JMIR Publications


The Karma system is currently undergoing maintenance (Monday, January 29, 2018).
The maintenance period has been extended to 8PM EST.

Karma Credits will not be available for redeeming during maintenance.

JMIR Public Health and Surveillance

Advertisement

Citing this Article

Right click to copy or hit: ctrl+c (cmd+c on mac)

Published on 01.02.18 in Vol 4, No 1 (2018): Jan-Mar

This paper is in the following e-collection/theme issue:

    Original Paper

    Objectively Measured Baseline Physical Activity Patterns in Women in the mPED Trial: Cluster Analysis

    1Department of Physiological Nursing/Institute for Health & Aging, University of California, San Francisco, San Francisco, CA, United States

    2Department of Industrial Engineering and Operations Research, University of California, Berkeley, Berkeley, CA, United States

    3Department of Epidemiology & Biostatistics, University of California, San Francisco, San Francisco, CA, United States

    4Stanford Prevention Research Center, Stanford University, Palo Alto, CA, United States

    5Department of Electrical Engineering and Computer Sciences, University of California, Berkeley, Berkeley, CA, United States

    Corresponding Author:

    Yoshimi Fukuoka, PhD, RN, FAAN

    Department of Physiological Nursing/Institute for Health & Aging

    University of California, San Francisco

    2 Koret Way, Box 0610

    San Francisco, CA, 94116

    United States

    Phone: 1 415 476 8419

    Fax:1 415 476 8899

    Email:


    ABSTRACT

    Background: Determining patterns of physical activity throughout the day could assist in developing more personalized interventions or physical activity guidelines in general and, in particular, for women who are less likely to be physically active than men.

    Objective: The aims of this report are to identify clusters of women based on accelerometer-measured baseline raw metabolic equivalent of task (MET) values and a normalized version of the METs ≥3 data, and to compare sociodemographic and cardiometabolic risks among these identified clusters.

    Methods: A total of 215 women who were enrolled in the Mobile Phone Based Physical Activity Education (mPED) trial and wore an accelerometer for at least 8 hours per day for the 7 days prior to the randomization visit were analyzed. The k-means clustering method and the Lloyd algorithm were used on the data. We used the elbow method to choose the number of clusters, looking at the percentage of variance explained as a function of the number of clusters.

    Results: The results of the k-means cluster analyses of raw METs revealed three different clusters. The unengaged group (n=102) had the highest depressive symptoms score compared with the afternoon engaged (n=65) and morning engaged (n=48) groups (overall P<.001). Based on a normalized version of the METs ≥3 data, the moderate-to-vigorous physical activity (MVPA) evening peak group (n=108) had a higher body mass index (P=.03), waist circumference (P=.02), and hip circumference (P=.03) than the MVPA noon peak group (n=61).

    Conclusions: Categorizing physically inactive individuals into more specific activity patterns could aid in creating timing, frequency, duration, and intensity of physical activity interventions for women. Further research is needed to confirm these cluster groups using a large national dataset.

    Trial Registration: ClinicalTrials.gov NCT01280812; https://clinicaltrials.gov/ct2/show/NCT01280812 (Archived by WebCite at http://www.webcitation.org/6vVyLzwft)

    JMIR Public Health Surveill 2018;4(1):e10

    doi:10.2196/publichealth.9138

    KEYWORDS



    Introduction

    Background

    Increasing physical activity is associated with a reduction in chronic illnesses and an increase in psychological well-being [1-3]. The 2008 Physical Activity Guidelines for Americans recommends US adults engage in a total 150 minutes of moderate-intensity aerobic activity (ie, brisk walking) each week or 75 minutes of vigorous-intensity aerobic activity each week, to be done in at least 10-minute bouts of activity [4]. The guidelines were developed mainly based on self-reported physical activity data in relation to health outcomes [5]. Since the current guidelines were issued, more objectively measured physical activity data in relation to health outcomes have become available. Recently, the US Department of Health and Human Services announced that they intend to publish new physical activity guidelines in 2018 [6].

    Recent investigations have shown that there is a large discrepancy between self-reported and objectively measured moderate-to-vigorous physical activity (MVPA) [7]. Although half of adults meet the current physical activity guidelines by self-report, only 3.5% of American adults meet these guidelines by accelerometer [8]. In particular, women and older adults are less likely to be physically active than men and younger adults regardless of measurement methods [9,10]. Dichotomization of meeting or not meeting the physical activity guidelines provides only one-dimensional information. However, identifying patterns of physical activity throughout the day may help develop more personalized interventions or physical activity guidelines in general, and in particular for women and older adults.

    Cluster analysis is a useful statistical technique that can allocate observations/individuals into groups based on similar characteristics [11]. In the past, a cluster analysis technique was used to cluster individuals based on self-reported physical activity and sedentary behavior. Few studies utilized objectively measured (ie, accelerometer) physical activity data. In a large cohort study in Hong Kong, two clusters were identified: (1) the active group characterized by a routine activity pattern on weekdays and a varied pattern on weekends and (2) the less active group characterized by a low activity pattern on weekdays and weekends [12]. A total of 72% of adults in this Hong Kong sample were classified as the less active group, and the daily average duration of MVPA in the active groups was two times greater than in the less active group. One of the limitations of this cohort study was that only four consecutive days of accelerometer data were used.

