The accuracy of personal activity monitoring devices

Author’s Accepted Manuscript The Accuracy of Personal Activity Monitoring Devices Andrew K. Battenberg, Steven Donohoe, Nicholas Robertson, Thomas P. Schmalzried www.elsevier.com/locate/sart

PII: DOI: Reference:

S1045-4527(17)30058-5 http://dx.doi.org/10.1053/j.sart.2017.07.006 YSART50763

To appear in: Seminars in Arthroplasty Cite this article as: Andrew K. Battenberg, Steven Donohoe, Nicholas Robertson and Thomas P. Schmalzried, The Accuracy of Personal Activity Monitoring Devices, Seminars in Arthroplasty, http://dx.doi.org/10.1053/j.sart.2017.07.006 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

The Accuracy of Personal Activity Monitoring Devices Andrew K. Battenberg, M.D.1, Steven Donohoe, M.D.2, Nicholas Robertson, M.D.1, and Thomas P. Schmalzried, M.D.1,3 1

Harbor-UCLA Medical Center Department of Orthopaedic Surgery Torrance, CA, USA 2

LAC+USC Medical Center Department of Orthopaedic Surgery Los Angeles, CA, USA 3

St. Vincent Medical Center Joint Replacement Institute Los Angeles, CA, USA

Corresponding Author: Andrew K. Battenberg, MD Harbor-UCLA Medical Center Department of Orthopaedic Surgery 1000 W Carson St, Box 422 Torrance, CA 90509 [email protected] Phone: 916-835-9205 Steven Donohoe, MD [email protected] Nicholas Robertson, MD [email protected] Thomas P. Schmalzried, MD [email protected]

Abstract Activity monitoring has important applications in orthopedic and medical research. There is a paucity of information on device accuracy. Thirty adults tested the accuracy of 10 devices in 5 activities: walk 400-M, run 400-M, walk 10-M, ascend and descend 10 steps. A second protocol tested slow walking speeds at 1.0, 1.5, 1.8, and 2.0MPH. The waist-based devices FitBit One™ and Omron HJ-321 were >90% accurate for all activities. The wristband devices and Smartphone Apps were <90% accurate for most activities. The StepWatch™ Activity Monitor was >95% accurate at lower cadence activities, but undercounted running by ~25%. Waist-based, dedicated activity monitors are highly accurate in a variety of activities.

Introduction Obesity and physical inactivity are significant public health concerns. Increased physical activity has been shown to decrease risk of cardiovascular disease,1 cancer,2 diabetes,3 stroke,4 and allcause mortality.5,6 Greater awareness of these health benefits has led to a surge in the use of activity monitors. In 2012, there were 156 million Smartphone fitness-tracking application downloads, with projections to 248 million by 2017.7 In 2013, 13 million dedicated activity monitors were sold, with expected growth to 130 million by 2018.8 There is increasing utility of activity tracking in orthopaedic and rehabilitative research. Pedometers and ankle-based accelerometers have been used to quantify arthroplasty patient activity levels pre- and post-operatively,9,10 measure early post-operative rehabilitation,11 and correlate activity to implant wear.12,13 Tracking devices provide objective, continuous data, which complements subjective, categorical measures like the UCLA activity score.14

