Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 Mar 6.
Published in final edited form as: J Meas Phys Behav. 2021 Apr 1;4(2):118–125. doi: 10.1123/jmpb.2020-0034

Estimates of Physical Activity in Older Adults Using the ActiGraph Low-Frequency Extension Filter

Hilary Hicks 1, Alexandra Laffer 2, Kayla Meyer 3, Amber Watts 4,*
PMCID: PMC11884512  NIHMSID: NIHMS2038083  PMID: 40051680

Body-worn accelerometers have become an increasingly popular means of assessing physical activity in older adults. These devices offer numerous advantages over self-report measures, such as increased accuracy (e.g., Grimm, Swartz, Hart, Miller, & Strath, 2012), and the ability to continuously measure activity in a free-living environment, which is especially important in an older adult population known to spend less time in higher intensity activity and more time being sedentary (e.g., Harvey, Chastin, & Skelton, 2013).

ActiGraph introduced the low frequency extension (LFE) filter in 2009, which reduces the movement threshold to capture low acceleration activity in populations that are known to engage in low levels of structured physical activity (ActiGraph, 2018). This filter may enable users to capture a wider range of activities and was intended to yield a more accurate reflection of older adults’ activity levels, for whom much of their activity may occur at the low end of the spectrum.

Effects of the LFE filter have been investigated in previous ActiGraph models such as the GT1M (Feito, Garner, & Bassett, 2015), GT3X (Feito et al., 2015), GT3X+ (Cain, Conway, Adams, Husak, & Sallis, 2013; Feito, Hornbuckle, Reid, & Crouter, 2017; Tudor-Locke, Barreira, & Schuna, 2015; Wallén, Nero, Franzén, & Hagströmer, 2014), and most recently, the GT9X (Toth et al., 2018), ActiGraph’s newest device released in 2014. Findings from this literature indicate that estimates of physical activity outcomes depend on factors such as device model, placement (i.e., hip versus wrist), sample characteristics, and testing environment (i.e., laboratory versus free-living). However, it appears that the LFE filter overestimates common physical activity outcomes (e.g., step counts, percent time spent in physical activity of varying intensity) when used in a free-living environment (Cain et al., 2013; Feito et al., 2015; Feito et al., 2017; Toth et al., 2018; Tudor-Locke et al., 2015; Wallén et al., 2014), and appears to do so even in populations with low acceleration outputs (e.g., Feito et al., 2017), which are those for which the LFE filter was designed (ActiGraph, 2018). To our knowledge, only one prior study has looked at the effects of the LFE filter in the GT9X model, and this was conducted in a healthy young adult population (Toth et al., 2018). There appears to be high comparability between the GT9X and the GT3X+ (Clevenger, Pfeiffer, & Montoye, 2020; Montoye et al., 2018), which suggests similar comparability in the LFE’s impact on physical activity outcomes generated by these devices. However, few studies have investigated use of this filter in older adults, despite this being the very population for which it was designed. It appears that no studies have looked at use of the LFE specifically with the GT9X in older adults, and so the current study aims to add to this literature by investigating the impact of the LFE filter on physical activity outcomes captured via the wrist-worn GT9X in older adults in a free-living environment.

In addition to the LFE filter, researchers have demonstrated that estimates of physical activity can be influenced by device placement using the GT9X (e.g., Loprinzi & Smith, 2017; Toth et al., 2018). To date, studies of device placement in older adults have only used the GT3X+ (Kerr et al., 2017; Korpan, Schafer, Wilson, & Webber, 2015). With ActiGraph products, steps are calculated by a proprietary algorithm based solely on vertical axis activity, and manufacturers have recommended that users interested in this outcome wear the device on the hip to maximize accuracy (ActiGraph Support, personal communication, May 1, 2018). However, non-dominant wrist placement of the device is common in research, especially to maximize adherence to wearing the device consistently. Thus, the current study quantified step count estimates with and without the LFE filter in older adults when the GT9X device is worn on the non-dominant wrist.

Because use of the LFE filter in the GT9X has been shown to overestimate steps when worn on the wrist of healthy young adults in a free-living environment (Toth et al., 2018), we hypothesized that it would also generate high counts of this activity outcome when worn by older adults in a similar setting. The present study contributes to fill a gap in the extant literature by reporting on step count estimates using wrist placement, expecting that the step counts will be elevated compared to reports of those worn on the waist or hip in this population (e.g., Tedesco et al., 2019).

A final goal of the present study was to explore and identify participant characteristics that are most appropriate for use of the LFE filter. Because the filter was designed for use in highly sedentary populations, characteristics that are associated with sedentary behavior such as older age, higher body mass index (BMI), and lower engagement in moderate physical activity (e.g., Diaz et al., 2016) were used to predict the discrepancy in step count estimates generated with and without the LFE filter.

Purpose

The purpose of this study was to investigate the influence of the LFE filter on physical activity outcomes in the GT9X when worn on the non-dominant wrist of older adults. We quantified step counts calculated with and without the filter and identified participant characteristics most appropriate for its use.

