Convergent Validity of 3 Low Cost Motion Sensors With the ActiGraph Accelerometer

Journal of Physical Activity and Health, 2010, 7, 662-670 2010 Human Kinetics, Inc. Convergent Validity of 3 Low Cost Motion Sensors With the ActiGraph Accelerometer James J. McClain, Teresa L. Hart, Renee S. Getz, and Catrine Tudor-Locke Background: This study evaluated the utility of several lower cost physical activity (PA) assessment instruments for detecting PA volume (steps) and intensity (time in MVPA or activity time) using convergent methods of assessment. Methods: Participants included 26 adults (9 male) age 27.3 ± 7.1 years with a BMI of 23.8 ± 2.8 kg/m 2. Instruments evaluated included the Omron HJ-151 (OM), New Lifestyles NL-1000 (NL), Walk4Life W4L Pro (W4L), and ActiGraph GT1M (AG). Participants wore all instruments during a laboratory phase, consisting of 10 single ute treadmill walking bouts ranging in speed from 40 to 112 m/, and immediate following the laboratory phase and during the remainder of their free-living day (11.3 ± 1.5 hours). Previously validated AG MVPA cutpoints were used for comparison with OM, NL, and W4L MVPA or activity time outputs during the laboratory and free-living phase. Results: OM and NL produced similar MVPA estimates during free-living to commonly used AG walking cutpoints, and W4L activity time estimates were similar to one AG lifestyle cutpoint evaluated. Conclusion: Current findings indicate that the OM, NL, and W4L, ranging in price from $15 to $49, can provide reasonable estimates of free-living MVPA or activity time in comparison with a range of AG walking and lifestyle cutpoints. Keywords: physical activity, objective assessment, intensity, adult Over the past 20 years, increased awareness of the health benefits of physical activity (PA) has fueled expanded interest in development, validation, and application of objective PA assessment instruments, among them body worn motion sensors such as accelerometers 1 and pedometers. 2 4 Most pedometers provide a user friendly and low cost estimate of the volume of PA participation, commonly outputted as the number of steps taken on a digital screen display. 4,5 As a result, pedometers have been widely used in both research and applied settings as tools for PA assessment and behavior change. 4,6 However, pedometers have not typically provided outputs related to PA intensity or movement time. In contrast, many accelerometers are capable of providing valid and reliable measures both for PA volume (ie, in the form of total activity counts or steps) 7,8 and time in specific PA intensities (eg, daily moderate-to-vigorous physical activity or MVPA utes); an output which is directly linked to current public health recommendations. 9 11 Unfortunately, current accelerometer costs (ie, 200 to 400+ dollars), and both time and technical expertise often required for data collection, processing, and analyses has McClain is with the Cancer Prevention Fellowship Program, National Cancer Institute, Bethesda, MD. Hart is with the Dept of Human Movement Sciences, University of Wisconsin Milwaukee, WI. Getz is with the Dept of Exercise & Wellness, Arizona State University, Mesa, AZ. Tudor-Locke is with the Walking Behavior Laboratory, Pennington Biomedical Research Center, Baton Rouge, LA. limited use of these instruments in applications beyond basic research. 12 In addition, many of the accelerometers currently on the market lack a display screen capable of providing immediate wearer feedback about PA accumulation over the course of the day. This type of feedback is considered to be an important component for success within pedometer-based PA interventions. 13 Recently, lower cost (ie, less than $50) instruments have been developed and marketed that are designed to detect both PA volume (steps/day) and PA intensity (time in MVPA), while also offering immediate feedback to the wearer via a screen display. The emergence of an instrument class with the above characteristics may allow broader utilization of objectively assessed PA intensity by public health researchers, clinicians, and practitioners. 12 As with past technology advancements, however, establishing validity and reliability of PA outcome measures from these new instruments is necessary before widespread acceptance and adoption. A number of methods have been employed by manufacturers of these low cost technologies capable of measuring movement time or time in MVPA. One approach that has been used is to enable existing pedometers with integrated digital time clocks to detere stepping rate (ie, steps/) and dichotomously classify intensity of each ute as below MVPA or MVPA based on a given steps/ threshold (eg, 100 steps/ute). Another approach is to modify existing accelerometer technology (or develop new accelerometers) to simplify instrument use and reduce costs associated with additional hardware 662

