Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

A Smartwatch Step-Counting App for Older Adults: Development and Evaluation Study

A Smartwatch Step-Counting App for Older Adults: Development and Evaluation Study Background: Older adults who engage in physical activity can reduce their risk of mobility impairment and disability. Short amounts of walking can improve quality of life, physical function, and cardiovascular health. Various programs have been implemented to encourage older adults to engage in physical activity, but sustaining their motivation continues to be a challenge. Ubiquitous devices, such as mobile phones and smartwatches, coupled with machine-learning algorithms, can potentially encourage older adults to be more physically active. Current algorithms that are deployed in consumer devices (eg, Fitbit) are proprietary, often are not tailored to the movements of older adults, and have been shown to be inaccurate in clinical settings. Step-counting algorithms have been developed for smartwatches, but only using data from younger adults and, often, were only validated in controlled laboratory settings. Objective: We sought to develop and validate a smartwatch step-counting app for older adults and evaluate the algorithm in free-living settings over a long period of time. Methods: We developed and evaluated a step-counting app for older adults on an open-source wrist-worn device (Amulet). The app includes algorithms to infer the level of physical activity and to count steps. We validated the step-counting algorithm in the lab (counting steps from a video recording, n=20) and in free-living conditions—one 2-day field study (n=6) and two 12-week field studies (using the Fitbit as ground truth, n=16). During app system development, we evaluated 4 walking patterns: normal, fast, up and down a staircase, and intermittent speed. For the field studies, we evaluated 5 different cut-off values for the algorithm, using correlation and error rate as the evaluation metrics. Results: The step-counting algorithm performed well. In the lab study, for normal walking (R =0.5), there was a stronger correlation between the Amulet steps and the video-validated steps; for all activities, the Amulet’s count was on average 3.2 (2.1%) steps lower (SD 25.9) than the video-validated count. For the 2-day field study, the best parameter settings led to an association between Amulet and Fitbit (R =0.989) and 3.1% (SD 25.1) steps lower than Fitbit, respectively. For the 12-week field study, the best parameter setting led to an R value of 0.669. Conclusions: Our findings demonstrate the importance of an iterative process in algorithm development before field-based deployment. This work highlights various challenges and insights involved in developing and validating monitoring systems in https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 1 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al real-world settings. Nonetheless, our step-counting app for older adults had good performance relative to the ground truth (a commercial Fitbit step counter). Our app could potentially be used to help improve physical activity among older adults. (JMIR Aging 2022;5(3):e33845) doi: 10.2196/33845 KEYWORDS step tracking; step counting; pedometer; wearable; smartwatch; older adults; physical activity; machine learning; walking; mHealth; mobile health; mobile app; mobile application; app; uHealth and have been shown to have wide error rates in such contexts Introduction [11,12]. A few smartwatch-based step-counting algorithms have been developed using data from young adults and validated in Older adults are faced with an increased risk of developing controlled settings only [13,14]. Matthies et al [15] developed multiple comorbid medical conditions, social isolation, and a smartwatch step-counting app for older adults who use a reduced physical function, which can lead to an increased risk walking frame equipped with wheels, which was evaluated of disability [1]. An inability to engage in activities of daily outdoors, but only in a controlled setting, with 5 older adults. living may increase mortality risk and premature nursing home To the best of our knowledge, a smartwatch step-counting app placement [2]. Hence, it is critical to encourage older adults for older adults has not been developed and validated in with multimorbidity to engage in interventions that improve free-living settings over a long period of time with a large health, including physical activity. In fact, simple sample. community-based walking programs and resistance-based programs [3] have effectively demonstrated reductions in the We previously developed GeriActive, an app that measures the long-term risk of disability [4]. Even short bouts of walking can daily activity levels (low, moderate, or vigorous) of older adults improve quality of life, physical function, and cardiovascular [16]. We aimed to develop and validate a smartwatch-based fitness in older adults [5]. step-counting algorithm for older adults that runs as an app on the Amulet device. The Amulet is an open-source wrist-worn Traditional consumer-based health-promoting endeavors, such device that has been used for various mobile health studies, as Silver Sneakers [6], have been scaled and widely such as stress and physical activity monitoring [17,18]. The disseminated across the United States. Randomized control Amulet platform enables developers to write energy- and trials have also shown the short- and long-term benefits of memory-efficient apps. physical activity. However, sustained motivation continues to be a challenge for many individuals. Simple encouragement by Methods clinicians may enhance engagement [7]. Yet, a study of accelerometry data demonstrated that only 7.6% of older adults Study Overview aged 65 years meet Physical Activity Guidelines for Americans We validated the step-counting algorithm with older adults in [8]. These pragmatic challenges demonstrate the need to the lab (using videorecordings as ground truth) and in 2 overcome the barriers of traditional health promotion to enhance free-living studies (using the Fitbit as ground truth) lasting 2 self-efficacy and behavioral change. and 12 days. Older adults are the fastest-growing group of technology users; Overview of Step Counter App one survey suggested that 61% of older adults use smartphones [9], an increase from 23% in 2013 [10]. In fact, over 50% of Our step-counting Amulet app estimates the number of steps older adults use social media [9]. Remote monitoring using taken over the course of a day and displays the information on fitness devices has now become ubiquitous in many countries the Amulet screen, similar to the functionality of a pedometer where technology is readily available. In both consumer-based or of other wearable devices (Figure 1). The app continuously and academic-focused trials, it continues to be challenging to estimates the number of steps for each 5-second window, find a balance between clinical accuracy and ease of use. Current updating the count viewed by the user. It uses a 2-step process: algorithms in consumer devices (such as Fitbit) are proprietary machine learning is used to determine if physical activity and often are not tailored to the group being evaluated in a occurred in the most recent 5 seconds of data, and then, the clinical setting, such as older adults in free-living conditions, number of steps is estimated by counting the number of peaks. https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 2 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al Figure 1. The Step Counter App with the step count displayed at the bottom. Dartmouth-Hitchcock Institutional Review Board (28905). All Activity-Detection Algorithm participants provided signed informed consent. We used a linear support vector machine that classifies each Participants 5-second window of accelerometer data into low, moderate, or vigorous activity [16]. We trained the algorithm on data Participants were recruited through the Center for Health and collected from older adults who performed various physical Aging at Dartmouth and primary care practices at activities: sitting, standing, lying down, walking, and running Dartmouth-Hitchcock using flyers, listservs, and posters. This [16,19]. Our evaluation of the algorithm produced an accuracy was a convenience sample; our results may not necessarily be of 91.7% using