Method

Participants

Participants were recruited from the University of Kansas Alzheimer’s Disease Center (KU-ADC) Registry, an ongoing longitudinal study that collects demographic, medical, psychological, and cognitive data annually to maintain a well-characterized registry of older adult participants with and without cognitive impairment due to Alzheimer’s disease. Participants are excluded from this Registry if they are less than 60 years old, exhibit significant depressive symptoms, or have untreated thyroid dysfunction that could account for cognitive symptoms. Participants are also excluded if they have significant visual or auditory impairment and any systemic illness that may prevent completion of the annual evaluations.

Participants whose data were used in these analyses were enrolled in an observational sub-study examining physical activity and sleep using wrist-worn accelerometry between September 2017 and August 2018. Participants were not included in analyses if they had cognitive impairment at the time of ActiGraph data collection (n = 15) or reported using a device to assist with ambulation (n = 3). Out of the remaining participants, 34 were randomly selected after a power analysis for the planned analyses indicated that this sample size would be sufficient to achieve power of 80% with a moderate effect size (.50).

Procedure

Starting in August of 2015, participants were invited once every other year at their annual visit to participate in an observational sub-study measuring activity levels and sleep in older adults. Participants were provided with informed consent, asked to self-report their hand dominance, and fitted with a watchband containing the GT9X at a regularly scheduled annual study visit. Participants were given written instructions directing them to wear the device at all times on their non-dominant wrist for a full 7 days. Participants were asked to keep the device fitted snugly to their wrists and refrain from removing it even in the shower or while sleeping. When initializing the device, the GT9X was programmed to display the time of day so that participants could use the device as a wristwatch. In addition to the GT9X, participants were provided with a sleep log and instructions on how to complete it. At the end of the 7 days, participants mailed the device and sleep log to the researchers in a prepaid padded envelope.

As part of their annual study visit, participants’ height and weight were recorded, which was used to calculate BMI. Race, education, and handedness were recorded upon initial enrollment in the KU-ADC Registry. The informed consent form and all study procedures were approved by the KU-ADC’s Institutional Review Board.

Physical Activity Monitors

ActiGraph GT9X Link accelerometers (Pensacola, FL) were used to calculate step count estimates and physical activity intensity levels of the research participants. The GT9X measures acceleration across three orthogonal planes (vertical—up and down, medio-lateral—side to side, and antero-posterior—front to back), which enables researchers to obtain information about activity counts across each of these three axes. Furthermore, the vector magnitude, a composite score of activity counts on all three axes (derived from the square root of the sum of the squares of the vertical, antero-posterior, and medio-lateral axes), can be calculated to provide additional information about wearers’ movement and activity.

Data Processing

The GT9X data were collected at a sample rate of 30 Hz, downloaded using ActiLife software (version 6.13.2, ActiGraph, LLC), and reintegrated into 60-second epochs. For each participant, raw data files were converted to activity counts per minute (CPM) and steps per minute both with and without the LFE filter enabled. Sleep data were processed using the Cole-Kripke algorithm (Cole, Kripke, Gruen, Mullaney, & Gillin, 1992), and ActiGraph-detected sleep periods were compared to sleep log records to denote sleep intervals. In the event of significant disagreement (> 30-minute difference) between ActiGraph and sleep logs, sleep intervals were manually calculated using a standardized protocol. Wear time was validated using the algorithm developed by Choi, Liu, Matthews, and Buchowski (2011), and a valid day was defined as a minimum of 10 hours of wear time. Participants with less than 4 days of valid accelerometry data were excluded from analyses. The cut points developed by Montoye and colleagues (2020) were used to classify physical activity by intensity level. According to this algorithm, < 2860 vector magnitude CPM is classified as sedentary, 2860 – 3940 as light, and ≥ 3941 as moderate to vigorous physical activity (MVPA). These cut points were developed for use in wrist-worn data, which makes them a more appropriate means of classifying activity intensity than alternative cut points developed for hip-worn data such as, the Freedson Adult Vector Magnitude (VM3) 2011 cut points (Sasaki, John, & Freedson, 2011).

Because the Montoye et al. (2020) cut points only recently became available to users, and it appears common practice to use the Freedson VM3 2011 cut points without regard for device placement, we opted to separately process our data with the Freedson VM3 2011 cut points to enable supplemental descriptive analyses that elucidate the limitations of these cut points. According to this algorithm, < 2690 vector magnitude CPM is classified as light, 2690 – 6166 as moderate, 6167 – 9642 as vigorous, and > 9642 as very vigorous. To date, there are no cut points specific to older adults.

Statistical Analyses

We correlated physical activity estimates (i.e., CPM, steps per minute, and percent time spent in different intensity levels) calculated with the LFE to their counterpart calculated without the filter. We used paired samples t-tests to evaluate the difference in CPM, average daily steps per minute, and percent time spent in different intensity levels of physical activity between files processed with and without the LFE filter. We used multiple regression analysis to predict the discrepancy in estimates of steps per minute generated with and without the LFE filter from participant age, BMI, and engagement in MVPA.