Validity of 3 Low Cost Motion Sensors 663 and/or software. For example, by eliating the need (or option) for computer connectivity and downloading capacity, a number of costs can be eliated including requisite downloading hardware (ie, such as infrared ports and readers or USB ports and cables) and software necessary for data processing. Convergent validity between 1) previously validated accelerometer-based measures of time in MVPA, and 2) pedometer measures of steps taken, would provide further support for these emerging low cost instruments. Currently, the ActiGraph (AG; ActiGraph LLC, Pensacola, FL) is the most commonly used accelerometer for measurement of PA intensity 14 and the Yamax SW-200 (YM; Yamasa, Tokyo, Japan) is widely accepted as a valid measure of steps taken. 2 Therefore, the purpose of this study was to compare low cost instrument outputs of 1) steps with manually counted steps (during laboratory testing only) and YM steps (during free-living), and 2) detected time in MVPA with AG detected time in MVPA (during both laboratory and free-living conditions) Methods A convenience sample of 26 adults, including 9 males (age = 25.8 ± 5.5 years; BMI= 24.7 ± 2.4 kg/m 2 ) and 17 females (age = 28.1 ± 7.8 years; BMI= 23.4 ± 3.0 kg/m 2 ) were recruited from a population of university students, faculty and staff, as well as surrounding community members. All aspects of this study were approved by the Arizona State University Institutional Review Board and participants provided written informed consent following an explanation of the study s purpose and monitoring procedures. Instruments The cost, features, and primary functions of the 5 instruments compared in this study are provided in Table 1. The instruments internal sensors include both hair spring and coiled spring suspended lever arm pedometers, as well as differing uniaxial piezoelectric accelerometers. In all instruments these sensors are used to detect steps, peak accelerations (ie, maximum acceleration), or activity counts (AG only; which quantifies the amplitude and frequency of detected accelerations) during movement. These outputs are then used to further detere activity intensity or activity time for each instrument, with the exception of the YM (ie, which outputs cumulative steps only). For the Omron HJ-151 (OM; OMRON Healthcare Inc., Bannockburn, IL), intensity is dichotomously assigned (ie, as below MVPA or MVPA) on a ute-byute basis based on estimated walking speed which, in turn, is internally modeled from steps taken and participant height (inputted during instrument set-up) using a proprietary algorithm. The output of moderate utes is the cumulative utes detected at MVPA. Likewise, the output of moderate intensity steps represents the cumulative steps taken during all utes classified as MVPA. The NL-1000 (NL; New Lifestyles Inc., Lee s Summit, MO) detects the maximum acceleration over each 4 second epoch (ie, sampling interval), and each epoch is then categorized into 1 of 11 activity intensity levels based on this value. A detailed description of the NL sensor mechanism and PA intensity categorization schema has been previously reported. 15 An embedded function accessible using the on-board keypad and screen display allows the NL MVPA intensity range to be modified if required. For example, the manufacturer default MVPA range used in the current study is 4 to 9, but there is evidence to suggest that a range of 5 to 9 might be useful for children age 10 to 11 years. 15 The output of activity time represents the cumulative time (displayed in hours, utes and seconds) detected within the selected MVPA range. The Walk4Life W4L Pro (W4L; Walk4Life Inc., Plainfield, IL) detects activity time (ie, irrespective of movement intensity) using a stopwatch mechanism. According to the manufacturers, the stopwatch begins counting activity time when 3 consecutive steps are detected and accumulates time until continuous movement stops. The W4L activity time output represents the cumulative duration of continuous movement bouts aggregated over the monitored period (eg, the full day). For the AG (GT1M model used in current study), time in specific intensities of activity can be derived from activity count outputs. Activity counts are summed over each epoch (ie, typically a ute in length in adults). 16 The sum of the activity counts in a given epoch is related to activity intensity and can be categorized (eg, light, moderate, vigorous) based on validated activity count cutpoints. 9 Detailed technical specification for the AG are provided elsewhere. 17 Procedures Within the current study, participants first completed a treadmill-based laboratory phase, immediately followed by a free-living phase. On the first morning, each participant s height (cm) and weight (kg) (without shoes) were measured using a portable stadiometer and electronic scale, respectively. Instruments were prepared for both studies by (1) initializing the AG; (2) synchronizing the researcher s timepiece, OM, NL, and AG internal clocks, with that of the PC used to initialize the AG (ie, the AG synchronizes automatically during initialization, the timepiece, OM and NL were manually synchronized to within ± 1 second of the PC); and (3) resetting instruments with a keypad accessible reset button (for the W4L and YM). Participants were then fitted with the elastic belt worn around the waist on which the 5 instruments under study were attached. The AG was worn at the midaxillary line of hip, with the remaining 4 instruments systematically rotated by participant from right to left on the waist at or near the midline of the thighs (ie, position 1 = slightly right of the midline of the right thigh, 2 = midline of right thigh, 3 = midline of left thigh, 4 = slightly left of midline of left thigh).