leave-one-subject-out cross-validation. If the applicable to other groups. output of the algorithm is moderate or vigorous, the Laboratory Study step-detection algorithm is run to determine the number of steps Data for the development and evaluation of the step-detection in the 5-second window. This 2-phase approach was necessary algorithm were collected at our Dartmouth campus laboratory. to reduce false positives by preventing various actions such as Older adults (n=20, age ≥65 years) were asked to perform random hand motions from being counted as steps. different types of walking (normal, fast, up and down a staircase, Step-Detection Algorithm and intermittent) while wearing an Amulet. The Amulet The step-detection algorithm estimates the number of steps in collected raw acceleration data at a frequency of 20 Hz and 5 seconds of acceleration data. The algorithm uses the magnitude logged the magnitude, which we later used to develop the step of the acceleration of the 3-axis accelerometer of the Amulet. algorithm. The participants were videotaped. The video was It is a 3-stage process consisting of zero-meaning, filtering, and independently reviewed to count steps by 2 individuals peak detection, using an approach similar to that described in independently (JAB, RKM) and any differences were later [13]. First, to ensure the data have a mean of zero, for each reconciled. We used these data for the preliminary development sample, we subtract the average of the preceding 20 data of the step-detection algorithm and evaluated the algorithm samples. Subsequently, a moving-average filter is applied, that using the error rate (the percentage difference between the is, each sample is replaced with the average of the 3 preceding algorithm’s estimated step count and the ground-truth step count samples. Finally, peaks in the filtered signal are identified by measured from the video). checking for change of slope. If the slope changes from positive 2-Day Field Study to negative, and the peak value is above a certain threshold, We conducted a 2-day field study in which older adults (n=7, then the peak is counted as a step. The cut-off threshold was age ≥65 years) wore an Amulet device (running our step counter initially empirically determined and then tuned. app) and a Fitbit Flex 2 device (Fitbit Inc) on the same wrist Ethics for 2 days. We compared each participant’s step count estimated Studies were approved by the Committee for the Protection of by their Amulet (exploring 5 different cut-off values) with their Human Subjects at Dartmouth College and the step count reported by the Fitbit (downloaded from the https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 3 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al individual’s research-based Fitbit account). The error rate Participants (Table 1) from both rounds wore an Amulet and between Fitbit’s steps and Amulet’s steps for each of the 5 peak Fitbit on the same wrist for 12 weeks. The Amulet logged the cut-off values was computed. summary steps on an SD memory card hourly and at the end of each day. A research assistant copied the data from each 12-Week Field Study participant’s SD card on a biweekly basis. The Fitbit logged a We conducted a field study (2 cohorts, 12 weeks each) to test summary of each day’s step count (computed by a proprietary the step-detection algorithm with the target population—older algorithm) to the Fitbit app on the participant’s Android tablet, adults with obesity. This study was a subset of a larger study which uploaded the data to the Fitbit servers; we subsequently that evaluated the impact of a combined weight loss and exercise used the Fitbit research API to download participants’ data. intervention in older adults with obesity [20]. The goal of this After 10 weeks of monitoring data from the first cohort, we analysis was to compare the Amulet’s step-count estimate with modified the step-detection peak cut-off value to minimize the the Fitbit’s step count over a long period in real-world settings. error rate relative to ground-truth step count data from the Fitbit. Table 1. Participant characteristics. Characteristic Value (n=16) Age (years) Mean (SD) 74.1 (5.6) Range 66-87 Sex, n (%) Male 4 (25) Female 12 (75) Marital status, n (%) Married 7 (44) Divorced 8 (50) Widowed 1 (6) Smoking history, n (%) None 13 (81) Formerly smoked 3 (19) Education, n (%) High school 2 (12) Some college 5 (31) College degree 3 (19) Postcollege degree 6 (38) Weight (kg), mean (SD) 97.06 (18.2) 36.8 (4.9) Body mass index (kg/m ), mean (SD) Multimorbidity, n (%) 14 (87) Comorbidities, n (%) Diabetes 2 (12) Fibromyalgia 1 (6) High cholesterol 6 (38) Hypertension 9 (56) Osteoarthritis 6 (38) We thus limited the data to days for which it was safe to assume Statistical Analysis that they were worn for nearly the same amount of time. The We calculated the mean with standard deviation for continuous Amulet was able to detect if it was worn each hour, so we measures and count with percentage for categorical measures. considered the Amulet to be worn if the Amulet was worn for Although participants were instructed to wear the Amulet and 75% of a 15-hour day (675 minutes); the Fitbit only reported Fitbit simultaneously, not all participants did so the entire time. daily step count, so we considered the Fitbit to be worn if the https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 4 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al step count was greater than 100. We selected these parameters between Amulet and Fitbit steps. For our analysis, we defined from an understanding of the distribution of wear time over the significance as P<.05. course of a 24-hour day and the distribution of steps expected for this population per day [21]. Comparisons between Amulet Results and Fitbit were limited to data from days for which both were Laboratory Study worn. We conducted bivariate linear regression to compare the association between Amulet and video-counted steps during There was a strong linear association when participants walked various activities (laboratory study) and between Amulet and normally (Figure 2; Table 2). For normal walking, the Amulet Fitbit steps (2-day and 12-week field studies). We compared step-detection algorithm undermeasured the number of steps Amulet steps to Fitbit steps using percentage difference taken by an average of 6.7 steps (SD 32.6). Across all activities, (difference between Amulet and Fitbit steps divided by Fitbit the Amulet was on average 3.2 steps lower (SD 25.9) or 2.1% steps).We used Bland-Altman plots to compare the agreement (SD 31.9%) lower than video-validated steps (Figure 2). The distribution was slightly positively skewed. Figure 2. Association between Amulet-estimated steps and video-validated steps. Table 2. Step count for different walking activities. Activity Video-validated, n (%) Amulet, n (%) Percentage error Fast 102.62 (14.7) 110.72 (22.7) 8.53 (20.02) Intermittent 101.02 (18.7) 99.48 (30.8) 1.71 (33.28) Normal 84.9 (31.9) 77.14 (33.9) –6.68 (32.55) Stairs 52.76 (13.4) 45.52 (13.2) –7.55 (36.97) https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 5 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al the highest association between Amulet and Fitbit steps 2-Day Field Study (R =0.989). Cut-off threshold number 2 had the smallest mean We discarded 1 participant’s data because the data indicated percentage difference between Amulet and Fitbit (–3.1%, SD the devices had not been worn much (step counts were less than 25.1) of all cut-off thresholds (threshold 1: mean 15.27%, SD 350 per day). The associations between each participant’s daily 33.19; threshold 3: mean –10.77, SD 23.43; threshold 4: mean step counts reported by the Amulet and Fitbit were high for all 7.18, SD 29.28; threshold 5: mean –4.11, SD 24.94) (Figure 4). cut-off thresholds (Figure 3). The third cut-off threshold had Figure 3. Association between Amulet and Fitbit steps for different cut-off values: 2-day field study. Figure 4. Distribution of percentage difference between Amulet and Fitbit steps by cut-off threshold. https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 6 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al association between Fitbit steps and Amulet steps improved 12-Week Field Study Results (first cut-off threshold: R =0.386; second cut-off threshold: Across both 12-day field studies, there were 297 participant-days for which both the Fitbit and Amulet had been worn. Cohort 2 R =0.669) (Figure 5). There was improved agreement between used the modified app. For the first cut-off threshold (version both measures with the second cut-off threshold (Figure 6). 