In an effort to clarify which participants are most appropriate for use of the LFE filter, we calculated a difference score between average daily steps per minute generated with and without the LFE filter to use as the outcome in two separate regression models using participant characteristics as predictors. Age, BMI, and percent time spent in MVPA were entered into a stepwise regression model predicting the difference score in number of steps per minute. We compared models where MVPA was calculated with the LFE enabled to models where MVPA was calculated with the LFE disabled. All statistical analyses were conducted with SPSS version 24 for Mac (SPSS, Inc.; Chicago, IL). The alpha level was set a priori at .05 (two tailed). The Benjamini-Hochberg false discovery rate procedure (Benjamini & Hochberg, 1995) was used to minimize chances of committing a Type I error.

Results

The final sample was 58.5% female, 91.2% Caucasian, and 94.1% right-hand dominant (see Table 1 for additional participant demographic characteristics). Physical activity data were normally distributed according to tests of skewness. Average daily wear time estimates generated with and without the LFE filter did not significantly differ, t (33) = 1.614, p = .116 (see Table 2).

Table 1.

Descriptive Statistics (N = 34)

Participant Characteristic Men (n = 14)
 Women (n = 20)
 Total
M SD M SD M (SD) Range
Age (years) 75.93 5.54 73.80 5.82 74.68 (5.72) 63 – 85
Education (years) 17.75 3.36 16.10 2.81 16.78 (3.11) 12 – 25
Height (inches)* 69.55 1.80 64.54 2.64
Weight (pounds)* 191.29 26.60 160.55 24.83
BMI 27.80 3.73 27.14 4.25
Open in a new tab
*

men and women differ at p ≤ .002

Note. BMI = body mass index

Table 2.

Average Daily Wear Time Estimates in Minutes (N = 34)

Filter Setting M SD Minimum Maximum
LFE On 962.21 51.61 875.00 1082.88
LFE Off 959.58 52.89 858.29 1082.88
Open in a new tab

Note. LFE = low frequency extension. Means do not significantly differ (p = .116).

Axes

Paired samples t-tests demonstrated significant differences between CPM on all three axes as well as the vector magnitude, a composite measure of all axes. Use of the LFE filter consistently yielded higher estimates (see Table 3).

Table 3.

Axis Counts Per Minute (N = 34)

LFE On
 LFE Off
Axis M SD M SD df t p
1 (vertical) 1024 304.48 956.72 291.70 33 21.704 < . 001
2 (antero-posterior) 1041.47 345.58 978.87 332.29 33 19.747 < . 001
3 (medial-lateral) 1333.98 427.09 1266.86 414.45 33 20.777 < . 001
Vector Magnitude 1984.04 612.73 1871.48 589.78 33 21.021 < . 001
Open in a new tab

Note. LFE = low frequency extension; Vector Magnitude = composite score of activity counts on all three axes

Steps

Participants took an average of 19,803 (SD = 5328) daily steps when the LFE was enabled and 10,138 (SD = 3421) daily steps when it was disabled. There was a significant positive correlation between average daily steps per minute generated with and without the LFE filter (r = .963, p < .001). Average daily steps per minute ranged from 10 to 33 when the LFE filter was on and from 5 to 19 when the LFE filter was off. A paired samples t-test demonstrated a significant difference between these estimates, t (33) = 25.79, p < .001, indicating that use of the LFE filter yielded approximately 90% higher average daily steps per minute (M = 21, SD = 1) than non-use (M = 11, SD = 1) (See Figure 1).

Figure 1.

Figure 1.

Open in a new tab

Mean steps per minute estimated with and without the low frequency extension (LFE) filter enabled. The difference is significant at p < .001. Error bars represent standard error of the mean.

Physical Intensity

Physical activity intensity estimates generated with the Montoye et al. (2020) cut points were used for the following analyses. There were significant positive correlations between physical activity calculated with and without the LFE filter for sedentary behavior (r = .998, p < .001), light physical activity (LPA; r = .995, p < .001), as well as MVPA (r = .998, p < .001). Paired samples t-tests revealed significant differences between physical activity estimates calculated with and without the LFE filter. Percentage of time spent in MVPA was significantly higher when calculated with the LFE filter (M = 5.03%, SD = 3.92%) than without (M = 4.27%, SD = 3.52%), t (33) = 9.259, p < .001. Similar findings were noted with time spent in LPA calculated with the LFE filter (M = 20.11%, SD = 7.38%) than without (M = 18.94%, SD = 7.45%), t (33) = 8.809, p < .001. In contrast, percentage of time spent in sedentary behavior was significantly lower when calculated with the LFE filter (M = 74.86%, SD = 10.23%) than without (M = 76.79%, SD = 9.97%), t (33) = −15.764, p < .001.