664 McClain et al Table 1 Comparison of the Cost, Functions, and Features for the Instruments Used in the Current Study Instrument feature/ function Cost (as of 12/1/2008) Omron HJ-151 $16 25 multiple distributors NewLifestyles NL-1000 Walk4Life W4L Pro Yamax SW-200 ActiGraph GT1M $48 single distributor $24 single distributor $17 22 multiple distributors $335 single distributor Internal Sensor uniaxial piezoelectric accelerometer uniaxial piezoelectric accelerometer hair spring suspended lever arm pedometer coil spring suspended lever arm pedometer dual-axial piezoelectric accelerometer (vertical axis default) Instrument Outputs 1) Steps 1) Steps 1) Steps 1) Steps 1) Activity counts 2) Distance (miles) 2) Distance (miles/km) 2) Distance (miles) 2) Steps derived 3) MVPA Steps 3) MVPA (hr::sec) 3) Movement time outputs: time in MVPA, etc 4) MVPA () (hr::sec) Epoch (MVPA/activity time) 1 4 sec 1 sec NA 1 60 sec (user defined) Memory 7 day (on instrument) 7 day (on instrument) 7 day (on instrument) None 170 days based on current mode and epoch) Display screen multifunction LCD multifunction LCD multifunction LCD LCD None Keypad functions 1) Set time and user height 1) Set time and user stide length select output view 1) Set time and user stide length 2) select output view 2) select output view 3) Reset (2 sec delay) 1) Reset None Dimensions 5.3 cm circumference 2.3 cm depth 7.0 4.0 2.5 cm 5.2 3.9 1.9 cm 5.2 3.9 1.9 cm 3.8 3.7 1.8 cm Attachment site midline of thigh midline of thigh midline of thigh midline of thigh midaxillary line of the hip During the laboratory phase, participants wore all instruments concurrently while walking on a calibrated treadmill at 0% grade for 10 separate 1 ute walking stages. Each stage was time synchronized with the internal clocks of the OM, NL, and AG and the researcher s electronic time piece. To clear all instruments with an internal sampling interval, participants stood stationary on the tracks of the treadmill for 2 utes before the first walking stage and in between each subsequent stage. The treadmill protocol began with participants walking at 40 m/ for 1 ute, with speed increasing by 8 m/ each stage, ending finally at 112 m/ for subsequence walking stages. Inclusion of the 40 m/ and 48 m/ walking stages within the current protocol provides evidence of instrument-specific step counting error at lower walking speeds than has previously been reported among adults. Two researchers were present during the laboratory testing phase. One researcher monitored time during the protocol on the electronic timepiece, prompted the participant to start and stop walking, and manipulated treadmill speed between walking stages. The second researcher counted the number of actual steps taken during each walking stage using a handheld manual counting device. Both researchers assisted with reading and recording those instruments with screen displays (ie, the OM, NL, W4L, and YM) of steps and time in MVPA or activity time immediately following each stage of the laboratory protocol. The free-living phase of data collection began immediately following completion of the laboratory study. Instruments were not cleared, reset, or reinitialized before the beginning of the free-living phase. Participants were instructed to wear all instruments for the remainder

Validity of 3 Low Cost Motion Sensors 665 of waking hours of their day and to proceed with their usual daily activities. At the end of the day, participants recorded steps and time in MVPA or activity time outputs from the OM, NL, W4L, and YM. Participants returned all instruments the following day and recorded values were checked against data stored in memory for the OM, and NL as a method of checking recording accuracy. No discrepancies were observed between participant recorded data for the OM and NL, and the researcher confirmed values from the instrument memory upon return. Data Processing Since instruments were not reset between subsequent stages of the laboratory study or between the laboratory and free-living phases, recorded data represented the aggregate total steps and detected time in MVPA or activity time at each specific recording period (ie, between walking stages or at the end of the free-living phase). Walking stage specific data for steps and time in MVPA or activity time for the OM, NL, W4L, and YM were calculated by subtracting previous stage data from the aggregate totals (eg, Stage 4 steps = aggregate steps recorded after stage 4 aggregate steps recorded after stage 3). The same procedure was used to calculate free-living steps and time in MVPA or activity time for the OM, NL, W4L, and YM, using participant recorded data at the end of the free-living phase and subtracting instrument matched data recorded after the last stage of laboratory testing. Both the NL MVPA data and W4L activity time data are displayed to the second (Hr:Min:Sec). During the laboratory phase, the NL and W4L raw data were reclassified into integer values of 0 and 1, representing 0 and 1 ute of MVPA, to generate comparable data to ute-by-ute OM and AG values. If detected MVPA or movement time was 48 seconds (ie, 80% of a 1 ute walking stage), then NL or W4L data were recoded as 1 ute of either MVPA or movement time, respectively. Alternatively, values of < 48 seconds were recoded as 0 utes. The finer grain of the 1 and 4 second epochs for the W4L and OM, respectively, would potentially allow short portions of the 1 ute stage to be classified at different intensities, such as at the beginning and end of the stage, dependent on participant reaction time to stage start and stop commands. As such, the above 80% criterion was established based on the research team s a priori best judgment for appropriate comparison. However, since aggregate data were compared over the entire free-living phase, raw outputs from the NL and WL during free-living were used without further manipulation. AG step and activity count data were outputted in 1 ute epochs for both the laboratory and free-living phases. These data were extracted from individual data files for synchronized output intervals corresponding to walking stages 1 to 10 of the laboratory phase and for the duration of the free-living phase. Step outputs from this latter phase of the study were aggregated to detere free-living AG detected steps. AG activity count data were recoded in the current study to produce MVPA outputs based on a range of available moderate intensity activity count cutpoints. These data were used for comparison of intensity outputs with OM, NL, and W4L detected MVPA and activity time. Functionally, activity counts from individual epochs were reclassified dichotomously as below MVPA or at least MVPA using validated cutpoints. The AG cutpoints used fall into 2 general categories of walking cutpoints (ie, providing estimates of time in imally moderate intensity walking behaviors as MVPA) and lifestyle cutpoints (ie, providing estimates of both time in moderate lifestyle behaviors not comprised of continuous walking, and moderate intensity walking behaviors as MVPA). 