1), we recorded 86 participant-days; the average Fitbit step There were 9 observations by 5 unique participants with count was 5797 steps (SD 3296), and the average Amulet step differences 2 standard deviations higher than the combined count was 9780 steps (SD 3719). For the second cut-off difference mean. These observations had an average Fitbit and threshold (version 2), we recorded 211 participant-days; the Amulet step count of 1373 (SD 1988) and 10,689 (SD 1971), average Fitbit step count was 6415 per day (SD 3751), and the which suggested that participants may have taken their Fitbits average Amulet step count was 7956 per day (SD 3324). The off at some point during the day before removing their Amulet. Figure 5. Association between Amulet and Fitbit steps by algorithm version. Each line is a linear regression for each participant, colored separately, with the overall linear regression in black. Version 1 represents the app used during the first 10 weeks of cohort 1, and version 2 represents the app used during the final 2 weeks of cohort 1 and all 12 weeks of cohort 2. https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 7 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al Figure 6. Bland-Altman plots of Amulet and Fitbit step measures by algorithm version. The blue line represents the mean difference between Amulet and Fitbit steps per day. Each red line represents 2 SD of the difference. Version 1 represents the app used during the first 10 weeks of cohort 1, and version 2 represents the app used during the final 2 weeks of cohort 1 and all 12 weeks of cohort 2. have been a result of the differences between the populations Discussion used for developing and evaluating the algorithm (older adults vs older obese adults). Older adults with obesity have a higher We found that an open-source platform and algorithm developed degree of comorbid conditions [22], along with differences in for older adults can capture daily step counts with reasonable stride length, cadence, and gait [23], which may impact either accuracy. Our findings demonstrate the importance of an algorithm. Additionally, the longer time period (12 weeks vs 2 iterative process in algorithm development before field days) could have allowed the occurrence of a greater number deployment. First, our lab-based data provided confidence in of confounding situations, such as both devices not being worn the algorithm’s step estimates, making a case for a real-world at all times or one device being off while the other was on. evaluation. We then tested the algorithm in a 2-day field study These results make a strong case for developing and refining before full-scale deployment. The step estimates from our algorithms with data from the target population and evaluating algorithm were highly correlated with the step counts from the algorithms in the conditions for which they were designed. Fitbit for all peak cut-off thresholds, with low error rates. These results provide evidence that the algorithm worked well in Based on our observations in the first 10 weeks of cohort 1, we free-living conditions, albeit for a short duration. We used the modified the peak cut-off threshold to minimize the error rate cut-off threshold with the lowest error rate for the subsequent and evaluated version 2 of the algorithm in a study of the same field studies conducted over longer periods. duration with different participants from the same target population (cohort 2). Version 2 exhibited better performance As with any user study, there are differences between field-based in terms of correlations and error rates. conditions and those in laboratory settings or short-duration studies. The poorer results in the 12-week study (cohort 1) may https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 8 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al One limitation of this work is that we used the Fitbit as ground Because we were evaluating predominantly intraperson truth. Given that the Fitbit device (and its proprietary algorithm) variability (ie, the purpose was not to evaluate the impact of was not specifically developed for this population (older adults any intervention), we did not measure baseline characteristics with obesity), it is difficult to say whether our algorithm of the participants (eg, disease regarding walking behavior, such performed better or worse than Fitbit’s algorithm relative to the as Parkinson disease, musculoskeletal disorders). Future studies ground truth. The ideal ground truth would have been to use should determine whether such characteristics could have an video recording, as we did in the lab study, but videorecording impact on our results. is not feasible for field studies due to privacy limitations. Hence, The use of an open-source system, such as the Amulet, a device such as the Fitbit was the best compromise. highlights researchers’ ability to develop algorithms that are Although Fitbit outputs have been shown to have high tailored and trained for a target population such as older adults. correlations with those of other step-counting devices when With the constant iteration of consumer devices and algorithms, used with older adults in free-living settings, results have varied it is difficult to ensure precision and accuracy for groups that depending on the specific Fitbit device used, device placement need to be more physically active, such as older adults. Hence, (wrist vs waist), and the comparison device used [24-27]. We it is important to develop and examine products that can meet hypothesized that the Fitbit would underestimate the steps of their needs. Providing imprecise or inaccurate information on older adults in comparison to the Amulet, because the Fitbit physical activity could undermine the motivation of this was developed using data from younger adults, and older adults population to increase their physical activity. We recommend move more slowly [28]. Thus, we expect the overall true step that future work demonstrate validity of algorithms in these count to be higher than Fitbit’s estimate. Hence, we settled for populations and identify situations where data collection can the case where our algorithm overestimated the steps but was be the most clinically relevant and actionable. highly correlated with Fitbit’s estimate. In addition, the Fitbit Our step-count algorithm performed well in comparison with data were captured daily, whereas Amulet data were captured Fitbit, with high correlations and low error rates. Overall, this hourly. Had hourly data been available from both, we could work highlights various challenges and insights involved in have performed a fine-grained comparison between the developing and validating monitoring systems in real-world algorithms. Finally, it was not possible to get a good sense of settings. wear time from the Fitbit as we did in the Amulet. We could only use the Amulet’s wear time estimate and a minimum Fitbit daily step count of 100 steps as the threshold for being worn. Acknowledgments This work was supported in part by the National Institute on Aging and Office of Dietary Supplements of the National Institutes of Health (K23AG051681 awarded to JAB). Support was also provided by the Dartmouth Health Promotion and Disease Prevention Research Center (U48DP005018) from the Centers for Disease Control and Prevention, and the Dartmouth Clinical and Translational Science Institute (UL1TR001086) from the National Center for Advancing Translational Sciences of the National Institutes of Health. CLP was supported by the Burroughs-Wellcome Fund: Big Data in the Life Sciences at Dartmouth. DK and GB were supported by National Science Foundation awards (CNS-1314281 and CNS-1619970) and by the National Institutes of Health/National Institute on Drug Abuse (P30DA029926). KLF was supported by the National Institute of Mental Health (K01MH117496). The content is solely the responsibility of the authors and does not necessarily represent the official views of, or represent the official position of, any of the sponsors. Conflicts of Interest KLF provides consulting through Social Wellness. References 1. Rubenstein LZ, Powers CM, MacLean