The model regressing discrepancy in steps per minute onto age, BMI, and MVPA calculated with the LFE enabled was significant, F (1, 32) = 8.792, p = .006, which was also the case for the model using MVPA calculated with the LFE filter disabled, F (1, 32) = 7.638, p = .009 (see Table 4). Age and BMI were removed from both models due to their insignificant predictive values. In contrast, percent time in MVPA significantly predicted step discrepancy in both models, indicating that as participants engage in more MVPA, the discrepancy between average daily steps per minute generated with the LFE filter on versus off increases (see Figure 2).

Table 4.

Regression Analysis Summary for Participant Characteristics Predicting Steps Per Minute Discrepancy (N = 34)

LFE On
 LFE Off
Predictors Standardized β t p Standardized β t p
Age (years) .167 1.023 .314 .165 .994 .328
BMI .079 .494 .624 .075 .461 .648
MVPA .464 2.965 .006 .439 2.764 .009
Open in a new tab

Note. LFE = low frequency extension; BMI = body mass index; MVPA = moderate to vigorous physical activity. LFE On: R2 = .216 (N = 34, p = .006), LFE Off: R2 = .193 (N = 34, p = .009).

Figure 2.

Figure 2.

Open in a new tab

Percent time spent in moderate to vigorous physical activity (MVPA) predicting the discrepancy in total daily average steps per minute generated with and without the low frequency extension (LFE) filter.

Supplemental descriptive analyses using estimates of physical activity intensity generated with the Freedson VM3 2011 cut points indicated that none of the participants engaged in very vigorous or vigorous activity. According to these cut points, participants spent ~70% of their time in LPA (LFE on: M = 71.67%, SD = 10.55%; LFE off: M = 73.68%, SD = 10.31%) and ~27% of their time in moderate physical activity (LFE on: M = 28.33%, SD = 10.55%; LFE off: M = 26.32%, SD = 10.31%).

Discussion

The purpose of the current study was to investigate the impact of the LFE filter on physical activity outcomes captured via the wrist-worn GT9X in older adults. To our knowledge, this was the first study to examine effects of this filter on physical activity estimates in this population (older adults) with regards to device (GT9X), device placement (wrist), and environment (free-living). Furthermore, it is the first study to use participant characteristics to predict for which participants the LFE filter may be more prone to overestimation.

LFE Impacts on Physical Activity Estimates

Our hypothesis that use of the LFE filter would significantly increase physical activity estimates was supported by our findings. Results indicated that enabling the LFE filter led to consistently higher CPM estimates across all three axes, therefore increasing the vector magnitude. In our sample, the average CPM was approximately 62 – 67 counts higher across each of the axes and 112 counts higher on the vector magnitude when the LFE filter was used. Because specific CPM are not reported in previous studies investigating the effects of the LFE filter in ActiGraph device models (Cain et al., 2013; Feito et al., 2015; Feito et al., 2017; Toth et al., 2018; Tudor-Locke et al., 2015; Wallén et al., 2014), it is not possible to directly compare these findings to related studies. However, the raw acceleration used to derive CPM is also used to calculate other commonly reported physical activity outcomes (e.g., step count estimates and physical activity intensities), which allows us to make some comparisons.

For example, the increased acceleration generated by the LFE filter that is reflected in the CPM directly impacts step count estimates, which are calculated from a proprietary algorithm based on acceleration captured across the vertical axis (axis 1). In our sample, we found that daily average steps per minute were increased by an average of 90% when the LFE filter was enabled. While somewhat more exaggerated, this significant discrepancy is consistent with findings by other studies that have had participants wear ActiGraph devices on the non-dominant wrist such as Toth et al. (2018) who found a 78% increase in step count estimates, and Tudor-Locke et al. (2015) who found a 67% increase. Researchers have also reported significant discrepancies in step counts generated with and without the LFE filter when the device is worn on the hip, which is what ActiGraph manufacturers recommend to those who have step count estimates as an outcome of interest. For example, researchers have reported discrepancies ranging from 61% (Feito et al., 2015) to 135% (Wallén et al., 2014) when participants wore ActiGraph devices on the hip in a free-living environment.

It is clear that the LFE filter yields significantly higher step count estimates measured in a free-living environment across several ActiGraph device models and populations, which is unsurprising as this is exactly what the LFE filter was designed to do. However, this becomes especially problematic when trying to compare step counts across studies that did and did not use the LFE filter, or do not report whether or not they used the filter. Linking actigraphy estimated step counts to objectively observed step counts is critical to understand any health outcomes that may be associated with a particular number of steps per day, such as the commonly used “10,000 steps a day” target that is widely mentioned.

The present study did not incorporate a criterion measure of step count estimates, so it is not possible to make the claim that the LFE filter overestimated step counts. However, based on the findings of other researchers, this is likely the case. For example, Toth and colleagues (2018) directly observed participant steps recorded by a GoPro video camera mounted on participants’ chests. This criterion measure allowed these researchers to demonstrate definitively that the LFE filter overestimated steps by 82% when they were recorded by the GT9X worn on the non-dominant wrist in a free-living environment. It is important to note that participants in Toth’s study were healthy young to middle-aged adults, populations known to engage in higher levels of physical activity than older adults. Without our own criterion measure with which to compare our findings, it is not possible to claim a comparable degree of overestimation in our older adult sample.