There are many available tracking devices and Smartphone applications. As of April 2015, Amazon.com, Inc. offered 1,870 distinct pedometers for sale and the Apple App and Google Play Stores combined to offer 856 distinct pedometer applications. Notably, manufacturers do not provide accuracy data. The goal of the current study was to assess the accuracy of several widely used activity trackers in walking, running, and stair-climbing. Methods A convenience sample of thirty healthy adult volunteers was studied. There were no specific exclusion criteria. Informed consent was obtained prior to participation in the study. Participant Characteristics There were 18 men and 12 women with mean age 25.6 years (SD: 2.5, range: 20-30), mean weight 73.4kg (SD: 14, range: 49.9-99.8), mean height 175.9cm (SD: 8.7, range: 154.9-190.5), and mean BMI 23.5 (SD: 2.9, range: 17.3-29.0). Mean waist circumference was 83.4cm (SD: 9.2, range: 67-99), mean hip circumference 102.9cm (SD: 5.7, range: 93-112), mean leg length 93.9cm (SD: 5.3, range: 75-103), and mean arm span 179.9cm (SD: 10.9, range: 149-199). Activity Tracking Devices 1. FitBit One™ (FitBit® Inc., San Francisco, CA, USA): A 3-axis accelerometer and altimeter worn on the waistband. 2. Omron HJ-321 Pedometer (Omron Corporation, Healthcare Division, Kyoto, Japan): A piezoelectric pedometer with 3-axis accelerometer worn on the waistband. 3. Sportline 340 Strider Pedometer (Sportline Corporation, Elmsford, NY, USA): A pendulumdesign pedometer worn on the waistband. 4. The Apple iPhone 5 (Apple Inc., Cupertino, CA, USA): A Smartphone with 3-axis accelerometer. Using the search term “Pedometer,” the two most downloaded Apple Apps were

selected: 1) Argus Motion and Fitness Tracker by Azumio. 2) Runtastic Pedometer Step Counter and Walking Tracker by Runtastic. 5. The Samsung Galaxy S IV (Samsung Group, Samsung Town, Seoul, South Korea): A Smartphone with 3-axis accelerometer. Using the search term “Pedometer,” the two most downloaded Google Play Apps were selected: 1) Noom Walk: Pedometer. 2) Runtastic Pedometer Step Counter and Walking Tracker by Runtastic. 6. FitBit Force™ (FitBit® Inc., San Francisco, CA, USA): A 3-axis accelerometer and altimeter worn on the wrist. 7. Nike+ Fuelband SE (Nike Inc. Washington County, OR, USA): A 3-axis accelerometer worn on the wrist. 8. StepWatch™ Activity Monitor (SAM; Cyma Corp., Seattle, WA, USA): A research-grade 2axis accelerometer worn on the ankle. Protocol 1 The FitBit One™ was worn on the waistband at the right anterior superior iliac spine (ASIS). The Omron HJ-321 and Sportline 340 Pedometer were worn on the waistband at the left ASIS. The Samsung Galaxy S IV and iPhone 5 were placed in holster cases and worn at the right and left ASIS, respectively. The Nike+ FuelBand SE and Fitbit Force™ were worn on the right and left wrist, respectively. The SAM was worn on the right ankle. Devices were worn simultaneously to provide an internal control for each trial. Each subject completed the following protocol 3 times: (1) walk briskly around a 400-M track (2) run 400-M (3) walk 10-M, approximating household pace (4) ascend 10 steps, and (5) descend 10 steps without the use of a railing. This protocol was adapted from Shepard et al. to

include running.15 Manual count was the gold standard (Champion Sports Tally Counter – Champion Sports, Marlboro, NJ, USA). Protocol 2 A subset of 20 consecutive subjects participated in a separate protocol to measure the efficacy of the Fitbit One™, Omron HJ-321, and Sportline 340 Strider at slow walking speeds. These devices were chosen because of superior performance of waist-based devices, as well as for comparison to previous studies of piezoelectric and pendulum-type pedometers showing significant undercounting at slow walking speeds.16,17,18 A treadmill two-minute walk test was performed at four speeds (1.0, 1.5, 1.8, 2.0-MPH) with two trials at each speed. Manual count was the gold standard.