CH. Quality indicators for the management and prevention of falls and mobility problems in vulnerable elders. Ann Intern Med 2001 Oct 16;135(8 Pt 2):686-693 [FREE Full text] [doi: 10.7326/0003-4819-135-8_part_2-200110161-00007] [Medline: 11601951] 2. Tinetti ME, Williams CS. Falls, injuries due to falls, and the risk of admission to a nursing home. N Engl J Med 1997 Oct 30;337(18):1279-1284. [doi: 10.1056/NEJM199710303371806] [Medline: 9345078] 3. Fisher KJ, Li F. A community-based walking trial to improve neighborhood quality of life in older adults: a multilevel analysis. Ann Behav Med 2004 Dec;28(3):186-194. [doi: 10.1207/s15324796abm2803_7] [Medline: 15576257] 4. Paterson DH, Warburton DE. Physical activity and functional limitations in older adults: a systematic review related to Canada's Physical Activity Guidelines. Int J Behav Nutr Phys Act 2010 May 11;7:38 [FREE Full text] [doi: 10.1186/1479-5868-7-38] [Medline: 20459782] 5. Murphy M, Nevill A, Neville C, Biddle S, Hardman A. Accumulating brisk walking for fitness, cardiovascular risk, and psychological health. Med Sci Sports Exerc 2002 Sep;34(9):1468-1474. [doi: 10.1097/00005768-200209000-00011] [Medline: 12218740] https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 9 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al 6. Greenwood-Hickman MA, Rosenberg DE, Phelan EA, Fitzpatrick AL. Participation in older adult physical activity programs and risk for falls requiring medical care, Washington state, 2005-2011. Prev Chronic Dis 2015 Jun 11;12:E90 [FREE Full text] [doi: 10.5888/pcd12.140574] [Medline: 26068411] 7. So C, Pierluissi E. Attitudes and expectations regarding exercise in the hospital of hospitalized older adults: a qualitative study. J Am Geriatr Soc 2012 Apr;60(4):713-718. [doi: 10.1111/j.1532-5415.2012.03900.x] [Medline: 22429118] 8. Tucker JM, Welk GJ, Beyler NK. Physical activity in U.S.: adults compliance with the Physical Activity Guidelines for Americans. Am J Prev Med 2011 Apr;40(4):454-461. [doi: 10.1016/j.amepre.2010.12.016] [Medline: 21406280] 9. Social media use in 2021. Pew Research Center. URL: https://www.pewresearch.org/internet/2021/04/07/ social-media-use-in-2021/ [accessed 2021-09-01] 10. Perrin A, Anderson M. Tech adoption climbs among older adults. Pew Research Center. 2017. URL: https://www. pewresearch.org/internet/2017/05/17/tech-adoption-climbs-among-older-adults/ [accessed 2021-09-01] 11. Singh AK, Farmer C, Van Den Berg ML, Killington M, Barr CJ. Accuracy of the FitBit at walking speeds and cadences relevant to clinical rehabilitation populations. Disabil Health J 2016 Apr;9(2):320-323. [doi: 10.1016/j.dhjo.2015.10.011] [Medline: 26905972] 12. Feehan LM, Geldman J, Sayre EC, Park C, Ezzat AM, Yoo JY, et al. Accuracy of Fitbit devices: systematic review and narrative syntheses of quantitative data. JMIR Mhealth Uhealth 2018 Aug 09;6(8):e10527 [FREE Full text] [doi: 10.2196/10527] [Medline: 30093371] 13. Cho Y, Cho H, Kyung C. Design and implementation of practical step detection algorithm for wrist-worn devices. IEEE Sensors J 2016 Aug 25:1. [doi: 10.1109/jsen.2016.2603163] 14. Genovese V, Mannini A, Sabatini AM. A smartwatch step counter for slow and intermittent ambulation. IEEE Access 2017;5:13028-13037. [doi: 10.1109/access.2017.2702066] 15. Matthies D, Haescher M, Nanayakkara S, Bieber G. Step detection for Rollator users with smartwatches. 2018 Presented at: Symposium on Spatial User Interaction; October 13, 2018; Berlin, Germany p. 163-167. [doi: 10.1145/3267782.3267784] 16. Boateng G, Batsis JA, Proctor P, Halter R, Kotz D. GeriActive: wearable app for monitoring and encouraging physical activity among older adults. 2018 Presented at: 15th International Conference on Wearable and Implantable Body Sensor Networks; March 3, 2018; Las Vegas, Nevada p. 46-49. [doi: 10.1109/BSN.2018.8329655] 17. Hester J, Peters T, Yun T, Peterson R, Skinner J, Golla B. Amulet: an energy-efficient, multi-application wearable platform. 2016 Presented at: ACM Conference on Embedded Network Sensor Systems; November 14, 2016; Stanford, California p. 216-229. 18. Boateng G, Motti V, Mishra V, Batsis J, Hester J, Kotz D. Experience: design, development and evaluation of a wearable device for mhealth applications. 2019 Presented at: 25th Annual International Conference on Mobile Computing and Networking; October 11,2019; Los Cabos, Mexico URL: https://doi.org/10.1145/3300061.3345432 [doi: 10.1145/3300061.3345432] 19. Boateng G, Batsis JA, Halter R, Kotz D. ActivityAware: an app for real-time daily activity level monitoring on the Amulet wrist-worn device. 2017 Presented at: International Conference on Pervasive Computing and Communications Workshops; March 2017; Hawaii. [doi: 10.1109/PERCOMW.2017.7917601] 20. Batsis JA, Petersen CL, Clark MM, Cook SB, Lopez-Jimenez F, Al-Nimr RI, et al. A weight loss intervention augmented by a wearable device in rural older adults with obesity: a feasibility study. J Gerontol A Biol Sci Med Sci 2021 Jan 01;76(1):95-100 [FREE Full text] [doi: 10.1093/gerona/glaa115] [Medline: 32384144] 21. Tudor-Locke C, Hart TL, Washington TL. Expected values for pedometer-determined physical activity in older populations. Int J Behav Nutr Phys Act 2009 Aug 25;6:59 [FREE Full text] [doi: 10.1186/1479-5868-6-59] [Medline: 19706192] 22. Lynch DH, Petersen CL, Fanous MM, Spangler HB, Kahkoska AR, Jimenez D, et al. The relationship between multimorbidity, obesity and functional impairment in older adults. J Am Geriatr Soc 2022 May;70(5):1442-1449. [doi: 10.1111/jgs.17683] [Medline: 35113453] 23. Ko SU, Stenholm S, Ferrucci L. Characteristic gait patterns in older adults with obesity--results from the Baltimore Longitudinal Study of Aging. J Biomech 2010 Apr 19;43(6):1104-1110 [FREE Full text] [doi: 10.1016/j.jbiomech.2009.12.004] [Medline: 20080238] 24. McVeigh J, Ellis J, Ross C, Tang K, Wan P, Halse RE. Convergent validity of the Fitbit Charge 2 to measure sedentary behavior and physical activity in overweight and obese adults. J Meas Phys Behav 2021 Jan;4(1):39-46. [doi: 10.1123/jmpb.2020-0014] 25. Farina N, Lowry RG. The validity of consumer-level activity monitors in healthy older adults in free-living conditions. J Aging Phys Act 2018 Jan 01;26(1):128-135. [doi: 10.1123/japa.2016-0344] [Medline: 28595019] 26. Collins JE, Yang HY, Trentadue TP, Gong Y, Losina E. Validation of the Fitbit Charge 2 compared to the ActiGraph GT3X+ in older adults with knee osteoarthritis in free-living conditions. PLoS One 2019 Jan 30;14(1):e0211231 [FREE Full text] [doi: 10.1371/journal.pone.0211231] [Medline: 30699159] 27. Paul SS, Tiedemann A, Hassett LM, Ramsay E, Kirkham C, Chagpar S, et al. Validity of the Fitbit activity tracker for measuring steps in community-dwelling older adults. BMJ Open Sport Exerc Med 2015:e000013 [FREE Full text] [doi: 10.1136/bmjsem-2015-000013] [Medline: 27900119] https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 10 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al 28. Mendes J, Borges N, Santos A, Padrão P, Moreira P, Afonso C, et al. Nutritional status and gait speed in a nationwide population-based sample of older adults. Sci Rep 2018 Mar 09;8(1):4227 [FREE Full text] [doi: 10.1038/s41598-018-22584-3] [Medline: 29523852] Abbreviations API: application programming interface Edited by J Wang; submitted 26.09.21; peer-reviewed by U Bork, S Rostam Niakan Kalhori; comments to author 28.10.21; revised version received 13.12.21; accepted 07.02.22; published 10.08.22 Please cite as: Boateng G, Petersen CL, Kotz D, Fortuna KL, Masutani R, Batsis JA JMIR Aging 2022;5(3):e33845 URL: https://aging.jmir.org/2022/3/e33845 doi: 10.2196/33845 PMID: ©George Boateng, Curtis L Petersen, David Kotz, Karen L Fortuna, Rebecca Masutani, John A Batsis. Originally published in JMIR Aging (https://aging.jmir.org), 10.08.2022. This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Aging, is properly cited. The complete bibliographic information, a link to the original publication on https://aging.jmir.org, as well as this copyright and license information must be included. https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 11 (page number not for citation purposes) XSL FO RenderX http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png JMIR Aging JMIR Publications

A Smartwatch Step-Counting App for Older Adults: Development and Evaluation Study

Download PDF
11 pages

Loading next page...