While it is not surprising that the LFE filter yields higher counts, the degree of difference between counts estimated with and without the filter is concerning. The significant discrepancy makes it difficult to interpret the step counts, because it is unclear whether the additional movement that is captured with the LFE filter is meaningful or predicts anything useful (e.g., health outcomes, fitness levels, etc.). When using the filter, one may appear to be sufficiently active (i.e., meeting the 10,000 steps a day goal), but this may not translate into actual health benefits. This leads us to question whether this filter is helpful for its users.

Similar to step count estimates, effects of increased CPM also extend to classifications of categories of physical activity levels based on activity cut points (e.g., sedentary, light, moderate-to-vigorous). We found that estimates for percent of day spent in MVPA differed, on average, by 0.76% such that use of the LFE filter led to higher estimates. Similar findings were noted for percent of day spent in LPA (i.e., estimates differed, on average, by 1.17%). In contrast, percent of day spent in sedentary behavior differed by 1.93%, such that the filter led to lower estimates. This finding suggests that more activity is being classified as LPA or MVPA and that less activity is being classified as sedentary behavior with the LFE filter enabled, suggesting that the same activity is being shifted from one category to the other.

These results align with those of other studies that have found that use of the LFE filter shifts activity out of the lowest intensity category into the next intensity classification (Cain et al., 2013; Feito et al., 2017; Wallén et al., 2014). It is important to note that the cut points (Freedson, Melanson, & Sirard, 1998; Matthews, 2005; Sasaki, John, & Freedson, 2011; Troiano et al., 2008) selected by researchers determine the lowest intensity category (i.e., sedentary or light), which can impact the interpretation of findings. Because the same activity is classified differently based on whether the LFE filter was enabled, it may be that cut point labels should be filter-specific to better reflect activity intensities.

Taken together, these findings suggest that the use of the LFE filter leads to significantly different physical activity outcomes captured via the wrist-worn GT9X in older adults compared to when the filter is disabled. The decision of whether or not to use this filter has important implications for physical activity outcomes including step count estimates and physical activity intensity classifications. With use of the filter, older adults may be classified as being more active than they actually are based on the amounts of activity needed to confer health benefits often studied. This result could have implications for the types of recommendations or guidelines for physical activity that are made for individuals within this population. While our results support the intent of the LFE filter (to lower the activity threshold to capture a wider range of activities), we have concerns about the extent to which the filter does so (e.g., the magnitude of the differences between the physical activity estimates calculated with and without the use of the filter) and the unequal distribution by which these estimates are inflated for some participants more than others.

Device Placement—Implications for Step Count Estimates

Our second aim of the study was to contribute to the extant literature by reporting on step count estimates using wrist placement in an older adult population. We hypothesized that the step counts in our sample would be elevated compared to step counts measured using waist or hip worn devices in other samples of older adults.

To our knowledge, there is no study that looks at the impact of device placement on GT9X step count estimates in older adults. Although limited in its interpretation, we can compare our findings calculated with the LFE filter disabled to those of Tedesco et al. (2019) who used the GT9X without the LFE filter to estimate step counts in a healthy older adult population that wore the device on their waists in a free-living environment. With comparable wear time estimates to our sample, we found that our participants averaged nearly 2000 steps more than those of Tedesco et al. This appears to support our hypothesis that wearing the device on the wrist results in higher step count estimates compared to the waist, although there certainly exist alternative explanations (e.g., our sample may be more active than that of Tedesco et al.).

Although there are no studies comparing the impact of device placement on step estimates using the GT9X, there are studies that look at this in other ActiGraph models (e.g., Chow, Thom, Wewege, Ward, & Parmenter, 2017; Mandigout et al., 2019). Results from these studies suggest that wrist-worn placement consistently yields higher step count estimates compared to waist-worn. Because we did not have our participants wear devices on both their waists and wrists, we cannot directly compare our findings to the above-cited studies.

Identifying Participant Characteristics for LFE Filter Use

The final aim of our study was to identify characteristics of participants that are most suitable for use of the LFE filter, which is marketed for use in sedentary populations. Three characteristics known to associate with sedentarism (i.e., older age, higher BMI, and low engagement in MVPA; Diaz et al., 2016) were used to predict the discrepancy between step count estimates generated with and without the LFE filter. Our correlational analyses indicated that step count estimates calculated with and without this filter were significantly associated with each other. However, our results suggest that the consistent difference between estimates disappears when specific participant characteristics are considered.

Percent time spent in MVPA (calculated either with or without the LFE filter) significantly predicted the difference score in steps per minute, which reflects a non-parallel relationship between these two variables (see Figure 2). As engagement in MVPA increased, the discrepancy between step count estimates calculated with and without the LFE filter widened. This finding suggests that use of the filter may be least appropriate in highly active individuals where the discrepancy in step counts is more pronounced. This supports ActiGraph’s original intent for the LFE filter, specifically to measure the physical activity of more sedentary populations. However, our findings also indicate that the LFE filter may not be universally appropriate for all older adults, a population that can exhibit variability in activity levels. Accordingly, we recommend that researchers be selective when choosing to use this filter and choose appropriate populations with which to compare results of activity counts.

