This paper is an invited commentary based on a keynote presentation at the 8th International Conference on Ambulatory Monitoring of Physical Activity and Movement (ICAMPAM) held in June 2022. The commentary reflects on the opportunities and the challenges afforded by the increasing use of devices to measure physical behaviors in epidemiologic studies.
Many decades of research have yielded convincing data showing that physical activity is essential for health and well-being (2018 Physical Activity Guidelines Advisory Committee, 2018). More recently, sedentary behavior—independent of low physical activity—also has been related to higher mortality rates, higher rates of noncommunicable diseases, and worse cardiometabolic risk profiles (Buffey et al., 2022; Ekelund et al., 2016, 2019; Katzmarzyk et al., 2019). These data have led to the development of national and international guidelines, recent examples of which include the 2018 U.S. Physical Activity Guidelines (U.S. Department of Health and Human Services, 2018) and the 2020 WHO Guidelines on Physical Activity and Sedentary Behaviour (World Health Organization, 2020).
The evidence that underpins these guidelines derives primarily from observational epidemiologic studies, using self-reported behaviors rather than randomized clinical trials. We anticipate, for reasons discussed below, that observational epidemiologic studies will continue to form the backbone of evidence for future iterations of guidelines, particularly for long-term health outcomes such as cardiovascular disease (CVD) and cancer—leading causes of death in the United States and worldwide (American Cancer Society, 2018). A major limitation of self-reports is the tendency for overestimation of physical activity and underestimation of sedentary behavior (Jefferis et al., 2016; Ndahimana & Kim, 2017). Thus, the recent proliferation of data from studies using device-assessed rather than self-reported behaviors (Paluch et al., 2022; Troiano et al., 2014) has been a welcome advance in the field. Of note is that published findings using these data may emanate from studies that were initiated years ago, employing devices which are now dated (Evenson et al., in press). What are some implications of using dated device data, and how can their usefulness be maximized for informing future public health guidelines?
In this paper, we provide reasons for why much of the evidence base in the field has come from observational studies rather than clinical trials. We next discuss why these epidemiologic studies can take a long time to mature, resulting in data that were collected using “dated” devices and technologies. What are some implications of using such dated device data, and how can their usefulness be maximized for informing future public health guidelines? We then use the Women’s Health Study (WHS; Lee et al., 2018b) to illustrate some translational aspects related to these questions.
Observational Epidemiologic Study Versus Randomized Controlled Trial in Physical Activity and Health Research
Observational epidemiologic studies have provided a large portion of the evidence base underlying the 2018 U.S. physical activity and sedentary behavior guidelines (2018 Physical Activity Guidelines Advisory Committee, 2018). For example, the expert panel reviewing data on all-cause mortality for the 2018 U.S. guidelines cited findings from 12 reviews of observational studies (2018 Physical Activity Guidelines Advisory Committee, 2018), all of which employed self-reports of physical activity. It is noteworthy that the 2018 guidelines, which represent the second edition of U.S. federal guidelines, relied almost exclusively on systematic reviews and pooled or meta-analyses, as opposed to primary studies, as had been the case with the first set of U.S. guidelines in 2008 (Physical Activity Guidelines Committee, 2008). This was due to the large body of evidence that had accumulated since the 2008 guidelines, making a review of primary studies onerous and impractical. In order to combine data meaningfully across studies, the physical activity measures need to be comparable; the relevance of comparability for device data is discussed below.
Why is there a dearth of data from randomized controlled trials, which are conventionally considered the best research study design since well-designed and conducted trials yield unbiased findings? There are several reasons for this. First, for randomized controlled trials, a crucial requirement for obtaining high-quality results is maintaining good compliance to the intervention(s). Imagine a simple trial conducted in an inactive population, where half the participants are randomized to continue their inactive habits (control); the other half, to exercise (intervention). In an extreme scenario, suppose that compliance was so compromised such that half the controls took up exercise while half the intervention participants stopped exercise. This would result in two groups with identical exercise behavior, yielding useless trial results for examining the effect of the exercise.
Maintaining compliance in trials over a long duration—necessary when studying health outcomes which take a long duration to develop, such as mortality, CVD, or cancer—is challenging and burdensome for researchers, as is retention of participants. Both erode over time. As an illustration, in the Generation 100 study, more than 1,500 participants aged 70–77 years were randomized to three arms to investigate effects on all-cause mortality over 5 years: control arm, moderate-intensity continuous training arm, or high-intensity interval training arm (Stensvold et al., 2020). Compliance with the assigned intervention in the three arms was 78%, 63%, and 50%, respectively, at Year 1. By Year 5, this had declined to 69%, 51%, and 47%, respectively. Meanwhile, crossover by control participants for doing high-intensity interval training was 23% at Year 1 and 18% at Year 5. Thus, the groups became more alike as the trial went on. Retention of participants also declined over time: The dropout rates at Year 1 were 8%, 15%, and 19% in the three arms, respectively; at Year 5, they were 20%, 26%, and 33%, respectively. As expected, erosion of compliance and retention were greater among participants in the more “difficult” exercise arms. We emphasize that this example is not meant to criticize the investigators—indeed, they are to be commended for undertaking this ambitious and important study—but, rather, to illustrate the difficulty of carrying out exercise trials of long duration, that is, the longer the trial duration, the lower the compliance and retention rates.