Statistical Analysis All trials were included in data analysis. Percentage error was calculated as 100 X ([device count – actual count]/actual count). Positive values indicate overcounting (extra steps detected), and negative values indicate undercounting (missed steps). Error and accuracy were calculated on a task-by-task basis. Statistical analysis was performed using Excel® (Version 12.0; Microsoft Corp, Redmond, WA, USA). Results Protocol 1 Waist-based Devices FitBit One™ and Omron HJ-321 were ≥95% accurate in the 400-M walk, 400-M run, and 10-M walk, and >91% accurate in the 10 stair ascent and descent (Table 1). Sportline 340 Strider was ≤ 92% accurate for all activities and undercounted on average (Table 1). It performed best in the 400-M run, 400-M walk, and 10 stair descent (mean error: -

5.4%, -7.3%, and -9.8%, respectively) and poorest in the 10-M walk and 10 stair ascent (mean error: -16.9% and -46.6%, respectively) (Figure 1, Table 2). Argus iPhone App was 97.2% accurate in the 400-M walk and 97.3% in the 400-M run, tending to over count (mean error 2.7% and 1.3%, respectively). It undercounted the 10-M walk, 10 stair ascent, and 10 stair descent (mean error -24.0%, -30.2%, and -42.9%, respectively). Runtastic iPhone App was 98.5% accurate in the 400-M walk, but undercounted the 400-M run (mean error -13.6%), and over counted the 10-M walk, 10 stair ascent and descent (mean error 13.8%, 24.0%, 18.4%, respectively). Runtastic Google App was 97.8% accurate in the 400-M walk, and had near 90% accuracy in the other 4 activities. Noom Walk Google App was 80% or less accurate for all activities except the 10 stair descent (91.3% accurate), and, on average, undercounted for all activities. Wrist-based Devices FitBit Force™ and Nike+ Fuelband SE undercounted the 400-M walk (mean error: -7.9% (95% CI: -10.1‒ -5.7) and -10.9% (95% CI -12.3‒ -9.4), respectively), and the 10-M walk (mean error: -11.5% (95% CI: -16.0‒ -7.1) and -35.3% (95% CI: -42.6‒ -28.1), respectively). They performed best on the 400-M run (FitBit Force 92.7% accuracy, Nike+ Fuelband SE 95.6% accuracy). The FitBit Force was less than 85% accurate for both stair ascent and descent, while the Nike+ Fuelband SE was less than 50% accurate for both. Ankle-based Device StepWatch™ Activity Monitor was >95% accurate in the non-running activities, but consistently undercounted the 400-M run: 74.4% accuracy, mean error -25.5% (95% CI: -27.2‒ -23.9).

Combining ascent and descent, it was the most accurate device in counting stairs (96.8% accurate vs 95.6% accurate in the FitBit One™, which was second). Protocol 2 Accuracy was directly related to walking speed (Table 3). Substantial undercounting was observed with all three devices at 1.0MPH. The FitBit One™ was >97% accurate at 1.5MPH, 1.8MPH, and 2.0MPH. The Omron HJ-321 was >95% accurate at 1.8MPH and 2.0MPH. The Sportline 340 Strider consistently undercounted and was 80% or less accurate at all speeds. Discussion To our knowledge, this is the first comparative study of activity monitoring devices to include running, and is the most comprehensive in terms of devices tested (10) and activities measured (5). Six of 10 devices were >95% accurate in the 400-M walk, demonstrating that a majority of devices tested are reliable for dedicated walking programs. Four of 10 were >95% accurate in the 400-M run. FitBit One™, Omron HJ-321, and Argus iPhone App were the only devices >95% accurate in both the 400-M walk and 400-M run. Greater variability was encountered for the other activities: 3 of 10, 3 of 10, and 5 of 10 devices were >90% accurate in the 10-M walk, 10 stair ascent and descent, respectively. The FitBit One™, Omron-HJ-321, and SAM were the only devices >90% accurate in these 3 activities. The StepWatch™ Activity Monitor (93.3%) was >95% accurate in all activities except the 400-M run at 74.4%. Undercounting was likely due to software “lockout,” a feature designed to prevent overcounting by temporarily halting the step count during non-ambulatory activity (e.g. heel tapping). High impact or high cadence activity can be misinterpreted as non-ambulatory activity and trigger a lockout.18 The Runtastic apps were >97% accurate for the 400-M walk, but much less accurate for all other activities. Their highest utility may be in dedicated walking programs. The Argus app was