Our findings indicated that age and BMI were not significant predictors of the discrepancy in step count estimates generated with and without the LFE filter. Both age and BMI are known to associate with engagement in sedentary behavior (e.g., Diaz et al., 2016), so this finding was unexpected. One possible explanation for the lack of significant findings could be related to our study’s exclusion criteria. Even though our participants ranged in age from 63 to 85 years and 67.6% had a BMI that could be classified as overweight or obese, we excluded participants with any systemic illness that could prevent completion of the annual evaluation, cognitive impairment, or who used a device to assist with ambulation. Therefore, our participants were much less likely to exhibit age and BMI-related physical factors known to influence physical activity resulting in a more homogeneous sample.

Supplementary Findings

Descriptive analyses using physical activity intensity estimates generated with the Freedson VM3 2011 cut points yielded estimates that are highly discordant with what is known about typical older adult physical activity patterns (e.g., Harvey, Chastin, & Skelton, 2013). Whether with or without the LFE filter enabled, our older adults appeared to engage in moderate physical activity for ~27% of the day, which clearly reflects the inappropriate use of these hip-worn cut points in wrist-worn data. Our findings indicate that researchers using accelerometry to measure physical activity intensity should carefully select cut points based on device placement.

Limitations

The findings of our study are limited by several factors. First, our sample consisted primarily of highly educated, Caucasian older adults. Participant characteristics such as race and educational attainment are known to associate with physical activity (He & Baker, 2005), thus our generalizability is limited. Furthermore, we sampled individuals who, at the time of data collection, lived in the greater metro-area and surrounding rural communities of a mid-size city on the border of two midwestern states. While we cannot comment on the specifics of their free-living environments, they are likely to vary from those of older adults who live in more urban settings or places with greater walkability. Because neighborhood characteristics such as these are known to influence physical activity patterns of older adults (Barnett et al., 2017), this further limits the generalizability of our findings.

Second, our sample consisted of individuals who had already volunteered to participate in a larger ongoing research study before being recruited for our sub-study. This could imply the presence of motivational factors that may make our sample unique. Participants like ours are often more educated, health-conscious, and physically active than the average older adult (Fry et al., 2017), so it is likely that our findings are not representative.

Future Directions

Future research investigating the influence of the LFE filter on accelerometry-assessed physical activity outcomes should aim to diversify study samples by including a larger and broader set of individuals with a range of demographic characteristics. Furthermore, researchers interested in older adult samples could incorporate participants with specific age-related diagnoses known to influence physical activity outcomes (Tudor-Locke, Washington, & Hart, 2009).

Future studies could significantly benefit from utilizing a criterion measure to definitively establish the accuracy and degree of overestimation resulting from the LFE filter. A criterion measure could also be used to evaluate the difference between physical activity outcomes assessed via hip and wrist-worn devices. To our knowledge, there are currently no studies that have used the GT9X in an older adult sample to analyze this difference. Finally, a criterion measure could facilitate the development of conversion algorithms to enable researchers to compare datasets developed using different devices, placement, and software settings.

Our findings have important implications for health outcomes and physical activity guidelines. A goal of 10,000 steps a day is commonly touted among the general public, without regard to how this physical activity outcome is calculated or how its achievement varies between devices and device placement. Numerous studies have demonstrated a positive association between step counts and physical health in older adults (for a review, see Tudor Locke et al., 2011). However, in these studies, little consideration is given to how step counts are calculated. We recommend that future studies strive to explore the relationship between actigraphy-measured physical activity estimates and health outcomes (e.g., cardiorespiratory fitness) to clarify the utility of step count goals and perhaps titrate goals based on specific models, software settings, and device placement. Surely a wearable device that enables its users to track their daily physical activity can benefit those who attempt to improve their physical fitness by increasing activity compared to personal baselines. However, our findings suggest that we should take care when offering specific step count goals that are so easily influenced by software settings.

Conclusion

To the best of our knowledge, the current study was the first to assess the impact of the LFE filter on physical activity outcomes captured via the wrist-worn GT9X in older adults in a free-living environment. Our findings suggest that the LFE filter yields significantly higher step count estimates and causes more activity to be classified as MVPA or LPA compared to sedentary. Findings from our exploratory analysis support ActiGraph’s initial intent for the LFE filter to be used in populations with low activity output but suggest that this filter becomes increasingly less appropriate in more active older individuals. Researchers using ActiGraph devices to estimate physical activity outcomes in a free-living environment should be aware of and consider the impact of the LFE filter on outcomes of interest including step counts.

Contributor Information

Hilary Hicks, The University of Kansas.

Alexandra Laffer, The University of Kansas.

Kayla Meyer, The University of Kansas Alzheimer’s Disease Center.