 
/lp/jmir-publications/a-smartwatch-step-counting-app-for-older-adults-development-and-5xuMcKn56g
Publisher
JMIR Publications
Copyright
Copyright © The Author(s). Licensed under Creative Commons Attribution cc-by 4.0
ISSN
2561-7605
DOI
10.2196/33845
Publisher site
See Article on Publisher Site

Abstract

Background: Older adults who engage in physical activity can reduce their risk of mobility impairment and disability. Short amounts of walking can improve quality of life, physical function, and cardiovascular health. Various programs have been implemented to encourage older adults to engage in physical activity, but sustaining their motivation continues to be a challenge. Ubiquitous devices, such as mobile phones and smartwatches, coupled with machine-learning algorithms, can potentially encourage older adults to be more physically active. Current algorithms that are deployed in consumer devices (eg, Fitbit) are proprietary, often are not tailored to the movements of older adults, and have been shown to be inaccurate in clinical settings. Step-counting algorithms have been developed for smartwatches, but only using data from younger adults and, often, were only validated in controlled laboratory settings. Objective: We sought to develop and validate a smartwatch step-counting app for older adults and evaluate the algorithm in free-living settings over a long period of time. Methods: We developed and evaluated a step-counting app for older adults on an open-source wrist-worn device (Amulet). The app includes algorithms to infer the level of physical activity and to count steps. We validated the step-counting algorithm in the lab (counting steps from a video recording, n=20) and in free-living conditions—one 2-day field study (n=6) and two 12-week field studies (using the Fitbit as ground truth, n=16). During app system development, we evaluated 4 walking patterns: normal, fast, up and down a staircase, and intermittent speed. For the field studies, we evaluated 5 different cut-off values for the algorithm, using correlation and error rate as the evaluation metrics. Results: The step-counting algorithm performed well. In the lab study, for normal walking (R =0.5), there was a stronger correlation between the Amulet steps and the video-validated steps; for all activities, the Amulet’s count was on average 3.2 (2.1%) steps lower (SD 25.9) than the video-validated count. For the 2-day field study, the best parameter settings led to an association between Amulet and Fitbit (R =0.989) and 3.1% (SD 25.1) steps lower than Fitbit, respectively. For the 12-week field study, the best parameter setting led to an R value of 0.669. Conclusions: Our findings demonstrate the importance of an iterative process in algorithm development before field-based deployment. This work highlights various challenges and insights involved in developing and validating monitoring systems in https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 1 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al real-world settings. Nonetheless, our step-counting app for older adults had good performance relative to the ground truth (a commercial Fitbit step counter). Our app could potentially be used to help improve physical activity among older adults. (JMIR Aging 2022;5(3):e33845) doi: 10.2196/33845 KEYWORDS step tracking; step counting; pedometer; wearable; smartwatch; older adults; physical activity; machine learning; walking; mHealth; mobile health; mobile app; mobile application; app; uHealth and have been shown to have wide error rates in such contexts Introduction [11,12]. A few smartwatch-based step-counting algorithms have been developed using data from young adults and validated in Older adults are faced with an increased risk of developing controlled settings only [13,14]. Matthies et al [15] developed multiple comorbid medical conditions, social isolation, and a smartwatch step-counting app for older adults who use a reduced physical function, which can lead to an increased risk walking frame equipped with wheels, which was evaluated of disability [1]. An inability to engage in activities of daily outdoors, but only in a controlled setting, with 5 older adults. living may increase mortality risk and premature nursing home To the best of our knowledge, a smartwatch step-counting app placement [2]. Hence, it is critical to encourage older adults for older adults has not been developed and validated in with multimorbidity to engage in interventions that improve free-living settings over a long period of time with a large health, including physical activity. In fact, simple sample. community-based walking programs and resistance-based programs [3] have effectively demonstrated reductions in the We previously developed GeriActive, an app that measures the long-term risk of disability [4]. Even short bouts of walking can daily activity levels (low, moderate, or vigorous) of older adults improve quality of life, physical function, and cardiovascular [16]. We aimed to develop and validate a smartwatch-based fitness in older adults [5]. step-counting algorithm for older adults that runs as an app on the Amulet device. The Amulet is an open-source wrist-worn Traditional consumer-based health-promoting endeavors, such device that has been used for various mobile health studies, as Silver Sneakers [6], have been scaled and widely such as stress and physical activity monitoring [17,18]. The disseminated across the United States. Randomized control Amulet platform enables developers to write energy- and trials have also shown the short- and long-term benefits of memory-efficient apps. physical activity. However, sustained motivation continues to be a challenge for many individuals. Simple encouragement by Methods clinicians may enhance engagement [7]. Yet, a study of accelerometry data demonstrated that only 7.6% of older adults Study Overview aged 65 years meet Physical Activity Guidelines for Americans We validated the step-counting algorithm with older adults in [8]. These pragmatic challenges demonstrate the need to the lab (using videorecordings as ground truth) and in 2 overcome the barriers of traditional health promotion to enhance free-living studies (using the Fitbit as ground truth) lasting 2 self-efficacy and behavioral change. and 12 days. Older adults are the fastest-growing group of technology users; Overview of Step Counter App one survey suggested that 61% of older adults use smartphones [9], an increase from 23% in 2013 [10]. In fact, over 50% of Our step-counting Amulet app estimates the number of steps older adults use social media [9]. Remote monitoring using taken over the course of a day and displays the information on fitness devices has now become ubiquitous in many countries the Amulet screen, similar to the functionality of a pedometer where technology is readily available. In both consumer-based or of other wearable devices (Figure 1). The app continuously and academic-focused trials, it continues to be challenging to estimates the number of steps for each 5-second window, find a balance between clinical accuracy and ease of use. Current updating the count viewed by the user. It uses a 2-step process: algorithms in consumer devices (such as Fitbit) are proprietary machine learning is used to determine if physical activity and often are not tailored to the group being evaluated in a occurred in the most recent 5 seconds of data, and then, the clinical setting, such as older adults in free-living conditions, number of steps is estimated by counting the number of peaks. https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 2 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al Figure 1. The Step Counter App with the step count displayed at the bottom. Dartmouth-Hitchcock Institutional Review Board (28905). All Activity-Detection Algorithm participants provided signed informed consent. We used a linear support vector machine that classifies each Participants 5-second window of accelerometer data into low, moderate, or vigorous activity [16]. We trained the algorithm on data Participants were recruited through the Center for Health and collected from older adults who performed various physical Aging at Dartmouth and primary care practices at activities: sitting, standing, lying down, walking, and running Dartmouth-Hitchcock using flyers, listservs, and posters. This [16,19]. Our evaluation of the algorithm produced an accuracy was a convenience sample; our results may not