To address the limitation of compliance and retention, investigators can study surrogate endpoints, or biomarkers of disease, such as cardiometabolic risk factors instead of CVD itself. There is a large body of data from trials of physical activity or sedentary behavior and surrogate endpoints (2018 Physical Activity Guidelines Advisory Committee, 2018; World Health Organization, 2020) which add valuable data to the evidence base. However, empirical data on the actual disease endpoint are also needed since beneficial changes in biomarkers may not always translate to reductions in disease rates. One example is the Look AHEAD trial, a randomized controlled trial of intensive lifestyle intervention versus usual care among patients with Type 2 diabetes and overweight or obesity (Look Ahead Research Group et al., 2013). While favorable changes in weight, waist circumference, physical fitness, and glycated hemoglobin occurred with intervention, this did not translate to any difference in CVD events.
A second possible explanation for the dearth of physical activity trials and long-term outcomes is that trials typically require greater participant burden compared with observational studies; thus, participants of trials tend to be highly selected and healthier resulting in limited statistical power. Using the Generation 100 study again, less than one quarter (23.5%) of eligible persons agreed to participate in the trial. Investigators had anticipated a 10% cumulative mortality rate over 5 years; however, the observed rate was less than half that expected (4.6%), likely because of self-selection of healthier individuals who were willing to participate.
In contrast, observational epidemiologic studies where participants are observed carrying out their typical habits can continue for many years. As an illustration, in a pooled analysis investigating the dose–response relationship between physical activity and all-cause mortality cited by the 2018 U.S. guideline expert panel (Arem et al., 2015), the median follow-up was more than 14 years. And, while the Generation 100 study included three arms or “doses” of exercise, this pooled analysis examined seven “doses” of self-reported physical activity, ranging from no leisure-time physical activity of moderate-to-vigorous intensity to a dose corresponding to 10 times or more of the recommended level.
Based on the discussion above, it is evident that both randomized trials and observational studies are needed to further our understanding of the relations between physical activity or sedentary behavior and good health and functioning. Neither study design is clearly superior (Westreich et al., 2019); each has unique strengths for physical activity research (summarized in Table 1), with both providing complementary data.
Summary of Typical Strengths: Randomized Clinical Trial Versus Observational Study Design in Physical Activity Research
| Randomized clinical trial | Observational study | |
|---|---|---|
| Less prone to bias | X | |
| Longer study duration | X | |
| Less selectivity of participants (greater generalizability) | X | |
| Feasibility for studying many “doses” of physical activity and/or sedentary behavior | X | |
| Reflects real life | X |
Note. X indicates the area(s) where each study design has the advantage.
Slow Pace of Discovery in Observational Epidemiologic Research Versus Rapid Pace of Advancement in Device Technology
The two leading causes of death in the United States, as well as globally, are CVD and cancer (American Cancer Society, 2018). These diseases take a long time to develop, and risk factors even during childhood predict future events (Jacobs et al., 2022). Thus, when investigating the relations of physical activity or sedentary behavior to these and other noncommunicable diseases, research participants need to be followed for sufficient length of time in order for incident cases of the disease to develop and to accrue in adequate numbers, such that statistical analyses are adequately powered.
Consequently, observational studies are better suited than randomized trials for examining research questions related to noncommunicable diseases, particularly in populations at usual risk (in high-risk populations, it is more feasible to conduct trials since rates of disease can be far higher). Even when trials are conducted, they typically last only a (relatively) short duration for the reasons discussed above, after which extended observational follow-up of trial participants can occur. For example, the first trial to investigate lifestyle (diet and exercise) modification for preventing Type 2 diabetes among individuals with impaired glucose tolerance was the Da Qing trial, which began in 1986 and lasted 6 years (Pan et al., 1997). Following the scheduled end of the trial, participants continued to be followed observationally for over 20 years, to allow investigation of lifestyle intervention on risks of other noncommunicable diseases and mortality with sufficient power (Gong et al., 2019).