>97% accurate in both the 400-M walk and run, but performed poorly in 10-M and stairs, making it most suitable for walking and running programs. The cause of poorer accuracy of Smartphone applications is unclear. Possibilities include hardware limitations, software limitations given the variable accuracy of different applications utilizing the same hardware, and the large device size versus case rigidity, which compromises sensor ability to detect heel strike. The wrist-based FitBit Force and Nike+ Fuelband SE were <95% accurate for all activities. Limitations of this study include the relatively homogeneous study population of young, nonobese, active individuals. This population allowed us to study running, a previously untested parameter in activity monitoring. Obesity has been shown to decrease pedometer accuracy.18,19 In this study of non-obese patients, error was not correlated to BMI. Ichinoseki-Sekine et al. studied elderly patients with gait disturbances with average walking speeds 1.23 – 1.64 MPH using the Omron HJ-720, finding mean error of -53.2%.17 This error is within the range observed in this study for the Omron HJ-321 at 1.0MPH and 1.5MPH, with mean error -82.6% and -13.3%, respectively. The FitBit One™ was the most accurate device tested in Protocol 2, with 97.6% accuracy at 1.5MPH. Accuracy decreased to 78.4% at 1.0MPH. The StepWatch™ Activity Monitor has >99% accuracy at 1.0 MPH; this ankle-based device remains the gold standard for slow speed and altered gait patterns.16,18 Another limitation is the sampling of available devices. This study aimed to test a representative portion of fitness tracking devices, with FitBit and Nike accounting for 78% of Smartphone-enabled tracker sales in 2013, and Apple and Samsung representing 68% of Smartphones owned in the US.20,21 In conclusion, waist-based devices FitBit One™ and Omron HJ-321 pedometer are highly accurate for quantifying a variety of activities, including running. Wrist-based devices are less

than 90% accurate for a majority of the tested activities. The ankle-based StepWatch™ Activity Monitor performs well in lower cadence, non-running activities, but consistently undercounts running. Smartphones are very popular, and some apps may be suitable for dedicated walking or running programs, but the tested Smartphone technology is inconsistent in measuring shortduration activities. REFERENCES 1. Tanasescu M, Leitzmann MF, Rimm EB, et al. Exercise Type and Intensity in relation to CHD in Men. JAMA 288 (16):1994-2000, 2003 2.

Arem H, Moore SC, Park Y, et al. Physical activity and cancer-specific mortality in the

NIH-AARP Diet and Health Study cohort. Int J Cancer 135(2):423-431, 2014 3.

Helmrich SP, Ragland DR, Leung RF, et al. Physical Activity and reduced occurrence of

non-insulin-dependent diabetes mellitus. N Engl J Med 325(3):147-152, 1991 4.

Lee CD, Folsom AR, Blair SN. Physical Activity and Stroke Risk: A Meta-analysis.

Stroke 34(10): 2475-81, 2003 5.

Hu FB, Willet WC, Li T, et al. Adiposity as Compared with Physical Activity in

predicting mortility among women. N Engl J Med 351(26):2694-2703, 2004 6.

Paffenbarger RS Jr, Hyde RT, Wing AL, et al. The Association of changes in physical

activity level and other lifestyle characteristics with mortality among men. N Engl J Med 328(8):538-45, 1993 7.

Information Handling Services Web site [Internet]. Douglas County (CO): Information

Handling Services (IHS), Inc. Sports and Fitness App Market to Expand by More Than 60 Percent in Five Years; 2013 July 11 [cited 2014 Nov 14]. Available from: http://press.ihs.com/ press-release/design-supply-chain/sports-and-fitness-app-market-expand-more-60-percentfive-years.

8.

Bhas, N. Juniper Research Web site [Internet]. Hampshire (UK): Juniper Research.