Amber Watts, The University of Kansas and The University of Kansas Alzheimer’s Disease Center.

References

  1. ActiGraph (2018, November 8). Low frequency extension explained. [Online support post]. Retrieved from https://actigraphcorp.force.com/support/s/article/Low-Frequency-Extension-Explained
  2. Barnett DW, Barnett A, Nathan A, Van Cauwenberg J, Cerin E, & Council on Environmental and Physical Activity (CEPA)-Older Adults working group. (2017). Built environmental correlates of older adults’ total physical activity and walking: A systematic review and meta-analysis. International Journal of Behavioral Nutrition and Physical Activity, 14: 103. doi: 10.1186/s12966-017-0558-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Benjamini Y & Hochberg Y (1995). Controlling the false discovery rate: A practical and powerful approach to multiple testing. Journal of the Royal Statistical Society: Series B (Methodological), 57, 289–300. 10.1111/j.2517-6161.1995.tb02031.x [DOI] [Google Scholar]
  4. Cain KL, Conway TL, Adams MA, Husak LE, & Sallis JF (2013). Comparison of older and newer generations of ActiGraph accelerometers with the normal filter and the low frequency extension. International Journal of Behavioral Nutrition and Physical Activity, 10, 1–6. http://www.ijbnpa.org/content/10/1/51 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Choi L, Liu Z, Matthews CE, & Buchowski MS (2011). Validation of accelerometer wear and nonwear time classification algorithm. Medicine & Science in Sports & Exercise, 43, 357–364. doi: 10.1249/MSS.0b013e3181ed61a3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Chow JJ, Thom JM, Wewege MA, Ward RE, & Parmenter BJ (2017). Accuracy of step count measured by physical activity monitors: The effect of gait speed and anatomical placement site. Gait Posture, 57, 199–203. doi: 10.1016/j.gaitpost.2017.06.012 [DOI] [PubMed] [Google Scholar]
  7. Clevenger KA, Pfeiffer KA, & Montoye AHK (2020). Cross-generational comparability of hip- and wrist-worn ActiGraph GT3X+, wGT3X-BT, and GT9X accelerometers during free-living in adults. Journal of Sports Sciences, 1–9. doi: 10.1080/02640414.2020.1801320 [DOI] [PubMed] [Google Scholar]
  8. Cole RJ, Kripke DF, Gruen W, Mullaney DJ, & Gillin JC (1992) Automatic sleep/wake identification from wrist activity. Sleep, 15, 461–469. doi: 10.1093/sleep/15.5.461 [DOI] [PubMed] [Google Scholar]
  9. Diaz KM, Howard VJ, Hutto B, Colabianchi N, Vena JE, Blair SN, & Hooker SP (2016). Patterns of sedentary behavior in US middle-age and older adults: The REGARDS study. Medicine & Science in Sports & Exercise, 48, 430–438. doi: 10.1249/MSS.0000000000000792 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Feito Y, Garner HR, & Bassett DR (2015). Evaluation of ActiGraph’s low-frequency filter in laboratory and free-living environments. Medicine & Science in Sports & Exercise, 211–217. doi: 10.1249/MSS.0000000000000395 [DOI] [PubMed] [Google Scholar]
  11. Feito Y, Hornbuckle LM, Reid LA, & Crouter SE (2017). Effect of ActiGraph’s low frequency extension for estimating steps and physical activity intensity. PloS ONE 12, e0188242. 10.1371/journal.pone.0188242 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Freedson PS, Melanson E, & Sirard J (1998). Calibration of the Computer Science and Applications, Inc. accelerometer. Medicine & Science in Sports & Exercise, 30, 777–781. [DOI] [PubMed] [Google Scholar]
  13. Fry A, Littlejohns TJ, Sudlow C, Doherty N, Adamska L, Sprosen T, Collins R, & Allen NE (2017). Comparison of sociodemographic and health-related characteristics of UK biobank participants with those of the general population. American Journal of Epidemiology, 186, 1026–1034. doi: 10.1093/aje/kwx246 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Grimm EK, Swartz AM, Hart T, Miller NE, & Strath SJ (2012). Comparison of the IPAQ-Short Form and accelerometry predictions of physical activity in older adults. Journal of Aging and Physical Activity, 20, 64–79. [DOI] [PubMed] [Google Scholar]