A further reason for desiring long follow-up in observational studies is to mitigate the effect of reverse causation bias. This bias refers to when physical activity is observed to be related to lower risk of, say, disease X; but, rather than physical activity causing less disease X, the reverse may be occurring. Individuals, ostensibly free from disease X, but who may have the disease without symptoms, may feel unwell and thus become inactive. In this situation, the reverse causation bias refers to that disease X caused inactivity, rather than inactivity caused disease X.
Such a bias operates more during early follow-up, so that with longer follow-up the bias is diluted. This can be illustrated by an investigation from Matthews et al. (2020) that examined device-assessed physical activity and all-cause mortality in National Health and Nutrition Examination Survey (NHANES). Investigators assessed differing lengths of follow-up, from 0–2 to 0–12 years, and found that bias from reverse causation appears to minimize after an approximately 6-year length of follow-up.
This discussion helps explains why observational epidemiologic studies can take years to mature and yield meaningful findings. This is contrasted with the rapid pace at which technology advances. The WHS, which we discuss below, collected device-assessed measures of physical activity and sedentary behavior from 2011 to 2015 (Lee et al., 2018b). By the standards of wearable devices, this may be considered an ancient period—the Jawbone UP and Nike FuelBand, devices no longer sold or supported, were released early during that period (Danova, 2015). Indeed, the first Apple Watch was released at the end of that period, many versions prior to the Series 8 that is expected to launch in 2022. Yet, the WHS is only recently beginning to accrue meaningful numbers of long-term health outcomes that can be investigated in relation to device-measured physical activity and sedentary behavior (Lee et al., 2018a, 2019). Can their “dated” (by necessity) measures be useful?
Maximizing the Utility and Comparability of Accelerometer Data From Observational Studies Initiated Years Ago: WHS Example
To quote Intille et al. (2012): “The only certainty about the future of activity-sensing technologies is that researchers must anticipate and plan for change.” This section uses the WHS to illustrate some ways in which researchers can plan for the reality that new devices and technologies will be used in research on physical activity and sedentary behavior over time. A key requirement is to have the ability to combine data from “old” studies, such as the WHS, with more recent and future data for meta-analytic purposes. As discussed above, guidelines today rely heavily on reviews and meta-analyses because of the large body of evidence currently available. Additionally, to be able to validly examine trends in physical behaviors over time—nationally and internationally—surveillance data from devices used over the years must be comparable.
Briefly, the WHS was a randomized controlled trial conducted from 1992 to 2004 that tested low-dose aspirin and vitamin E for the prevention of cancer and CVD among 39,876 women aged ≥45 years, enrolled from throughout the United States (Cook et al., 2005; Lee et al., 2005; Ridker et al., 2005). After an average follow-up of 10 years, the trial ended as scheduled. At trial completion, women were invited to participate in an observational follow-up study, and 89% of those alive (n = 33,682) consented to do so. Women currently continue to be followed with an annual health questionnaire (including questions on physical activity and sedentary behavior); self-reports of cancer and CVD are adjudicated by study physicians using medical records and death certificates.
Between 2011 and 2015, an ancillary study was conducted where eligible and willing women were asked to wear an ActiGraph GT3X+ on the hip for 7 days during all waking hours (Lee et al., 2018b). The devices were set to record data at 30 Hz. A total of 17,708 women wore and returned their devices, and data were downloaded from 17,466 devices (device failure occurred in the remaining 242). Participants were from throughout the United States, including Alaska, Hawaii, and Puerto Rico; their mean age was 71.5 (SD 5.7) years (Evenson et al., 2021). Almost all women (17,062; 97.7%) had at least one adherent day of wear (≥10 hr) and 16,742 (95.9%) had at least four adherent days. Among women with at least four adherent days, the mean wear time on adherent days was 14.9 (1.3) hr. Using vector magnitude cut points calibrated for older women (Evenson et al., 2015), the mean time spent in sedentary behavior was 510.6 (98.8) min/day; moderate-to-vigorous physical activity, 91.9 (45.4) min/day.
One framework to consider the data collected is shown in Figure 1 (adapted from Bai et al., 2016). The raw data collected by accelerometers can be summarized across axes and time into summary measures that have been utilized by most studies of exposure health outcomes to date. Commonly, the summary measure used in such studies has been the activity count per time epoch, for example, counts per minute or “CPM.” This metric has been very useful for investigators seeking to describe dose–response relationships (e.g., Ekelund et al., 2019). However, “counts” has no precise meaning. Counts are an aggregate measure of both amount and intensity of activity performed over a specific time period. When counts are provided, each device manufacturer generates counts using its own proprietary algorithm (or what has largely been proprietary), making use of different transducers, amplifiers, sampling rates, signal filters, and so forth (John et al., 2019). Thus, counts are not comparable across devices. For example, in the meta-analysis referenced above that investigated the dose–response relations of physical activity and sedentary behavior with all-cause mortality (Ekelund et al., 2019), moderate-to-vigorous physical activity was defined as ≥1,952 CPM on the vertical axis for studies using the ActiGraph device and >1,535 CPM on the vertical axis for the Actical device (Phillips Respironics); all studies used hip-worn devices. Different wear locations—such as hip versus wrist wear—result in different counts generated, further contributing to the lack of standardization.