Mobile Smart Wearable Device Shipments to Approach 130 Million by 2018, Juniper Research Finds; 2013 Dec 2 [cited 2014 Oct 14]. Available from: http://www.juniperresearch.com/ viewpressrelease. php?pr=414. 9.

Harding P, Holland AE, Delany C, Hinman RS. Do activity levels increase after total hip

and knee arhtroplasty? Clin Orthop Relat Res. 472(5):1502-1511, 2014 10. Kuhn M, Harris-Hayes M, Steger-May K, Pashos G, Clohisy JC. Total hip arthroplasty in patients 50 years of less: do we improve activity profiles? J Arthroplasty 28(5):872-876, 2013 11.

Toogood PA, Abdel MP, Spear JA, et al. The monitoring of activity at home after total

hip arthroplasty. Bone Joint J. 98-B(11):1450-1454, 2016 12. Battenberg, AK, Hopkins JS, Kupiec AD, Schmalzried TP. The 2012 Frank Stinchfield Award: Decreasing Patient Activity With Aging: Implications for Crosslinked Polyethylene Wear. Clin Orthop Relat Res. 471(2): 386-392, 2013 13. Schmalzried TP, Shepherd EF, Dorey FJ, et al. The John Charnley Award. Wear is a function of use, not time. Clin Orthop Relat Res. (381): 36-46, 2000 14.

Zahiri, CA, Schmalzried, TP, Amstutz, HC, et al. Assessing activity in joint replacement

patients. J. Arthroplasty 13:890-895, 1998 15. Shepherd EF, Toloza E, McClung CD, Schmalzried TP. Step Activity Monitor: Increased Accuracy in Quantifying Ambulatory Activity. J Orthop Res. 17(5):703-708, 1999 16. Fulk GD, Combs SA, Danks KA. Accuracy of 2 activity monitors in detecting steps in people with stroke and traumatic brain injury. Phys Ther 92(2):222-229, 2014 17. Ichinoseki-Sekine N, Kuwae Y, Higashi Y. Improving the accuracy of pedometer used by the elderly with the FTT algorithm. Med Sci Sports Exerc 38(9):1674-1681, 2006

18. Karabulut M, Crouter SE, Bassett DR. Comparison of two waist-mounted and two anklemounted electronic pedometers. Eur J Appl Physiol 95(4):335-343, 2005 19. Melanson EL, Knoll JR, Bell ML. Commercially Available Pedometers: Considerations for accurate step counting. Prev Med 39 (2):361-368, 2004 20. Dolan, B. MobiHealthNews Web site [Internet]. Boston (MA): MobiHealthNews. Fitbit, Jawbone, Nike had 97 percent of fitness tracker retail sales in 2013; 2014 Jan 15 [cited 2014 Dec 6]. Available from: http://mobihealthnews.com/28825/fitbit-jawbone-nike-had-97-percentof-fitness-tracker-retail-sales-in-2013/. 21. The NPD Group Web site [Internet]. Port Washington (NY): The NPD Group, Inc. Apple and Samsung Grow to Represent 68 Percent of Smartphones Owned in the US, According to The NPD Group; 2014 Jan 16 [cited 2014 Jul 6]. Available from: https://www.npd.com/ wps/portal /npd/us/news/press-releases/ apple-and-samsung-grow-to-represent-68-percent-ofsmartphones-owned-in%20 the-u-s-according-to-the- npd-group.