9 Three walking cutpoints were used in the current study including 1) Freedson 3 MET (AG_F3; 1952 counts/), 2) Freedson 3.5 MET (AG_F3.5; 2592 counts/), and 3) a recently used cutpoint for NHANES AG data (AG_NHANES; 2020 counts/ ). 17,18 Two lifestyle cutpoints were used in the current study including (1) Swartz (AG_Swartz; 574 counts/ ) and (2) Matthews (AG_Matthews; 760 counts/ ). 9,19 In addition, AG ute-by-ute step outputs were recoded into dichotomous intensity categories (ie, below MVPA or MVPA) using a previously validated moderate intensity steps/ cutpoint established by Tudor-Locke et al. (AG_Steps; 100 steps/). 20 These data were used to investigate the potential of raw steps/ data to approximate current accelerometer activity count threshold-based methods for MVPA deteration Within the laboratory phase, OM and NL MVPA outputs were compared against 2 AG walking cutpoints (ie, AG_F3 and AG_F3.5 which represent the lowest and highest AG walking cutpoints evaluated, respectively), and W4L activity time outputs were compared against both AG lifestyle cutpoints. Comparisons of OM and NL vs. AG walking cutpoints, and W4L vs. AG lifestyle cutpoints were detered a priori based on predicted maximal agreement. Post hoc comparisons confirmed lower levels of agreement between alternate comparisons (eg, W4L vs. AG walking cutpoints) of intensity outputs during the laboratory phase (data not shown). Within the free-living phase, OM, NL, W4L, and AG_Steps MVPA and activity time outputs were compared against all 3 AG walking and 2 AG lifestyle cutpoints. Statistical Analyses Laboratory Phase. Means and standard deviations for steps were calculated for all instruments and the manually counted steps for each stage of the laboratory phase. For the steps data from each instrument, percent error [((instrument steps counted steps) / counted steps) 100] and absolute percent error [(( instrument steps counted steps ) / counted steps) 100] of instrument detected vs. manually counted steps were calculated for each stage and overall across all stages. A repeated-measures ANOVA was used to compare mean difference scores (pedometer detected steps manually

666 McClain et al counted steps) for steps taken during the laboratory phase. 2 Separately, classification agreement (ie, % agreement for dichotomous below MVPA or at least MVPA outputs) was calculated by stage and overall across all stages between OM, NL, and W4L MVPA and activity time outputs vs. AG MVPA (ie, % agreement for OM and NL MVPA outputs calculated against AG walking cutpoints and % agreement for W4L activity time calculated against AG Lifestyle cutpoints based on expected patterns of concordance). Free-Living Phase. Means and standard deviations for steps (ie, for all 5 instruments) and detected time in MVPA or activity time (ie, for OM, NL, W4L, AG_Steps, and the 5 AG walking or lifestyle cutpoints) were calculated for the free-living phase of the study. Separate repeated-measures ANOVAs were used to compare means between steps and MVPA or activity time outputs during the free-living phase. Post hoc pairwise comparisons were used to assess differences between specific MVPA or activity time outputs from instruments or AG cutpoints during free-living. Correlations between instrument outputs of free-living MVPA or movement time were also calculated. SPSS version 13.0 was used to complete all analyses and an alpha level of 0.05 was used to interpreted significance. Laboratory Phase Results The repeated-measures ANOVA of mean step difference scores yielded a significant main effect for instrument (F = 73.74, P <.001), and a significant instrument walking speed interaction (F = 7.91, P <.001). Percent error and absolute percent error for instrument detected steps (vs. manually counted steps) overall and by walking stage are presented in Table 2. With the exception of OM, all mean percent error values by instrument and walking stage reflect a relative undercounting of steps vs. manually counted steps. In general, percent error and absolute percent error values were more divergent (ie, further from 0.0%) at the slower walking speeds and more convergent (ie, nearer to 0.0%) at higher walking speeds. The overall percent error and absolute percent error values (representing the mean error across all 10 walking stages) should be interpreted with caution, since these values are a reflection of both the number of stages and the specific walking speed selected during the protocol. Classification agreement for walking intensity (below MVPA vs. at least MVPA) at each of the walking stages, between OM, NL, and W4L vs. selected AG cutpoints is presented in Table 3. Classification agreement for MVPA during the laboratory phase of the study was high overall (ie, the mean across all 10 walking stages) ranging from 84.6% to 88.9% in comparisons between OM and NL vs. AG walking cutpoints, and from 86.2% to 88.1% in the comparison between W4L vs. AG lifestyle cutpoints. Classification agreement tended to be lower (ie, below 80%) for the OM, NL, and W4L walking at speeds of 72 to 80 m/, 72 to 88, m/, and <64 m/, respectively. Free-Living Phase Average wearing time (based on AG data) among participants during the free-living phase of the study was 11.3 ± 1.5 hours. Average steps/day detected by the YM, Table 2 Percent Error and Absolute % Error for Instrument Detected vs. Manual Counted Steps During Separate Walking Stages and Overall Across Stages During the Laboratory Phase 40 m/ 48 m/ 56 m/ 64 m/ 72 m/ 80 m/ 88 m/ 96 m/ 104 m/ 112 m/ Overall % error Yamax 14.8 44.7 33.9 24.9 16.3 11.5 7.8 3.8 2.8 1.7 1.0 Omron 0.2 9.3 2.1 3.5 2.4 2.4 2.4 0.3 1.0 0.2 2.2 NL-1000 7.7 36.3 20.2 11.2 4.8 1.8 0.7 1.0 0.8 0.2 0.3 Walk-4-Life 15.3 44.4 29.7 24.8 16.6 13.0 5.8 5.4 5.4 4.3 3.5 ActiGraph 18.7 65.2 51.2 39.8 18.9 5.6 2.1 1.3 1.5 1.3 0.5 Absolute % error Yamax 17.3 47.3 39.8 28.7 19.5 14.9 9.8 5.7 3.3 2.6 1.5 Omron 5.4 22.4 9.2 5.0 3.4 3.8 3.2 0.8 1.6 1.1 3.1 NL-1000 8.9 37.8 23.0 13.0 6.5 2.6 1.6 1.1 1.1 1.1 0.8 Walk-4-Life 16.1 44.7 31.1 25.7 17.3 13.8 7.1 6.3 5.7 5.0 3.9 ActiGraph 19.1 65.2 51.2 39.8 18.9 6.5 2.4 1.3 1.5 1.6 2.7 Note. % error = [((instrument detected steps manually counted steps) / manually counted steps) 100]; absolute error = [(( instrument detected steps manually counted steps ) / manually counted steps) 100].