  15. Harvey JA, Chastin SFM, & Skelton DA (2013). Prevalence of sedentary behavior in older adults: A systematic review. International Journal of Environmental Research and Public Health, 10(12), 6645–6661. 10.3390/ijerph10126645 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. He XZ & Baker DW (2005). Differences in leisure-time, household, and work-related physical activity by race, ethnicity, and education. Journal of General Internal Medicine, 20, 259–266. doi: 10.1111/j.1525-1497.2005.40198.x [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Kerr J, Marinac CR, Ellis K, Godbole S, Hipp A, Glanz K, Mitchell J, Laden F, James P, & Berrigan D (2017). Comparison of accelerometry methods for estimating physical activity. Medicine & Science in Sports & Exercise, 49, 617–624. doi: 10.1249/MSS.0000000000001124 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Korpan SM, Schafer JL, Wilson KCS, & Webber SC (2015). Effect of ActiGraph GT3X+ position and algorithm choice on step count accuracy in older adults. Journal of Aging and Physical Activity, 23, 377–382. 10.1123/japa.2014-0033 [DOI] [PubMed] [Google Scholar]
  19. Loprinzi PD & Smith B (2017). Comparison between wrist-worn and waist-worn accelerometry. Journal of Physical Activity and Health, 14(7), 539–545. doi: 10.1123/jpah.2016-0211 [DOI] [PubMed] [Google Scholar]
  20. Mandigout S, Lacroix J, Perrochon A, Svoboda Z, Aubourg T, & Vuillerme N (2019). Comparison of step count assessed using wrist- and hip-worn Actigraph GT3X in free-living conditions in young and older adults. Frontiers in Medicine, 6, 252. doi: 10.3389/fmed.2019.00252 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Matthews CE (2005). Calibration of accelerometer output for adults. Medicine & Science in Sports & Exercise, 37, S512–S522. doi: 10.1249/01.mss.0000185659.11982.3d [DOI] [PubMed] [Google Scholar]
  22. Montoye AHK, Clevenger KA, Pfeiffer KA, Nelson MB, Bock JM, Imboden MT, & Kaminsky LA (2020). Development of cut-points for determining activity intensity from a wrist-worn ActiGraph accelerometer in free-living adults. Journal of Sports Sciences, 1–10. doi: 10.1080/02640414.2020.1794244 [DOI] [PubMed] [Google Scholar]
  23. Montoye AHK, Nelson MB, Bock JM, Imboden MT, Kaminsky LA, Mackintosh KA, McNarry MA, & Pfeiffer KA (2018). Raw and count data comparability of hip-worn ActiGraph GT3X+ and Link accelerometers. Medicine & Science in Sports & Exercise, 50, 1103–1112. doi: 10.1249/MSS.0000000000001534 [DOI] [PubMed] [Google Scholar]
  24. Rhudy MB, Dreisbach SB, Moran MD, Ruggiero MJ, & Veerabhadrappa P (2020). Cut points of the ActiGraph GT9X for moderate and vigorous intensity physical activity at four different wear locations. Journal of Sports Sciences, 38, 503–510. doi: 10.1080/02640414.2019.1707956 [DOI] [PubMed] [Google Scholar]
  25. Sasaki JE, John D, & Freedson PS (2011). Validation and comparison of ActiGraph activity monitors. Journal of Science and Medicine in Sport, 14, 411–416. doi: 10.1016/j.jsams.2011.04.003 [DOI] [PubMed] [Google Scholar]
  26. Tedesco S, Sica M, Ancillao A, Timmons S, Barton J, & O’Flynn B (2019). Validity evaluation of the Fitbit Charge2 and the Garmin Vivosmart HR+ in free-living environments in an older adult cohort. Journal of Medical Internet Research Mhealth Uhealth, 7(6), e13084. doi: 10.2196/13084 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Toth LP, Park S, Springer CM, Feyerabend MD, Steeves JA, & Bassett DR (2018). Video-recorded validation of wearable step counters under free-living conditions. Medicine & Science in Sports & Exercise, 1315–1322. doi: 10.1249/MSS.0000000000001569 [DOI] [PubMed] [Google Scholar]
  28. Troiano RP, Berrigan D, Dodd KW, Masse LC, Tilert T, & McDowell M (2008). Physical activity in the United States measured by accelerometer. Medicine & Science in Sports & Exercise, 40, 181–188. 10.1249/mss.0b013e31815a51b3 [DOI] [PubMed] [Google Scholar]
  29. Tudor-Locke C, Barreira TV, & Schuna JM (2015). Comparison of step outputs for waist and wrist accelerometer attachment sites. Medicine & Science in Sports & Exercise, 839–842. doi: 10.1249/MSS.0000000000000476 [DOI] [PubMed] [Google Scholar]
  30. Tudor-Locke C, Craig CL, Aoyagi Y, Bell RC, Croteau KA, De Bourdeaudhuij I, ⋯ Blair SN (2011). How many steps/day are enough? For older adults and special populations. International Journal of Behavioral Nutrition and Physical Activity, 8, 80. doi: 10.1186/1479-5868-8-80 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Tudor-Locke C, Washington TL, & Hart TL (2009). Expected values for steps/day in special populations. Preventive Medicine, 49(1), 3–11. doi: 10.1016/j.ypmed.2009.04.012 [DOI] [PubMed] [Google Scholar]
  32. Wallén MB, Nero H, Franzén E, & Hagströmer M (2014). Comparison of two accelerometer filter settings in individuals with Parkinson’s disease. Physiological Measurement, 35, 2287–2296. doi: 10.1088/0967-3334/35/11/2287 [DOI] [PubMed] [Google Scholar]

RESOURCES