TABLE 1. Device Accuracy by Activity*

400-M Walk 400-M Run 10-M Walk 10 Stair Ascent 10 Stair Decent

StepWatch Activity Monitor

Fitbit One

Omron HJ-321

Sportline 340 Strider

Argus iPhone App

99.3% (99.1 - 99.5)

99.5% (99.4 - 99.6)

99.3% (99.0 - 99.6)

92.2% (89.2 - 95.3)

97.2% (96.3 - 98.1)

74.4% (72.8 - 76.0)

96.7% (95.7 - 97.7)

97.9% (97.4 - 98.4)

92.0% (88.3 - 95.6)

97.3% (96.5 - 98.1)

96.7% (95.9 - 97.5)

97.8% (97.1 - 98.5)

94.9% (93.9 - 95.9)

79.1% (74.4 - 83.8)

75.5% (73.1 - 78.0)

98.2% (97.0 - 99.4)

94.3% (92.0 - 96.6)

92.2% (89.8 - 94.6)

48.8% (41.9 - 55.6)

67.6% (63.5 - 71.6)

95.3% (93.2 - 97.5)

96.8% (93.7 - 99.9)

91.3% (88.7 - 94.0)

84.0% (79.6 - 88.4)

57.1% (52.2 - 62.0)

Runtastic iPhone App

Runtastic Google App

Noom Walk Google App

FitBit Force

Nike+ FuelBand SE

97.8% (97.0 - 98.7)

81.0% (77.3 - 84.6)

91.6% (89.4 - 93.7)

89.1% (87.7 - 90.6)

91.2% (88.8 - 93.7)

66.3% (60.3 - 72.3)

92.7% (90.8 - 94.6)

95.6% (94.9 - 96.2)

88.3% (85.3 - 91.3)

72.6% (67.0 - 78.2)

82.6% (79.0 - 86.1)

56.8% (51.7 - 62.0)

87.3% (83.9 - 90.8)

78.9% (73.5 - 84.3)

81.0% (76.4 - 85.6)

42.2% (33.7 - 50.7)

90.2% (86.0 - 94.5)

91.3% (88.3 - 94.4)

83.9% (79.6 - 88.2)

42.4% (33.8 - 51.1)

400-M 98.2% (97.9 - 98.6) Walk 400-M 85.8% (83.1 - 88.4) Run 10-M 84.5% (81.2 - 87.9) Walk 10 Stair 75.6% (71.4 - 79.7) Ascent 10 Stair 77.8% (73.6 - 81.9) Decent *Accuracy Percent (95% confidence interval)

TABLE 2. Mean Percent Error of Device by Activity*

400-M Walk 400-M Run 10-M Walk 10 Stair Ascent 10 Stair Decent

400-M Walk 400-M Run 10-M Walk 10 Stair Ascent 10 Stair Decent

StepWatch Activity Monitor

Fitbit One

Omron HJ-321

Sportline 340 Strider

Argus iPhone App

0.33 ± 0.69, (-1.85 ‒ 2.39) -25.5 ± 6.05, (-32.8 ‒ 10.6) 1.85 ± 2.52, (-2.56 ‒ 6.67) 0.89 ± 2.89, (-6.67 ‒ 6.67) 3.78 ± 7.77, (0.00 ‒ 40.0)

0.29 ± 0.37, (-0.37 ‒ 1.25) -0.07 ± 4.14, (-17.0 ‒ 5.71) -0.36 ± 2.42, (-6.00 ‒ 4.76) 3.67 ± 6.68, (-26.7 ‒ 16.7) -0.78 ± 6.47, (-33.3 ‒ 3.33)

0.41 ± 0.93, (-1.97 ‒ 4.38) 1.40 ± 1.87, (-2.38 ‒ 7.00) -1.23 ± 4.50, (-8.61 ‒ 6.98) 1.33 ± 10.1, (-40.0 ‒ 16.7) 3.56 ± 11.9, (-40.0 ‒ 20.0)

-7.31 ± 13.2, (-51.5 ‒ 2.67) -5.44 ± 13.4, (-55.3 ‒ 6.98) 17.0 ± 23.6, (-86.7 ‒ 19.1) -46.6 ± 37.1, (-96.7 ‒ 30.0) -9.78 ± 21.9, (-73.3 ‒ 13.3)

2.71 ± 4.15, (0.15 ‒ 24.4) 1.34 ± 3.53, (-14.3 ‒ 7.70) -24.0 ± 8.25, (-40.4 ‒ 5.83) -30.2 ± 14.6, (-66.7 ‒ 3.33) -42.9 ± 17.2, (-80.0 ‒ 20.0)