Validity of 3 Low Cost Motion Sensors 667 Table 3 Classification Agreement Between Specific ActiGraph Walking and Lifestyle MVPA Cutpoints With Omron, NL-1000, and Walk4Life Detected MVPA or Activity Time During Separate Walking Stages and Overall Across Stages During the Laboratory Phase 40 m/ 48 m/ 56 m/ 64 m/ % agreement MVPA Overall ActiGraph Freedson 3 MET vs. Omron 88.5 100.0 96.2 96.2 80.7 50.0 65.4 96.2 100.0 100.0 100.0 NL-1000 84.6 100.0 100.0 100.0 80.8 53.8 46.2 65.4 100.0 100.0 100.0 ActiGraph Freedson 3.5 MET vs. Omron 88.9 100.0 96.2 96.2 84.6 65.4 53.8 92.3 100.0 100.0 100.0 NL-1000 87.3 100.0 100.0 100.0 92.3 84.6 34.6 61.5 100.0 100.0 100.0 ActiGraph Matthews vs. Walk-4-Life 86.2 53.9 66.5 69.2 92.3 92.3 100.0 96.2 96.2 100.0 100.0 ActiGraph Swartz vs. Walk-4-Life 88.1 53.9 65.4 80.8 96.2 92.3 100.0 96.2 96.2 100.0 100.0 Note. Percent agreement is based on dichotomous classification of each 1 ute walking stage as below MVPA or MVPA for all instruments and specific ActiGraph cutpoints. ActiGraph MVPA cutpoints included Freedson 3 MET ( 1952 counts/; based on Freedson et al, 1998), Freedson 3.5 MET ( 2592 counts/; based on Freedson et al, 1998), Matthews ( 760 counts/; based on Mathews, 2005), and Swartz ( 574 counts/; based on Swartz et al, 2000). 72 m/ 80 m/ 88 m/ 96 m/ 104 m/ 112 m/ OM, NL, W4L, and AG during the free-living phase were 8861 ± 3659, 9284 ± 3070, 10029 ± 3169, 8236 ± 4026, and 9359 ± 2994, respectively. The overall test of instrument-detected free-living steps/day by repeated measure ANOVA was nonsignificant (F = 0.98, P =.42). Correlations between MVPA or activity time outputs from OM, NL, W4L, and AG_Steps with AG walking and lifestyle cutpoints are presented in Table 4. Correlations between OM, NL, and AG_Steps outputs vs. AG walking cutpoints were high (ranging from 0.82 to 0.93). Correlations between W4L vs. AG lifestyle cutpoints were lower (0.54 and 0.57 vs. AG_Matthews and AG_Swartz, respectively). Minutes of MVPA or activity time detected during free-living are provided in Table 5 for the 9 intensity outputs evaluated during free-living (ie, OM, NL, W4L, AG_Steps, and both AG walking and lifestyle cutpoints). There was a significant overall difference (F = 30.43, P <.0001) based on repeated-measures ANOVA between the 9 intensity outputs during the free-living phase. Based on post hoc pairwise comparisons, no differences were observed between any possible pairings of OM, NL, AG_Steps or AG walking cutpoints; and W4L was not different from AG_Matthews. However, all pairwise comparisons were significantly different (P <.0001) between outputs from OM, NL, AG_Steps or AG walking cutpoints versus outputs from W4L or AG lifestyle cutpoints. In addition, AG_Swartz yielded higher estimates of MVPA when compared against either W4L (P =.0015, Δ=24.9 ) or AG_Matthews (P =.0095, Δ=20.2 ). Discussion The current study evaluated the capability of several lower cost PA assessment instruments for detecting PA volume (ie, steps) and PA intensity (ie, time in MVPA or activity time) when compared against criterion (and/or accepted standards) and convergent methods of PA assessment. Pedometers have provided a method in recent years for simple and low cost objective assessment of free-living PA volume. 2 However, objective assessment of free-living PA intensity has typically required the use of higher cost accelerometers and/or more cumbersome instruments such as heart rate monitors, most of which require individual calibration and a sensor be worn against the skin on a chest strap. 16,21 The laboratory phase of the study was designed to compare instrument step outputs against criterion (manually counted steps) and convergent (AG detected intensity) outcomes under controlled walking bouts at specific speeds. Consistent with previous research, 22 the OM and NL piezoelectric pedometers tended to more accurately detect steps at slower walking speeds than did the YM or W4L hair spring and coiled spring suspended lever arm models (ie, with lower absolute percent error at every walking stage from 40 to 104 m/ for the OM and NL vs. YM and W4L). In addition, generally high levels of overall agreement (ie, over 80%) were observed between OM and NL MVPA and AG walking cutpoints, and between W4L activity time an AG lifestyle cutpoints. However, decreased classification agreement (ie, below

