Lox C, Rose P, Reid S, Kirk EP. One-set resistance training elevates energy expenditure for 72 h similar to three sets. Eur J Appl Physiol. 2011 Mar; 111(3):477–84. https://doi.org/10.1007/ s00421-010-1666-5 PMID: 20886227 25. Mu¨ller MJ, Langemann D, Gehrke I, Later W, Heller M, Glu¨er CC, et al. Effect of Constitution on Mass of Individual Organs and Their Association with Metabolic Rate in Humans—A Detailed View on Allometric Scaling. PLOS ONE. 2011 Jul 26; 6(7):e22732. https://doi.org/10.1371/journal.pone.0022732 PMID: 21818376 26. Darr KC, Bassett DR, Morgan BJ, Thomas DP. Effects of age and training status on heart rate recovery after peak exercise. Am J Physiol. 1988 Feb; 254(2 Pt 2):H340–343. https://doi.org/10.1152/ajpheart. 1988.254.2.H340 PMID: 3344824 27. Chaput J-P, Klingenberg L, Rosenkilde M, Gilbert J-A, Tremblay A, Sjo ¨din A. Physical activity plays an important role in body weight regulation. J Obes. 2011; 2011. 28. Garber CE, Blissmer B, Deschenes MR, Franklin BA, Lamonte MJ, Lee I-M, et al. American College of Sports Medicine position stand. Quantity and quality of exercise for developing and maintaining cardiorespiratory, musculoskeletal, and neuromotor fitness in apparently healthy adults: guidance for prescribing exercise. Med Sci Sports Exerc. 2011 Jul; 43(7):1334–59. https://doi.org/10.1249/MSS. 0b013e318213fefb PMID: 21694556 29. Meijer EP, Westerterp KR, Verstappen FT. Effect of exercise training on total daily physical activity in elderly humans. Eur J Appl Physiol. 1999 Jun; 80(1):16–21. 30. Evenson KR, Goto MM, Furberg RD. Systematic review of the validity and reliability of consumer-wear- able activity trackers. Int J Behav Nutr Phys Act. 2015;12(Journal Article):159–159. 31. Schumann M, Yli-Peltola K, Abbiss CR, Hakkinen K. Cardiorespiratory Adaptations during Concurrent Aerobic and Strength Training in Men and Women. Plos One. 2015 Sep 29; 10(9):e0139279. https:// doi.org/10.1371/journal.pone.0139279 PMID: 26418015 32. Westerterp K, Wouters L, Lichtenbelt W. The Maastricht Protocol for the Measurement of Body-Compo- sition and Energy-Expenditure with Labeled Water. Obes Res. 1995 Mar; 3:49–57. 33. Wang Z, Deurenberg P, Heymsfield SB. Cellular-level body composition model. A new approach to studying fat-free mass hydration. Ann N Y Acad Sci. 2000 May; 904:306–11. https://doi.org/10.1111/j. 1749-6632.2000.tb06472.x PMID: 10865761 34. Wang Z, Heshka S, Gallagher D, Boozer CN, Kotler DP, Heymsfield SB. Resting energy expenditurefat-free mass relationship: new insights provided by body composition modeling. Am J Physiol Endocrinol Metab. 2000 Sep; 279(3):E539–545. https://doi.org/10.1152/ajpendo.2000.279.3.E539 PMID: 10950820 35. Schofield WN. Predicting basal metabolic rate, new standards and review of previous work. Hum Nutr Clin Nutr. 1985; 39 Suppl 1:5–41. 36. Kinnunen H, Nissila S. Method and equipment for human-related measuring [Internet]. US20010023320A1, 2001 [cited 2019 Feb 1]. https://patents.google.com/patent/US20010023320A1/ en?oq=US2001023320 37. Matthews CE, Chen KY, Freedson PS, Buchowski MS, Beech BM, Pate RR, et al. Amount of time spent in sedentary behaviors in the united states, 2003–2004. Am J Epidemiol. 2008; 167(7):875–81. https://doi.org/10.1093/aje/kwm390 PMID: 18303006 38. Hautala A, Martinmaki K, Kiviniemi A, Kinnunen H, Virtanen P, Jaatinen J, et al. Effects of habitual physical activity on response to endurance training. J Sports Sci. 2012; 30(6):563–9. https://doi.org/10.1080/ 02640414.2012.658080 PMID: 22315980 39. Sandvei M, Jeppesen PB, Stoen L, Litleskare S, Johansen E, Stensrud T, et al. Sprint interval running increases insulin sensitivity in young healthy subjects. Arch Physiol Biochem. 2012; 118(3):139–47. https://doi.org/10.3109/13813455.2012.677454 PMID: 22540332 Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 18 / 19 40. Koehler K, Braun H, de Mare ´es M, Fusch G, Fusch C, Schaenzer W. Assessing energy expenditure in male endurance athletes: validity of the SenseWear Armband. Med Sci Sports Exerc. 2011 Jul; 43 (7):1328–33. https://doi.org/10.1249/MSS.0b013e31820750f5 PMID: 21131865 41. Wallen MP, Gomersall SR, Keating SE, Wisløff U, Coombes JS. Accuracy of Heart Rate Watches: Implications for Weight Management. PLOS ONE. 2016 May 27; 11(5):e0154420. https://doi.org/10. 1371/journal.pone.0154420 PMID: 27232714 42. Westerterp K, Brouns F, Saris WHM, ten Hoor F. Comparison of doubly labeled water with respirometry at low- and high-activity levels. J Appl Physiol Bethesda Md 1985. 1988 Aug 1; 65:53–6. 43. Drenowatz C, Grieve GL, DeMello MM. Change in energy expenditure and physical activity in response to aerobic and resistance exercise programs. SpringerPlus. 2015 Dec 1; 4(1):798. Training-induced changes in daily energy expenditure measured with wearable devices and doubly labeled water PLOS ONE | https://doi.org/10.1371/journal.pone.0219563 July 10, 2019 19 / 19