Runtastic iPhone App

Runtastic Google App

Noom Walk Google App

FitBit Force

Nike+ FuelBand SE

0.79 ± 1.98 (-4.80 ‒ 5.21) -13.6 ± 12.3, (-45.1 ‒ 5.19) 13.8 ± 12.8, (-4.6 ‒ 47.6) 24.0 ± 14.5, (-3.33 ‒ 53.3) 18.4 ± 17.2, (-20.0 ‒ 66.7)

-0.36 ± 4.08, (-8.81 ‒ 17.0) -7.41 ± 9.57, (-41.3 ‒ 2.65) -8.17 ± 10.9, (-35.6 ‒ 10.5) -0.07 ± 19.2, (-50.0 ‒ 53.3) 4.89 ± 20.4, (-33.3 ‒ 86.7)

-16.6 ± 17.8, (-56.9 ‒ 13.5) -33.2 ± 25.9, (-95.5 ‒ 2.21) -26.5 ± 18.5, (-58.9 ‒ 0.0) -18.2 ± 20.2, (-73.3 ‒ 20.0) -4.00 ± 10.2, (-30.0 ‒ 20.0)

-7.9 ± 10.2, (-37.2 ‒ 1.10) -2.92 ± 8.78, (-32.2 ‒ 6.32) -11.5 ± 15.6, (-61.4 ‒ 11.9) 0.78 ± 22.1, (-70.0 ‒ 46.7) -6.33 ± 20.6 (-70.0 ‒ 23.3)

-10.9 ± 6.46, (-34.2 ‒ 3.45) -3.41 ± 3.00, (-9.61 ‒ 4.28) -35.3 ± 21.1, (-80.9 ‒ 25.6) -55.1 ± 29.6, (-100 ‒ 6.67) -52.7 ± 26.5, (-100 ‒ 0.00)

*Mean percent error ± SD, (Range)

TABLE 3. Accuracy and Mean Error at Slow Walking Speeds Device Accuracy Percent* Walking Speed

Fitbit One

Omron HJ-321

Sportline 340 Strider

1.0 MPH 1.5 MPH 1.8 MPH 2.0 MPH

78.4% (69.5 ‒ 87.4) 97.6% (95.7 ‒ 99.4) 99.0% (98.8 ‒ 99.3) 99.4% (99.2 ‒ 99.5)

17.0% (9.7 ‒ 24.3) 75.3% (67.9 ‒ 82.6) 95.7% (94.4 ‒ 97.1) 97.9% (97.3 ‒ 98.4)

34.0% (25.4 ‒ 42.5) 63.7% (52.7 ‒ 74.7) 74.7% (64.9 ‒ 84.4) 80.8% (73.2 ‒ 88.5)

Device Mean Percent Error‡ Walking Speed

Fitbit One

Omron HJ-321

Sportline 340 Strider

1.0 MPH 1.5 MPH 1.8 MPH 2.0 MPH

-19.8% ± 30.9, (-100 ‒ 13.6) -1.72% ± 5.83, (-31.8 ‒ 2.86) 0.27% ± 1.30, (-2.45 ‒ 3.52) -0.04% ± 0.86, (-1.75 ‒ 2.06)

-82.6% ± 25.6, (-100 ‒ 8.33) -13.3% ± 30.9, (-90.4 ‒ 43.1) 0.88% ± 6.01, (-13.1 ‒ 19.4) 0.80% ± 2.70, (-3.85 ‒ 7.18)

-65.57% ± 29.3, (-100 ‒ 10.0) -36.3% ± 33.8, (-100 ‒ 0.74) -24.7% ± 31.9, (-99.5 ‒ 6.18) -17.9% ± 26.2, (-98.8 ‒ 12.9)

*Accuracy percent (95% confidence interval) ‡Mean percent error ± SD, (Range)