668 McClain et al Table 4 Correlations Between Free-Living Omron, NL-1000, Walk4Life, and ActiGraph Steps/Min Detected MVPA or Activity Time Outputs vs. ActiGraph Walking and Lifestyle MVPA Outputs Freedson 3 MET Freedson 3.5 MET NHANES Swartz Matthews Omron 0.88 0.87 0.88 0.33 0.43 NL-1000 0.88 0.82 0.87 0.47 0.57 ActiGraph steps/ 0.88 0.94 0.89 0.25 0.35 Walk-4-Life 0.32 0.18 0.21 0.57 0.54 Note. ActiGraph MVPA cutpoints included Freedson 3 MET ( 1952 counts/; based on Freedson et al, 1998), Freedson 3.5 MET ( 2592 counts/ ; based on Freedson et al, 1998), NHANES ( 2020 counts/; based on Troiano et al, 2008), Steps/ ( 100 steps/; based on Tudor-Locke et al, 2005), Matthews ( 760 counts/; based on Matthews, 2005), and Swartz ( 574 counts/; based on Swartz et al, 2000). Table 5 Mean and Range of Detected MVPA or Activity Time From Omron, NL-1000, Walk4Life, ActiGraph Steps/Min, and ActiGraph Walking and Lifestyle MVPA Cutpoints During the Free-Living Phase Instrument Cutpoints Mean MVPA () MVPA range () Walking Omron 38.0 ± 19.7 a 0.0 75.0 NL-1000 44.6 ± 22.3 a 14.3 91.9 ActiGraph Freedson (3 MET) 47.9 ± 20.6 a 22.0 84.0 Freedson (3.5 MET) 39.3 ± 20.3 a 2.0 80.0 NHANES 46.8 ± 20.4 a 21.0 84.0 Steps/ 36.0 ± 20.3 a 7.0 77.0 Lifestyle Walk-4-Life 90.6 ± 41.2 b 15.1 187.0 ActiGraph Swartz 115.5 ± 39.4 55.0 240.0 Matthews 95.3 ± 34.5 b 44.0 199.0 Note. Mean MVPA values with matching subscripts are not significantly different. All other pairwise comparisons are significantly different (P <.01). ActiGraph MVPA cutpoints included Freedson 3 MET ( 1952 counts/; based on Freedson et al, 1998), Freedson 3.5 MET ( 2592 counts/; based on Freedson et al, 1998), NHANES ( 2020 counts/; based on Troiano et al, 2008), Steps/ ( 100 steps/; based on Tudor-Locke et al, 2005), Matthews ( 760 counts/; based on Matthews, 2005), and Swartz ( 574 counts/; based on Swartz et al, 2000). 80%) was observed between OM, NL, and WL vs. AG cutpoints at walking speeds near instrument-specific moderate thresholds. For example, when comparing OM and NL vs. AG walking cutpoints, agreement at speeds < 72 m/ and > 88 m/ were near perfect (ranging from 80.7% to 100%) but agreement at speeds between 72 to 88 m/ tended to be lower (ranging from 34.6% to 96.2%). Upon closer inspection of the patterns of disagreement, it was apparent that the majority of discordant pairings for the OM and NL vs. the AG_F3 walking cutpoint occurred when the AG_F3 classified a stage as at least MVPA but the OM and NL classified it as below MVPA (ie, 70% and 97.5% for OM and NL vs. AG_F3, respectively). This suggests that in comparison with the AG_F3 (a common cutpoint used in AG research), both the OM and NL have slightly higher moderate intensity thresholds. Findings from the laboratory study help us further understand potential discrepancies that exist between specific instrument outputs during highly controlled conditions. However, it is important to remember that the primary purpose and function of instruments such as the OM, NL, and W4L is to assess volume and/or intensity of free-living behaviors. As a result, the freeliving phase of the study serves to provide a real life comparison between instrument outputs under imal researcher imposed constraints (ie, allowing participant self-selection of individual activities, walking speed, and length of activity bouts).