necessarily be of 91.7% using leave-one-subject-out cross-validation. If the applicable to other groups. output of the algorithm is moderate or vigorous, the Laboratory Study step-detection algorithm is run to determine the number of steps Data for the development and evaluation of the step-detection in the 5-second window. This 2-phase approach was necessary algorithm were collected at our Dartmouth campus laboratory. to reduce false positives by preventing various actions such as Older adults (n=20, age ≥65 years) were asked to perform random hand motions from being counted as steps. different types of walking (normal, fast, up and down a staircase, Step-Detection Algorithm and intermittent) while wearing an Amulet. The Amulet The step-detection algorithm estimates the number of steps in collected raw acceleration data at a frequency of 20 Hz and 5 seconds of acceleration data. The algorithm uses the magnitude logged the magnitude, which we later used to develop the step of the acceleration of the 3-axis accelerometer of the Amulet. algorithm. The participants were videotaped. The video was It is a 3-stage process consisting of zero-meaning, filtering, and independently reviewed to count steps by 2 individuals peak detection, using an approach similar to that described in independently (JAB, RKM) and any differences were later [13]. First, to ensure the data have a mean of zero, for each reconciled. We used these data for the preliminary development sample, we subtract the average of the preceding 20 data of the step-detection algorithm and evaluated the algorithm samples. Subsequently, a moving-average filter is applied, that using the error rate (the percentage difference between the is, each sample is replaced with the average of the 3 preceding algorithm’s estimated step count and the ground-truth step count samples. Finally, peaks in the filtered signal are identified by measured from the video). checking for change of slope. If the slope changes from positive 2-Day Field Study to negative, and the peak value is above a certain threshold, We conducted a 2-day field study in which older adults (n=7, then the peak is counted as a step. The cut-off threshold was age ≥65 years) wore an Amulet device (running our step counter initially empirically determined and then tuned. app) and a Fitbit Flex 2 device (Fitbit Inc) on the same wrist Ethics for 2 days. We compared each participant’s step count estimated Studies were approved by the Committee for the Protection of by their Amulet (exploring 5 different cut-off values) with their Human Subjects at Dartmouth College and the step count reported by the Fitbit (downloaded from the https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 3 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al individual’s research-based Fitbit account). The error rate Participants (Table 1) from both rounds wore an Amulet and between Fitbit’s steps and Amulet’s steps for each of the 5 peak Fitbit on the same wrist for 12 weeks. The Amulet logged the cut-off values was computed. summary steps on an SD memory card hourly and at the end of each day. A research assistant copied the data from each 12-Week Field Study participant’s SD card on a biweekly basis. The Fitbit logged a We conducted a field study (2 cohorts, 12 weeks each) to test summary of each day’s step count (computed by a proprietary the step-detection algorithm with the target population—older algorithm) to the Fitbit app on the participant’s Android tablet, adults with obesity. This study was a subset of a larger study which uploaded the data to the Fitbit servers; we subsequently that evaluated the impact of a combined weight loss and exercise used the Fitbit research API to download participants’ data. intervention in older adults with obesity [20]. The goal of this After 10 weeks of monitoring data from the first cohort, we analysis was to compare the Amulet’s step-count estimate with modified the step-detection peak cut-off value to minimize the the Fitbit’s step count over a long period in real-world settings. error rate relative to ground-truth step count data from the Fitbit. Table 1. Participant characteristics. Characteristic Value (n=16) Age (years) Mean (SD) 74.1 (5.6) Range 66-87 Sex, n (%) Male 4 (25) Female 12 (75) Marital status, n (%) Married 7 (44) Divorced 8 (50) Widowed 1 (6) Smoking history, n (%) None 13 (81) Formerly smoked 3 (19) Education, n (%) High school 2 (12) Some college 5 (31) College degree 3 (19) Postcollege degree 6 (38) Weight (kg), mean (SD) 97.06 (18.2) 36.8 (4.9) Body mass index (kg/m ), mean (SD) Multimorbidity, n (%) 14 (87) Comorbidities, n (%) Diabetes 2 (12) Fibromyalgia 1 (6) High cholesterol 6 (38) Hypertension 9 (56) Osteoarthritis 6 (38) We thus limited the data to days for which it was safe to assume Statistical Analysis that they were worn for nearly the same amount of time. The We calculated the mean with standard deviation for continuous Amulet was able to detect if it was worn each hour, so we measures and count with percentage for categorical measures. considered the Amulet to be worn if the Amulet was worn for Although participants were instructed to wear the Amulet and 75% of a 15-hour day (675 minutes); the Fitbit only reported Fitbit simultaneously, not all participants did so the entire time. daily step count, so we considered the Fitbit to be worn if the https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 4 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al step count was greater than 100. We selected these parameters between Amulet and Fitbit steps. For our analysis, we defined from an understanding of the distribution of wear time over the significance as P<.05. course of a 24-hour day and the distribution of steps expected for this population per day [21]. Comparisons between Amulet Results and Fitbit were limited to data from days for which both were Laboratory Study worn. We conducted bivariate linear regression to compare the association between Amulet and video-counted steps during There was a strong linear association when participants walked various activities (laboratory study) and between Amulet and normally (Figure 2; Table 2). For normal walking, the Amulet Fitbit steps (2-day and 12-week field studies). We compared step-detection algorithm undermeasured the number of steps Amulet steps to Fitbit steps using percentage difference taken by an average of 6.7 steps (SD 32.6). Across all activities, (difference between Amulet and Fitbit steps divided by Fitbit the Amulet was on average 3.2 steps lower (SD 25.9) or 2.1% steps).We used Bland-Altman plots to compare the agreement (SD 31.9%) lower than video-validated steps (Figure 2). The distribution was slightly positively skewed. Figure 2. Association between Amulet-estimated steps and video-validated steps. Table 2. Step count for different walking activities. Activity Video-validated, n (%) Amulet, n (%) Percentage error Fast 102.62 (14.7) 110.72 (22.7) 8.53 (20.02) Intermittent 101.02 (18.7) 99.48 (30.8) 1.71 (33.28) Normal 84.9 (31.9) 77.14 (33.9) –6.68 (32.55) Stairs 52.76 (13.4) 45.52 (13.2) –7.55 (36.97) https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 5 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al the highest association between Amulet and Fitbit steps 2-Day Field Study (R =0.989). Cut-off threshold number 2 had the smallest mean We discarded 1 participant’s data because the data indicated percentage difference between Amulet and Fitbit (–3.1%, SD the devices had not been worn much (step counts were less than 25.1) of all cut-off thresholds (threshold 1: mean 15.27%, SD 350 per day). The associations between each participant’s daily 33.19; threshold 3: mean –10.77, SD 23.43; threshold 4: mean step counts reported by the Amulet and Fitbit were high for all 7.18, SD 29.28; threshold 5: mean –4.11, SD 24.94) (Figure 4). cut-off thresholds (Figure 3). The third cut-off threshold had Figure 3. Association between Amulet and Fitbit steps for different cut-off values: 2-day field study. Figure 4. Distribution of percentage difference between Amulet and Fitbit steps by cut-off threshold. https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 6 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al association between Fitbit steps and Amulet steps improved 12-Week Field Study Results (first cut-off threshold: R =0.386; second cut-off threshold: Across both 12-day field studies, there were 297 participant-days for which both the Fitbit and Amulet had been worn. Cohort 2 R =0.669) (Figure 5). There was improved agreement between used the modified app. For the first cut-off threshold (version both measures with the second cut-off threshold (Figure 6). 