—A framework for considering data collected from studies using accelerometers to measure physical behavior. Note. MIMS-unit = Monitor-Independent Movement Summary unit; AAI = Accelerometer Activity Index. Adapted from “An Activity Index for Raw Accelerometry Data and Its Comparison With Other Activity Metrics,” by J. Bai, C. Di, L. Xiao, K.R. Evenson, A.Z. LaCroix, C.M. Crainiceanu, and D.M. Buchner, 2016, PLoS One, 11(8), Article e0160644.
Citation: Journal for the Measurement of Physical Behaviour 6, 1; 10.1123/jmpb.2022-0035
To overcome this lack of standardization, several methods have been proposed that yield summary measures comparable across devices from different manufacturers and devices of the same manufacturer over time, using high-resolution raw accelerometry data which are now available from many research-grade accelerometers. Below, we discuss some alternate metrics (a selected list) using open-source methods that are being considered in the WHS.
John et al. (2019) have proposed the Monitor-Independent Movement Summary unit (MIMS-unit; “R Package MIMS-Unit Algorithm,” 2022) derived from raw acceleration data. Accelerometer data from the U.S. NHANES, 2011–2014, have been released for public use in MIMS-units in contrast to data from previous cycles which were released as counts. A key consideration behind the MIMS-unit algorithm is that different devices collect data within different dynamic ranges (generally ±2 to 16 g; most human activities are within ±2 g; John et al., 2019). Additionally, device sampling rates can vary (e.g., WHS collected data at 30 Hz; NHANES, 80 Hz; most human movements are 0.3–5 Hz; John et al., 2019). The algorithm extrapolates data falling outside the device dynamic range and interpolates data collected at lower sampling rates to “fill in” the gaps. A band-pass filter is then applied to filter out signal artifacts (i.e., “noise” rather than true movement). John et al. showed that the MIMS-unit was similar across simulated data, representing different devices collecting data at varying dynamic ranges (2–16 g) and sampling rates (20–100 Hz).
Although this represents an advance toward comparability, more work is needed to translate the metric to an interpretable concept for the public. What does, for example, 1,279 MIMS-units/day less—shown to be associated with hearing loss in NHANES (Martinez-Amezcua et al., 2022)—mean? A further complication is that wear location matters: While MIMS-units can be generated for devices at any wear location, they are similar across devices only when worn at the same location. That is, they will not be comparable for two devices if one is worn on the wrist and the other on the hip.
Bai et al. (2014, 2016) proposed a summary measure, the Activity Index (“R Package ActivityIndex,” 2021), which also has been called the Accelerometer Activity Index (AAI) by Wang et al. (2022). The AAI is a measure of the variability (or intensity) of the raw acceleration signal, relative to the signal variability when the device is not perceived to be moving (“rest”). It is calculated as the SD of the raw accelerometer signal during activity divided by the SD of the signal at rest. Thus, the AAI will have a similar interpretation across devices. When Wang et al. compared AAI versus counts per 15-s epoch, AAI predicted activity intensity better than did counts among older women (R2 = .74 vs. .54).
Another summary measure, the Mean Amplitude Deviation (MAD; “R Package GGIR,” 2022) has been proposed by Vähä-Ypyä et al. (2015). This uses a similar concept to AAI, with MAD calculated as the typical distance of acceleration signal data points about the signal mean. Vähä-Ypyä et al. showed similar MAD measures for different sedentary activities, as well as walking/running at different speed across three device brands, worn on the hip.
A final summary measure that we will mention was proposed for measuring human physical activity by van Hees et al. (2013). This is the Euclidean Norm, or vector magnitude, which has been used more extensively than the other metrics discussed above. Raw acceleration signals contain three basic components: an individual’s movement (the component of interest), gravity, and “noise.” The Euclidean Norm Minus One (1 g, or one Earth standard gravitational unit, to remove the gravity component; “R Package GGIR,” 2022) has previously been used to investigate data from the UK Biobank (e.g., Ramakrishnan et al., 2021).
It is worth emphasizing that as with the MIMS-units, all these other metrics are comparable across devices but only for the same wear location, not for different wear locations.