Validity of 3 Low Cost Motion Sensors 669 These distinct differences between the controlled laboratory phase and free-living phase are perhaps best highlighted by the steps data. As noted above, there were no significant differences between instrument-detected steps during free-living. Interestingly, the highest and lowest overall absolute percent error in steps during the laboratory phase were detected by the AG (absolute % error = 19.1%) and OM (absolute % error = 5.4%), respectively. However, during the free-living phase, the mean difference in AG and OM detected steps (ie, AG-OM) was only 75 steps or 0.8% of OM detected steps. Still, it is important to remember that our sample consisted of generally healthy young to middle age adults. Free-living comparisons of instrument steps outputs within a population with slower self-selected walking speeds (such as older adults) or higher abdoal obesity (which may result in larger instrument tilt angles from the optimal vertical position) may produce more divergent results. 22 Results from the free-living phase of the study provide evidence, although preliary, for the potential utility of low cost instruments such as the OM, NL, and W4L for assessing intensity of free-living PA. Within our sample, there were no significant differences between OM and NL vs. AG walking based cutpoints and W4L was not significantly different from the AG_Matthews lifestyle cutpoint. For example, on average the NL detected 4.8% fewer utes in MVPA than the AG_NHANES cutpoint. These data suggest that there is relatively limited difference between estimated MVPA obtained using the NL when compared with AG_NHANES, a moderate AG cutpoint recently used to generate national prevalence estimates for PA participation among United States adults in the National Health and Nutrition Exaation Survey. 18 Present findings are in agreement with a previous instrument comparison between the Kenz Lifecorder EX (Suzuken Co. Ltd., Nagoya, Japan), a downloadable accelerometer with the same internal sensor and intensity classification scheme as the NL, and AG walking cutpoints which reported < 7% difference (nonsignificant) between-instrument time in MVPA among free-living healthy adults. 23 In addition, the simple activity time function on the W4L pedometer (which counts time using a simple stopwatch function during continuous movement) produced outputs during free-living which were only 5.0% lower than AG_Matthews, an AG lifestyle cutpoint. Convergence of OM, NL, and W4L instruments with the above mentioned respective AG walking and lifestyle cutpoints lends a degree of support for their utility in assessing PA intensity outcomes in applications such as interventions, program evaluation, and surveillance. The availability of such an instrument class helps to lower the financial and data processing burdens typically associated with objectively measured PA intensity. As a result, a wider range of researchers and practitioners may benefit from the opportunity to use an objective measure of PA intensity in their work, an outcome that is directly linked to recommendations for PA and health. 11 Among the low cost devices evaluated, a relative advantage of the NL is the ability to modify the instrument s moderate intensity range using an imbedded keypad function. The manufacturer s default MVPA threshold was used in the current study (ie, intensity 4 to 9), which according to previous validation research corresponds to an estimated 3.6 MET threshold for moderate intensity. 24 The NL modifiable intensity range would allow the NL-1000 to be used to assess both moderate lifestyle and moderate walking PA in a similar fashion as that used for lifestyle or walking based AG moderate intensity cutpoints. 9 In fact, the above mentioned study comparing MVPA estimates from the AG and the Lifecorder EX found that use of a MVPA range of 1 to 9 produced convergent estimates with AG lifestyle cutpoints (ie, using the same AG_Swartz and AG_Matthews cutpoints used herein). 23 The performance of a previously established moderate intensity steps/ cutpoint (ie, 100 step/) was also evaluated in the current study in a proof of concept method using AG steps data. Since the AG is capable of storing epoch level steps data, the raw steps data detected during free-living can be recoded using these steps/ cutpoints in a manner similar to using an activity counts/ cutpoint to detere intensity. There was no significant difference between AG_Steps data with either the OM, NL, or AG walking cutpoints. Still, AG_Steps produced the lowest mean MVPA estimate (36.0 ± 20.3 ) of all instruments or AG cutpoints. This suggests that convergence of a steps/ based MVPA threshold with AG walking cutpoints may be optimized, at least within our sample, by using a lower steps/ threshold. In this study, to approach a similar mean MVPA estimate as AG walking cutpoints, the steps/ threshold had to be lowered to 80 steps/, resulting in an average of 43.4 ± 19.8 utes of MVPA or 7.3% less time in MVPA than AG_NHANES. However, the current study suggests that the previously established 100 steps/ cutpoint 20 provides a sufficiently high steps/ target for achieving a moderate walking pace, as assessed convergently with AG walking based intensity outputs. Future research may further refine steps/ MVPA cutpoints for specific populations (eg, specific age or BMI groups). It is important to interpret the current findings with an appropriate degree of caution due to the limitation of the study sample size and characteristics. In the future, assessing convergent validity of these low cost instrument outputs within a larger and more diverse sample would allow for further stratified comparisons (eg, by factors such as age and gender). In addition, the study sample consisted of generally healthy young to middle age adults with no ambulatory limitations. Further evidence of convergence of instrument outputs will be necessary before the current finding can be generalized to obese individuals, older adults, or individuals with ambulatory limitations. In addition, current findings among adults do not constitute a basis for conclusions about a similar degree of convergence between these instrument outputs among children or adolescents. Finally, validation of low