1), we recorded 86 participant-days; the average Fitbit step There were 9 observations by 5 unique participants with count was 5797 steps (SD 3296), and the average Amulet step differences 2 standard deviations higher than the combined count was 9780 steps (SD 3719). For the second cut-off difference mean. These observations had an average Fitbit and threshold (version 2), we recorded 211 participant-days; the Amulet step count of 1373 (SD 1988) and 10,689 (SD 1971), average Fitbit step count was 6415 per day (SD 3751), and the which suggested that participants may have taken their Fitbits average Amulet step count was 7956 per day (SD 3324). The off at some point during the day before removing their Amulet. Figure 5. Association between Amulet and Fitbit steps by algorithm version. Each line is a linear regression for each participant, colored separately, with the overall linear regression in black. Version 1 represents the app used during the first 10 weeks of cohort 1, and version 2 represents the app used during the final 2 weeks of cohort 1 and all 12 weeks of cohort 2. https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 7 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al Figure 6. Bland-Altman plots of Amulet and Fitbit step measures by algorithm version. The blue line represents the mean difference between Amulet and Fitbit steps per day. Each red line represents 2 SD of the difference. Version 1 represents the app used during the first 10 weeks of cohort 1, and version 2 represents the app used during the final 2 weeks of cohort 1 and all 12 weeks of cohort 2. have been a result of the differences between the populations Discussion used for developing and evaluating the algorithm (older adults vs older obese adults). Older adults with obesity have a higher We found that an open-source platform and algorithm developed degree of comorbid conditions [22], along with differences in for older adults can capture daily step counts with reasonable stride length, cadence, and gait [23], which may impact either accuracy. Our findings demonstrate the importance of an algorithm. Additionally, the longer time period (12 weeks vs 2 iterative process in algorithm development before field days) could have allowed the occurrence of a greater number deployment. First, our lab-based data provided confidence in of confounding situations, such as both devices not being worn the algorithm’s step estimates, making a case for a real-world at all times or one device being off while the other was on. evaluation. We then tested the algorithm in a 2-day field study These results make a strong case for developing and refining before full-scale deployment. The step estimates from our algorithms with data from the target population and evaluating algorithm were highly correlated with the step counts from the algorithms in the conditions for which they were designed. Fitbit for all peak cut-off thresholds, with low error rates. These results provide evidence that the algorithm worked well in Based on our observations in the first 10 weeks of cohort 1, we free-living conditions, albeit for a short duration. We used the modified the peak cut-off threshold to minimize the error rate cut-off threshold with the lowest error rate for the subsequent and evaluated version 2 of the algorithm in a study of the same field studies conducted over longer periods. duration with different participants from the same target population (cohort 2). Version 2 exhibited better performance As with any user study, there are differences between field-based in terms of correlations and error rates. conditions and those in laboratory settings or short-duration studies. The poorer results in the 12-week study (cohort 1) may https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 8 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al One limitation of this work is that we used the Fitbit as ground Because we were evaluating predominantly intraperson truth. Given that the Fitbit device (and its proprietary algorithm) variability (ie, the purpose was not to evaluate the impact of was not specifically developed for this population (older adults any intervention), we did not measure baseline characteristics with obesity), it is difficult to say whether our algorithm of the participants (eg, disease regarding walking behavior, such performed better or worse than Fitbit’s algorithm relative to the as Parkinson disease, musculoskeletal disorders). Future studies ground truth. The ideal ground truth would have been to use should determine whether such characteristics could have an video recording, as we did in the lab study, but videorecording impact on our results. is not feasible for field studies due to privacy limitations. Hence, The use of an open-source system, such as the Amulet, a device such as the Fitbit was the best compromise. highlights researchers’ ability to develop algorithms that are Although Fitbit outputs have been shown to have high tailored and trained for a target population such as older adults. correlations with those of other step-counting devices when With the constant iteration of consumer devices and algorithms, used with older adults in free-living settings, results have varied it is difficult to ensure precision and accuracy for groups that depending on the specific Fitbit device used, device placement need to be more physically active, such as older adults. Hence, (wrist vs waist), and the comparison device used [24-27]. We it is important to develop and examine products that can meet hypothesized that the Fitbit would underestimate the steps of their needs. Providing imprecise or inaccurate information on older adults in comparison to the Amulet, because the Fitbit physical activity could undermine the motivation of this was developed using data from younger adults, and older adults population to increase their physical activity. We recommend move more slowly [28]. Thus, we expect the overall true step that future work demonstrate validity of algorithms in these count to be higher than Fitbit’s estimate. Hence, we settled for populations and identify situations where data collection can the case where our algorithm overestimated the steps but was be the most clinically relevant and actionable. highly correlated with Fitbit’s estimate. In addition, the Fitbit Our step-count algorithm performed well in comparison with data were captured daily, whereas Amulet data were captured Fitbit, with high correlations and low error rates. Overall, this hourly. Had hourly data been available from both, we could work highlights various challenges and insights involved in have performed a fine-grained comparison between the developing and validating monitoring systems in real-world algorithms. Finally, it was not possible to get a good sense of settings. wear time from the Fitbit as we did in the Amulet. We could only use the Amulet’s wear time estimate and a minimum Fitbit daily step count of 100 steps as the threshold for being worn. Acknowledgments This work was supported in part by the National Institute on Aging and Office of Dietary Supplements of the National Institutes of Health (K23AG051681 awarded to JAB). Support was also provided by the Dartmouth Health Promotion and Disease Prevention Research Center (U48DP005018) from the Centers for Disease Control and Prevention, and the Dartmouth Clinical and Translational Science Institute (UL1TR001086) from the National Center for Advancing Translational Sciences of the National Institutes of Health. CLP was supported by the Burroughs-Wellcome Fund: Big Data in the Life Sciences at Dartmouth. DK and GB were supported by National Science Foundation awards (CNS-1314281 and CNS-1619970) and by the National Institutes of Health/National Institute on Drug Abuse (P30DA029926). KLF was supported by the National Institute of Mental Health (K01MH117496). The content is solely the responsibility of the authors and does not necessarily represent the official views of, or represent the official position of, any of the sponsors. Conflicts of Interest KLF provides consulting through Social Wellness. References 1. Rubenstein LZ, Powers CM, MacLean CH. Quality indicators for the management and prevention of falls and mobility