We compared these summary measures in a pilot study of 100 women randomly sampled from the WHS, for illustrative purposes. Using raw acceleration signals, we computed the data in 5-s epochs for each metric and summed the 5-s metric to the day level (Table 2). We then calculated Spearman correlations between combinations of metrics within each woman (since each person contributed multiple data points) and averaged person-specific correlations across all women. At the day level, MIMS-units, AAI, and MAD were highly correlated among each other (r = .79−.94); Euclidean Norm Minus One was less correlated to the other summary measures (r = .57−.68). Counts were only moderately correlated to the other summary measures (r = .33−.51).
Spearman Correlation Coefficients Among Various Summary Measures From Accelerometer Data, Women’s Health Studya
| Counts | MIMS-unit | AAI | MAD | ENMO | |
|---|---|---|---|---|---|
| Counts | — | .51 | .46 | .40 | .33 |
| MIMS-unit | — | .93 | .79 | .57 | |
| AAI | — | .89 | .63 | ||
| MAD | — | .68 | |||
| ENMO | — |
Note. MIMS-unit = Monitor-Independent Movement Summary unit; AAI = Accelerometer Activity Index; MAD = Mean Amplitude Deviation; ENMO = Euclidean Norm Minus One.
aFrom pilot study of 100 randomly sample women; data were computed in 5-s epochs and summed to the day level.
Thus, the traditional counts metric and the other summary metrics do not appear to represent identical measures or constructs. Which is “better?” The other metrics are superior, in terms of comparability, as discussed above. However, further work is needed to investigate associations with health outcomes using the different metrics and to compare their magnitudes of association, as well as their ability to predict outcomes in the WHS. And, as noted above, the interpretability of all the summary metrics is currently limited—that is, what is “x” MIMS-units, or AAI, and so forth? Further work is needed to translate these metrics to measures that are understandable.
Referring again to Figure 1, the framework shows that raw acceleration data can also be analyzed directly instead of being reduced and summarized. Rich details can be obtained from such analyses, and research is active in this area, particularly for identifying specific activities being carried out. This is useful since early studies that used accelerometers typically employed a hip-wear protocol, which does not detect sedentary postures well using the counts metric; thus, algorithms that can better identify such postures are necessary. For example, a machine learning algorithm, Two-Level Behavioral Classification, was developed by Ellis et al. (Ellis et al., 2016; Rosenberg et al., 2017) to improve the detection of specific types of sedentary behaviors and physical activities (sitting, riding in a vehicle, standing still, standing moving, and walking/running) using accelerometer data from hip-worn devices in older women. In another example, Greenwood-Hickman et al. (2021) developed the Convolutional Neural Network Hip Accelerometer Posture classification method using a deep neural network to classify sitting versus nonsitting postures, as well as sit-to-stand transitions among older individuals wearing hip devices. It is again worth emphasizing the importance of device wear location; these algorithms were developed for hip-worn devices and may not be transferrable to data from devices worn at other locations.
Conclusions
There is a large body of evidence over many decades showing that physical activity is related to good health and functioning. Many of the studies underpinning the evidence base for long-term health outcomes have been observational studies, rather than randomized clinical trials which have primarily been used to study biomarkers of the long-term health outcomes (e.g., cardiometabolic risk factors, rather than CVD itself). In this paper, we have discussed reasons why the observational study design is better suited for investigating long-term outcomes.
More recently, we have entered an exciting era of physical activity and sedentary behavior research where many observational studies now use devices to measure these behaviors instead of relying on self-reports, yielding more precise and detailed information. However, while device technology is rapidly advancing, data collection in such studies can be plodding for valid reasons. To maximize the use of data that may have been collected years ago, investigators who study the effect of physical behaviors on health urgently need clarity and consensus on the most efficient metrics and algorithms, using open-source methods, which can be compared across devices (van Hees et al., 2016; Wijndaele et al., 2015). This will allow maximal leverage of device data to inform future guidelines.
Acknowledgments
This study was supported by grants CA154647, CA047988, CA182913, HL043851, HL080467, and HL099355 from the National Institutes of Health (NIH), United States; and CA227122 from the Office of the Director, Office of Disease Prevention, and Office of Behavioral and Social Sciences Research, National Cancer Institute, NIH, United States. Moore was supported by T32-HL007055, National Heart, Lung, and Blood Institute, NIH, United States.
References
2018 Physical Activity Guidelines Advisory Committee. (2018). 2018 physical activity guidelines advisory committee scientific report. U.S. Department of Health and Human Services.
American Cancer Society. (2018). Global cancer facts & figures (4th ed.). American Cancer Society.