670 McClain et al cost instrument outputs of PA intensity against a criterion measure in future studies would provide an additional level of evidence for accuracy of these devices within research. In summary, the current study evaluated the capability of several lower cost PA assessment instruments for detecting PA volume (steps) and intensity (time in MVPA or activity time) when compared against criterion and/or accepted standards and convergent methods of PA assessment. Although large differences were observed during the laboratory phase of the study for instrument error in detected steps (with the OM and NL displaying the lowest absolute percent error overall), there were no statistically significant differences in steps detected during the free-living phase of the study between instruments. Results from the study provide encouragement, although preliary, for the utility of low cost instruments such as the OM, NL, and W4L for assessing intensity of freeliving PA. Specifically, there were no significant freeliving differences between OM and NL vs. AG walking cutpoints and W4L was not significantly different from the AG_Matthews lifestyle cutpoint. Current findings indicate that the OM, NL, and W4L, ranging in price from $15 to $49, can provide reasonable estimates of free-living MVPA or activity time in comparison with a range of AG walking and lifestyle cutpoints. The availability of a class of instruments such as these helps to lower the financial and data processing burden typically associated with objectively measured PA intensity. As a result, a wider range of researchers and practitioners may benefit from the opportunity to use an objective measure of PA intensity in their work. References 1. Troiano RP. A timely meeting: objective measurement of physical activity. Med Sci Sports Exerc. 2005;37(11, Suppl):S487 S489. 2. Schneider PL, Crouter SE, Bassett DR. Pedometer measures of free-living physical activity: comparison of 13 models. Med Sci Sports Exerc. 2004;36(2):331 335. 3. Crouter SE, Schneider PL, Karabulut M, Bassett DR, Jr. Validity of 10 electronic pedometers for measuring steps, distance, and energy cost. Med Sci Sports Exerc. 2003;35(8):1455 1460. 4. Tudor-Locke CE, Myers AM. Methodological considerations for researchers and practitioners using pedometers to measure physical (ambulatory) activity. Res Q Exerc Sport. 2001;72(1):1 12. 5. Bassett DR, Jr, Strath SJ. Use of Pedometers to Assess Physical Activity. In: Welk GJ, ed. Physical activity assessment for health-related research. Champaign, IL: Human Kinetics; 2002:163 178. 6. Tudor-Locke C, Williams JE, Reis JP, Pluto D. Utility of pedometers for assessing physical activity: construct validity. Sports Med. 2004;34(5):281 291. 7. Le Masurier GC, Lee SM, Tudor-Locke C. Motion sensor accuracy under controlled and free-living conditions. Med Sci Sports Exerc. 2004;36(5):905 910. 8. Hoos MB, Plasqui G, Gerver WJ, Westerterp KR. Physical activity level measured by doubly labeled water and accelerometry in children. Eur J Appl Physiol. 2003;89(6):624 626. 9. Matthew CE. Calibration of accelerometer output for adults. Med Sci Sports Exerc. 2005;37(11, Suppl):S512 S522. 10. Freedson P, Pober D, Janz KF. Calibration of accelerometer output for children. Med Sci Sports Exerc. 2005;37(11, Suppl):S523 S530. 11. Pate RR, Pratt M, Blair SN, et al. Physical activity and public health. A recommendation from the Centers for Disease Control and Prevention and the American College of Sports Medicine. JAMA. 1995;273(5):402 407. 12. Ward DS, Evenson KR, Vaughn A, Rodgers AB, Troiano RP. Accelerometer use in physical activity: best practices and research recommendations. Med Sci Sports Exerc. 2005;37(11, Suppl):S582 S588. 13. Richardson CR, Newton TL, Abraham JJ, Sen A, Jimbo M, Swartz AM. A meta-analysis of pedometer-based walking interventions and weight loss. Ann Fam Med. 2008;6(1):69 77. 14. Treuth MS, Sherwood NE, Butte NF, et al. Validity and reliability of activity measures in African-American girls for GEMS. Med Sci Sports Exerc. 2003;35(3):532 539. 15. McClain JJ, Sisson SB, Washington TL, Craig CL, Tudor-Locke C. Comparison of Kenz Lifecorder EX and ActiGraph accelerometers in 10-yr-old children. Med Sci Sports Exerc. 2007;39(4):630 638. 16. Chen KY, Bassett DR, Jr. The technology of accelerometry-based activity monitors: current and future. Med Sci Sports Exerc. 2005;37(11, Suppl):S490 S500. 17. Freedson PS, Melanson E, Sirard J. Calibration of the Computer Science and Applications, Inc. accelerometer. Med Sci Sports Exerc. 1998;30(5):777 781. 18. Troiano RP, Berrigan D, Dodd KW, Masse LC, Tilert T, McDowell M. Physical activity in the United States measured by accelerometer. Med Sci Sports Exerc. 2008;40(1):181 188. 19. Swartz AM, Strath SJ, Bassett DR, Jr, O Brien WL, King GA, Ainsworth BE. Estimation of energy expenditure using CSA accelerometers at hip and wrist sites. Med Sci Sports Exerc. 2000;32(9, Suppl):S450 S456. 20. Tudor-Locke C, Sisson SB, Collova T, Lee SM, Swan PD. Pedometer-detered step count guidelines for classifying walking intensity in a young ostensibly healthy population. Can J Appl Physiol. 2005;30(6):666 676. 21. Strath SJ, Swartz AM, Bassett DR, Jr, O Brien WL, King GA, Ainsworth BE. Evaluation of heart rate as a method for assessing moderate intensity physical activity. Med Sci Sports Exerc. 2000;32(9, Suppl):S465 S470. 22. Crouter SE, Schneider PL, Bassett DR, Jr. Springlevered versus piezo-electric pedometer accuracy in overweight and obese adults. Med Sci Sports Exerc. 2005;37(10):1673 1679. 23. McClain JJ, Craig CL, Sisson SB, Tudor-Locke C. Comparison of Lifecorder EX and ActiGraph accelerometers under free-living conditions. Appl Physiol Nutr Metab. 2007;32(4):753 761. 24. Kumahara H, Schutz Y, Ayabe M, et al. The use of uniaxial accelerometry for the assessment of physicalactivity-related energy expenditure: a validation study against whole-body indirect calorimetry. Br J Nutr. 2004;91(2):235 243.