problems in vulnerable elders. Ann Intern Med 2001 Oct 16;135(8 Pt 2):686-693 [FREE Full text] [doi: 10.7326/0003-4819-135-8_part_2-200110161-00007] [Medline: 11601951] 2. Tinetti ME, Williams CS. Falls, injuries due to falls, and the risk of admission to a nursing home. N Engl J Med 1997 Oct 30;337(18):1279-1284. [doi: 10.1056/NEJM199710303371806] [Medline: 9345078] 3. Fisher KJ, Li F. A community-based walking trial to improve neighborhood quality of life in older adults: a multilevel analysis. Ann Behav Med 2004 Dec;28(3):186-194. [doi: 10.1207/s15324796abm2803_7] [Medline: 15576257] 4. Paterson DH, Warburton DE. Physical activity and functional limitations in older adults: a systematic review related to Canada's Physical Activity Guidelines. Int J Behav Nutr Phys Act 2010 May 11;7:38 [FREE Full text] [doi: 10.1186/1479-5868-7-38] [Medline: 20459782] 5. Murphy M, Nevill A, Neville C, Biddle S, Hardman A. Accumulating brisk walking for fitness, cardiovascular risk, and psychological health. Med Sci Sports Exerc 2002 Sep;34(9):1468-1474. [doi: 10.1097/00005768-200209000-00011] [Medline: 12218740] https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 9 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al 6. Greenwood-Hickman MA, Rosenberg DE, Phelan EA, Fitzpatrick AL. Participation in older adult physical activity programs and risk for falls requiring medical care, Washington state, 2005-2011. Prev Chronic Dis 2015 Jun 11;12:E90 [FREE Full text] [doi: 10.5888/pcd12.140574] [Medline: 26068411] 7. So C, Pierluissi E. Attitudes and expectations regarding exercise in the hospital of hospitalized older adults: a qualitative study. J Am Geriatr Soc 2012 Apr;60(4):713-718. [doi: 10.1111/j.1532-5415.2012.03900.x] [Medline: 22429118] 8. Tucker JM, Welk GJ, Beyler NK. Physical activity in U.S.: adults compliance with the Physical Activity Guidelines for Americans. Am J Prev Med 2011 Apr;40(4):454-461. [doi: 10.1016/j.amepre.2010.12.016] [Medline: 21406280] 9. Social media use in 2021. Pew Research Center. URL: https://www.pewresearch.org/internet/2021/04/07/ social-media-use-in-2021/ [accessed 2021-09-01] 10. Perrin A, Anderson M. Tech adoption climbs among older adults. Pew Research Center. 2017. URL: https://www. pewresearch.org/internet/2017/05/17/tech-adoption-climbs-among-older-adults/ [accessed 2021-09-01] 11. Singh AK, Farmer C, Van Den Berg ML, Killington M, Barr CJ. Accuracy of the FitBit at walking speeds and cadences relevant to clinical rehabilitation populations. Disabil Health J 2016 Apr;9(2):320-323. [doi: 10.1016/j.dhjo.2015.10.011] [Medline: 26905972] 12. Feehan LM, Geldman J, Sayre EC, Park C, Ezzat AM, Yoo JY, et al. Accuracy of Fitbit devices: systematic review and narrative syntheses of quantitative data. JMIR Mhealth Uhealth 2018 Aug 09;6(8):e10527 [FREE Full text] [doi: 10.2196/10527] [Medline: 30093371] 13. Cho Y, Cho H, Kyung C. Design and implementation of practical step detection algorithm for wrist-worn devices. IEEE Sensors J 2016 Aug 25:1. [doi: 10.1109/jsen.2016.2603163] 14. Genovese V, Mannini A, Sabatini AM. A smartwatch step counter for slow and intermittent ambulation. IEEE Access 2017;5:13028-13037. [doi: 10.1109/access.2017.2702066] 15. Matthies D, Haescher M, Nanayakkara S, Bieber G. Step detection for Rollator users with smartwatches. 2018 Presented at: Symposium on Spatial User Interaction; October 13, 2018; Berlin, Germany p. 163-167. [doi: 10.1145/3267782.3267784] 16. Boateng G, Batsis JA, Proctor P, Halter R, Kotz D. GeriActive: wearable app for monitoring and encouraging physical activity among older adults. 2018 Presented at: 15th International Conference on Wearable and Implantable Body Sensor Networks; March 3, 2018; Las Vegas, Nevada p. 46-49. [doi: 10.1109/BSN.2018.8329655] 17. Hester J, Peters T, Yun T, Peterson R, Skinner J, Golla B. Amulet: an energy-efficient, multi-application wearable platform. 2016 Presented at: ACM Conference on Embedded Network Sensor Systems; November 14, 2016; Stanford, California p. 216-229. 18. Boateng G, Motti V, Mishra V, Batsis J, Hester J, Kotz D. Experience: design, development and evaluation of a wearable device for mhealth applications. 2019 Presented at: 25th Annual International Conference on Mobile Computing and Networking; October 11,2019; Los Cabos, Mexico URL: https://doi.org/10.1145/3300061.3345432 [doi: 10.1145/3300061.3345432] 19. Boateng G, Batsis JA, Halter R, Kotz D. ActivityAware: an app for real-time daily activity level monitoring on the Amulet wrist-worn device. 2017 Presented at: International Conference on Pervasive Computing and Communications Workshops; March 2017; Hawaii. [doi: 10.1109/PERCOMW.2017.7917601] 20. Batsis JA, Petersen CL, Clark MM, Cook SB, Lopez-Jimenez F, Al-Nimr RI, et al. A weight loss intervention augmented by a wearable device in rural older adults with obesity: a feasibility study. J Gerontol A Biol Sci Med Sci 2021 Jan 01;76(1):95-100 [FREE Full text] [doi: 10.1093/gerona/glaa115] [Medline: 32384144] 21. Tudor-Locke C, Hart TL, Washington TL. Expected values for pedometer-determined physical activity in older populations. Int J Behav Nutr Phys Act 2009 Aug 25;6:59 [FREE Full text] [doi: 10.1186/1479-5868-6-59] [Medline: 19706192] 22. Lynch DH, Petersen CL, Fanous MM, Spangler HB, Kahkoska AR, Jimenez D, et al. The relationship between multimorbidity, obesity and functional impairment in older adults. J Am Geriatr Soc 2022 May;70(5):1442-1449. [doi: 10.1111/jgs.17683] [Medline: 35113453] 23. Ko SU, Stenholm S, Ferrucci L. Characteristic gait patterns in older adults with obesity--results from the Baltimore Longitudinal Study of Aging. J Biomech 2010 Apr 19;43(6):1104-1110 [FREE Full text] [doi: 10.1016/j.jbiomech.2009.12.004] [Medline: 20080238] 24. McVeigh J, Ellis J, Ross C, Tang K, Wan P, Halse RE. Convergent validity of the Fitbit Charge 2 to measure sedentary behavior and physical activity in overweight and obese adults. J Meas Phys Behav 2021 Jan;4(1):39-46. [doi: 10.1123/jmpb.2020-0014] 25. Farina N, Lowry RG. The validity of consumer-level activity monitors in healthy older adults in free-living conditions. J Aging Phys Act 2018 Jan 01;26(1):128-135. [doi: 10.1123/japa.2016-0344] [Medline: 28595019] 26. Collins JE, Yang HY, Trentadue TP, Gong Y, Losina E. Validation of the Fitbit Charge 2 compared to the ActiGraph GT3X+ in older adults with knee osteoarthritis in free-living conditions. PLoS One 2019 Jan 30;14(1):e0211231 [FREE Full text] [doi: 10.1371/journal.pone.0211231] [Medline: 30699159] 27. Paul SS, Tiedemann A, Hassett LM, Ramsay E, Kirkham C, Chagpar S, et al. Validity of the Fitbit activity tracker for measuring steps in community-dwelling older adults. BMJ Open Sport Exerc Med 2015:e000013 [FREE Full text] [doi: 10.1136/bmjsem-2015-000013] [Medline: 27900119] https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 10 (page number not for citation purposes) XSL FO RenderX JMIR AGING Boateng et al 28. Mendes J, Borges N, Santos A, Padrão P, Moreira P, Afonso C, et al. Nutritional status and gait speed in a nationwide population-based sample of older adults. Sci Rep 2018 Mar 09;8(1):4227 [FREE Full text] [doi: 10.1038/s41598-018-22584-3] [Medline: 29523852] Abbreviations API: application programming interface Edited by J Wang; submitted 26.09.21; peer-reviewed by U Bork, S Rostam Niakan Kalhori; comments to author 28.10.21; revised version received 13.12.21; accepted 07.02.22; published 10.08.22 Please cite as: Boateng G, Petersen CL, Kotz D, Fortuna KL, Masutani R, Batsis JA JMIR Aging 2022;5(3):e33845 URL: https://aging.jmir.org/2022/3/e33845 doi: 10.2196/33845 PMID: ©George Boateng, Curtis L Petersen, David Kotz, Karen L Fortuna, Rebecca Masutani, John A Batsis. Originally published in JMIR Aging (https://aging.jmir.org), 10.08.2022. This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Aging, is properly cited. The complete bibliographic information, a link to the original publication on https://aging.jmir.org, as well as this copyright and license information must be included. https://aging.jmir.org/2022/3/e33845 JMIR Aging 2022 | vol. 5 | iss. 3 | e33845 | p. 11 (page number not for citation purposes) XSL FO RenderX

Journal

JMIR Aging – JMIR Publications

Published: Aug 10, 2022

Keywords: step tracking; step counting; pedometer; wearable; smartwatch; older adults; physical activity; machine learning; walking; mHealth; mobile health; mobile app; mobile application; app; uHealth

There are no references for this article.