Arem, H., Moore, S.C., Patel, A., Hartge, P., Berrington de Gonzalez, A., Visvanathan, K., … Matthews, C.E. (2015). Leisure time physical activity and mortality: A detailed pooled analysis of the dose-response relationship. JAMA Internal Medicine, 175(6), 959–967. https://doi.org/10.1001/jamainternmed.2015.0533
Bai, J., Di, C., Xiao, L., Evenson, K.R., LaCroix, A.Z., Crainiceanu, C.M., & Buchner, D.M. (2016). An activity index for raw accelerometry data and its comparison with other activity metrics. PLoS One, 11(8), Article e0160644. https://doi.org/10.1371/journal.pone.0160644
Bai, J., He, B., Shou, H., Zipunnikov, V., Glass, T.A., & Crainiceanu, C.M. (2014). Normalization and extraction of interpretable metrics from raw accelerometry data. Biostatistics, 15(1), 102–116. https://doi.org/10.1093/biostatistics/kxt029
Buffey, A.J., Herring, M.P., Langley, C.K., Donnelly, A.E., & Carson, B.P. (2022). The acute effects of interrupting prolonged sitting time in adults with stnding and light-intensity walking on biomarkers of cardiometabolic health in adults: A systematic review and meta-analysis. Sports Medicine, 52(8), 1765–1787. https://doi.org/10.1007/s40279-022-01649-4
Cook, N.R., Lee, I.M., Gaziano, J.M., Gordon, D., Ridker, P.M., Manson, J.E., … Buring, J.E. (2005). Low-dose aspirin in the primary prevention of cancer: The women’s health study: A randomized controlled trial. Journal of the American Medical Association, 294(1), 47–55. https://doi.org/10.1001/jama.294.1.47
Danova, T. (2015). The entire history of the smartwatch and fitness-band market in one infographic. https://www.businessinsider.com/the-smartwatch-and-fitness-band-market-2015-1
Ekelund, U., Steene-Johannessen, J., Brown, W.J., Fagerland, M.W., Owen, N., Powell, K.E., … Lancet Sedentary Behaviour Working Group. (2016). Does physical activity attenuate, or even eliminate, the detrimental association of sitting time with mortality? A harmonised meta-analysis of data from more than 1 million men and women. Lancet, 388(10051), 1302–1310. https://doi.org/10.1016/S0140-6736(16)30370-1
Ekelund, U., Tarp, J., Steene-Johannessen, J., Hansen, B.H., Jefferis, B., Fagerland, M.W., … Lee, I.M. (2019). Dose-response associations between accelerometry measured physical activity and sedentary time and all cause mortality: Systematic review and harmonised meta-analysis. The BMJ, 366, Article l4570. https://doi.org/10.1136/bmj.l4570
Ellis, K., Kerr, J., Godbole, S., Staudenmayer, J., & Lanckriet, G. (2016). Hip and wrist accelerometer algorithms for free-living behavior classification. Medicine & Science in Sports & Exercise, 48(5), 933–940. https://doi.org/10.1249/MSS.0000000000000840
Evenson, K.R., Bellettiere, J., Cuthbertson, C.C., Di, C., Dushkes, R., Howard, A.G., … LaCroix, A.Z. (2021). Cohort profile: The women’s health accelerometry collaboration. BMJ Open, 11(11), Article e052038. https://doi.org/10.1136/bmjopen-2021-052038
Evenson, K.R., Scherer, E., Peter, K.M., Cuthbertson, C.C., & Eckman, S. (2022, November 21). Historical development of accelerometry measures and methods for physical activity and sedentary behavior research worldwide: A scoping review of observational studies of adults. PLoS One. 17(11), Article e0276890. https://doi.org/10.1371/journal.pone.0276890
Evenson, K.R., Wen, F., Herring, A.H., Di, C., LaMonte, M.J., Tinker, L.F., … Buchner, D.M. (2015). Calibrating physical activity intensity for hip-worn accelerometry in women age 60 to 91 years: The Women’s Health Initiative OPACH Calibration Study. Preventive Medicine Reports, 2, 750–756. https://doi.org/10.1016/j.pmedr.2015.08.021
Gong, Q., Zhang, P., Wang, J., Ma, J., An, Y., Chen, Y., … Da Qing Diabetes Prevention Study Group. (2019). Morbidity and mortality after lifestyle intervention for people with impaired glucose tolerance: 30-year results of the Da Qing diabetes prevention outcome study. Lancet Diabetes & Endocrinology, 7(6), 452–461. https://doi.org/10.1016/S2213-8587(19)30093-2
Greenwood-Hickman, M.A., Nakandala, S., Jankowska, M.M., Rosenberg, D.E., Tuz-Zahra, F., Bellettiere, J., … Natarajan, L. (2021). The CNN Hip Accelerometer Posture (CHAP) method for classifying sitting patterns from hip accelerometers: A validation study. Medicine & Science in Sports & Exercise, 53(11), 2445–2454. https://doi.org/10.1249/MSS.0000000000002705
Intille, S.S., Lester, J., Sallis, J.F., & Duncan, G. (2012). New horizons in sensor development. Medicine & Science in Sports & Exercise, 44(1, Suppl. 1), S24–S31. https://doi.org/10.1249/MSS.0b013e3182399c7d
Jacobs, D.R., Jr., Woo, J.G., Sinaiko, A.R., Daniels, S.R., Ikonen, J., Juonala, M., … Dwyer, T. (2022). Childhood cardiovascular risk factors and adult cardiovascular events. New England Journal of Medicine, 386(20), 1877–1888. https://doi.org/10.1056/NEJMoa2109191
Jefferis, B.J., Sartini, C., Ash, S., Lennon, L.T., Wannamethee, S.G., & Whincup, P.H. (2016). Validity of questionnaire-based assessment of sedentary behaviour and physical activity in a population-based cohort of older men; comparisons with objectively measured physical activity data. The International Journal of Behavioral Nutrition and Physical Activity, 13, 14. https://doi.org/10.1186/s12966-016-0338-1
John, D., Tang, Q., Albinali, F., & Intille, S. (2019). An open-source monitor-independent movement summary for accelerometer data processing. Journal for the Measurement of Physical Behaviour, 2(4), 268–281. https://doi.org/10.1123/jmpb.2018-0068
Katzmarzyk, P.T., Powell, K.E., Jakicic, J.M., Troiano, R.P., Piercy, K., Tennant, B., & Physical Activity Guidelines Advisory Committee. (2019). Sedentary behavior and health: Update from the 2018 physical activity guidelines advisory committee. Medicine & Science in Sports & Exercise, 51(6), 1227–1241. https://doi.org/10.1249/MSS.0000000000001935
Lee, I.M., Cook, N.R., Gaziano, J.M., Gordon, D., Ridker, P.M., Manson, J.E., … Buring, J.E. (2005). Vitamin E in the primary prevention of cardiovascular disease and cancer: The Women’s health study: A randomized controlled trial. Journal of the American Medical Association, 294(1), 56–65. https://doi.org/10.1001/jama.294.1.56
Lee, I.M., Shiroma, E.J., Evenson, K.R., Kamada, M., LaCroix, A.Z., & Buring, J.E. (2018a). Accelerometer-measured physical activity and sedentary behavior in relation to all-cause mortality: The women’s health study. Circulation, 137(2), 203–205. https://doi.org/10.1161/CIRCULATIONAHA.117.031300
Lee, I.M., Shiroma, E.J., Evenson, K.R., Kamada, M., LaCroix, A.Z., & Buring, J.E. (2018b). Using devices to assess physical activity and sedentary behavior in a large cohort study, the women’s health study. Journal for the Measurement of Physical Behaviour, 1(2), 60–69. https://doi.org/10.1123/jmpb.2018-0005
Lee, I.M., Shiroma, E.J., Kamada, M., Bassett, D.R., Matthews, C.E., & Buring, J.E. (2019). Association of step volume and intensity with all-cause mortality in older women. JAMA Internal Medicine, 179(8), 1105–1112. https://doi.org/10.1001/jamainternmed.2019.0899
Look Ahead Research Group, Wing, R.R., Bolin, P., Brancati, F.L., Bray, G.A., Clark, J.M., … Yanovski, S.Z. (2013). Cardiovascular effects of intensive lifestyle intervention in type 2 diabetes. New England Journal of Medicine, 369(2), 145–154. https://doi.org/10.1056/NEJMoa1212914
Martinez-Amezcua, P., Dooley, E.E., Reed, N.S., Powell, D., Hornikel, B., Golub, J.S., … Palta, P. (2022). Association of hearing impairment and 24-hour total movement activity in a representative sample of US adults. JAMA Netw Open, 5(3), Article e222983. https://doi.org/10.1001/jamanetworkopen.2022.2983
Matthews, C.E., Troiano, R.P., Salerno, E.A., Berrigan, D., Patel, S.B., Shiroma, E.J., & Saint-Maurice, P.F. (2020). Exploration of confounding due to poor health in an accelerometer-mortality study. Medicine & Science in Sports & Exercise, 52(12), 2546–2553. https://doi.org/10.1249/MSS.0000000000002405
Ndahimana, D., & Kim, E.K. (2017). Measurement methods for physical activity and energy expenditure: A review. Clinical Nutrition Research, 6(2), 68–80. https://doi.org/10.7762/cnr.2017.6.2.68
Paluch, A.E., Bajpai, S., Bassett, D.R., Carnethon, M.R., Ekelund, U., Evenson, K.R., … The Steps for Health Collaborative. (2022). Daily steps and all-cause mortality: A meta-analysis of 15 international cohorts. Lancet Public Health, 7(3), e219–e228. https://doi.org/10.1016/S2468-2667(21)00302-9
Pan, X.R., Li, G.W., Hu, Y.H., Wang, J.X., Yang, W.Y., An, Z.X., … Howard, B.V. (1997). Effects of diet and exercise in preventing NIDDM in people with impaired glucose tolerance. The Da Qing IGT and diabetes study. Diabetes Care, 20(4), 537–544. https://doi.org/10.2337/diacare.20.4.537
Physical Activity Guidelines Committee. (2008). Physical activity guidelines advisory committee report. Department of Health and Human Services.
R Package ActivityIndex. (2021). ActivityIndex: Activity index calculation using raw ‘accelerometry’ data. https://cran.rstudio.com/web/packages/ActivityIndex/index.html
R Package GGIR. (2022). GGIR: Raw accelerometer data analysis. https://cran.r-project.org/web/packages/GGIR/index.html
R Package MIMS-Unit Algorithm. (2022). MIMSunit: Algorithm to compute monitor independent movement summary unit (MIMS-Unit). https://mhealthgroup.github.io/MIMSunit/
Ramakrishnan, R., Doherty, A., Smith-Byrne, K., Rahimi, K., Bennett, D., Woodward, M., … Dwyer, T. (2021). Accelerometer measured physical activity and the incidence of cardiovascular disease: Evidence from the UK Biobank cohort study. PLoS Medicine, 18(1), Article e1003487. https://doi.org/10.1371/journal.pmed.1003487
Ridker, P.M., Cook, N.R., Lee, I.M., Gordon, D., Gaziano, J.M., Manson, J.E., … Buring, J.E. (2005). A randomized trial of low-dose aspirin in the primary prevention of cardiovascular disease in women. New England Journal of Medicine, 352(13), 1293–1304. https://doi.org/10.1056/NEJMoa050613
Rosenberg, D., Godbole, S., Ellis, K., Di, C., Lacroix, A., Natarajan, L., & Kerr, J. (2017). Classifiers for accelerometer-measured behaviors in older women. Medicine & Science in Sports & Exercise, 49(3), 610–616. https://doi.org/10.1249/MSS.0000000000001121
Stensvold, D., Viken, H., Steinshamn, S.L., Dalen, H., Stoylen, A., Loennechen, J.P., … Wisloff, U. (2020). Effect of exercise training for five years on all cause mortality in older adults-the Generation 100 study: Randomised controlled trial. The BMJ, 371, Article m3485. https://doi.org/10.1136/bmj.m3485
Troiano, R.P., McClain, J.J., Brychta, R.J., & Chen, K.Y. (2014). Evolution of accelerometer methods for physical activity research. British Journal of Sports Medicine, 48(13), 1019–1023. https://doi.org/10.1136/bjsports-2014-093546
U.S. Department of Health and Human Services. (2018). Physical activity guidelines for Americans (2nd ed.).
Vähä-Ypyä, H., Vasankari, T., Husu, P., Suni, J., & Sievanen, H. (2015). A universal, accurate intensity-based classification of different physical activities using raw data of accelerometer. Clinical Physiology and Functional Imaging, 35, 64–70.
van Hees, V.T., Gorzelniak, L., Dean Leon, E.C., Eder, M., Pias, M., Taherian, S., … Brage, S. (2013). Separating movement and gravity components in an acceleration signal and implications for the assessment of human daily physical activity. PLoS One, 8(4), Article e61691. https://doi.org/10.1371/journal.pone.0061691
van Hees, V.T., Thaler-Kall, K., Wolf, K.H., Brond, J.C., Bonomi, A., Schulze, M., … Horsch, A. (2016). Challenges and opportunities for harmonizing research methodology: Raw accelerometry. Methods of Information in Medicine, 55(6), 525–532. https://doi.org/10.3414/ME15-05-0013
Wang, G., Wu, S., Evenson, K.R., Kang, I., LaMonte, M.J., Bellettiere, J., … Di, C. (2022). Calibration of an accelerometer activity index among older women and its association with cardiometabolic risk factors. Journal for the Measurement of Physical Behaviour, 5(3), 145–155. https://doi.org/10.1123/jmpb.2021-0031
Westreich, D., Edwards, J.K., Lesko, C.R., Cole, S.R., & Stuart, E.A. (2019). Target validity and the hierarchy of study designs. American Journal of Epidemiology, 188(2), 438–443. https://doi.org/10.1093/aje/kwy228
Wijndaele, K., Westgate, K., Stephens, S.K., Blair, S.N., Bull, F.C., Chastin, S.F., … Healy, G.N. (2015). Utilization and harmonization of adult accelerometry data: Review and expert consensus. Medicine & Science in Sports & Exercise, 47(10), 2129–2139. https://doi.org/10.1249/MSS.0000000000000661
World Health Organization. (2020). WHO guidelines on physical activity and sedentary behaviour.