    Goals of This Study

    Our research team had a unique opportunity to analyze seven consecutive days of accelerometer data in women who were screened and completed the run-in period of the Mobile Phone Based Physical Activity Education (mPED) randomized controlled trial (RCT). To our knowledge, no study has used cluster analyses to explore daily patterns of physical activity using seven consecutive days of accelerometer data in female adults. The aims of this paper are (1) to identify clusters of women who enrolled in the mPED study based on overall accelerometer-measured baseline physical activity and MVPA and (2) to compare sociodemographic and cardiometabolic risks among these identified clusters.


    Methods

    Study Design and Sample

    The mPED study is a RCT of the app-based physical activity intervention in physically inactive women (trial registration: ClinicalTrials.gov NCT01280812). Detailed descriptions of the study design and study protocol have been published previously [7,13,14]. The study protocol was approved by the University of California, San Francisco Committee on Human Research and the Data and Safety Monitoring Board. All participants provided written consent prior to study enrollment. In this paper, we analyzed only the sociodemographic, clinical, and self-reported questionnaires data collected at the screening/baseline study visit and accelerometry data collected during the last 7 days of the run-in period prior to a randomization visit.

    Initial inclusion criteria for the mPED trial were (1) physically inactive at work and/or during leisure time based on the Stanford Brief Activity Survey [15], (2) intent to be physically active, (3) female aged 25 to 69 years, (4) access to a home telephone or mobile phone, (5) speak and read English, (6) body mass index (BMI) of 18.5 to 43.0 kg/m2, and (7) no mild cognitive impairment screened by the Mini-Cog test [16,17]. Initial exclusion criteria were (1) known medical conditions or physical problems that require special attention in an exercise program, (2) planning an international trip during the next 4 months (which could interfere with daily server uploads of mobile phone data), (3) pregnant/gave birth during the past 6 months, (4) severe hearing or speech problem, (5) history of eating disorder, (6) current substance abuse, (7) current participation in lifestyle modification programs or research studies that may confound study results, and (8) history of bariatric surgery or plans for bariatric surgery in the next 12 months.

    In total, 318 women came in for a screening/baseline visit. Of those, 57 did not start or complete the run-in period and 46 did not have sufficient accelerometer wear time of at least 8 hours per day for the last 7 days prior to the randomization visit. The remaining 215 participants were analyzed in this report.

    Measures

    A triaxial accelerometer (HJA-350IT, Active Style Pro, Omron Healthcare Co, Ltd) was used to assess objectively measured physical activity [18,19]. Its dimensions are 74×46×34 mm (width/height/depth) including the clip, and it weighs 60 grams (2.1 oz). Throughout the run-in period, participants were asked to wear the accelerometer all day on their waist, except when showering, bathing, swimming, or sleeping, from the time they got up in the morning until they went to bed at night. All participants were also instructed to engage in their regular daily activity and not increase this activity during the run-in period. The accelerometer displayed only date and time. To avoid providing any feedback and to collect the clean baseline activity data, neither the step counts nor metabolic equivalent of task (MET) values were displayed. Activity data were stored minute by minute for the entire duration of the run-in period, and the accelerometer’s data was automatically reset at midnight. A trained research staff downloaded the data to a personal computer with the software program provided by the manufacturer in the research office prior to randomization visit. In this paper, only recorded accelerometer data during the seven consecutive days prior to the randomization visit were used to identify patterns of physical activity. In order for accelerometer data to be valid, all 7 days of accelerometer activity needed to indicate at least 8 hours per day of recorded wear time for the device. The METs determined by this accelerometer are closely correlated with METs calculated using energy expenditure measured by indirect calorimetry [20,21]. This accelerometer was programed to collect physical activity intensity every 10 seconds per minute and the mean intensity value of a 1-minute epoch was calculated as the mean of six 10-second epochs. Moderate- or vigorous-intensity activity was defined as ≥3 to <6 or ≥6 METs, respectively, using the Compendium of Physical Activities [20,21].

    The Center for Epidemiological Studies Depression Scale (CES-D) is a 20-item questionnaire widely used for assessing symptoms of depression [22]. Scores can range from 0 to 60, with higher scores indicating more depressive symptoms. The 12-item Short-Form Health Survey (SF-12) is an instrument derived from the longer 36-item Short-Form Survey, which was designed to measure general health functioning [23]. The SF-12 provides two summary scores, the Physical Component Summary and the Mental Component Summary. Scores are standardized; the mean score in the population is 50 with a standard deviation of 10 points. Higher scores indicate better functioning in physical function or mental status. The Television/Computer Usage Scale is a semistructured interview that estimates an individual’s time spent (1) using a computer, Internet, or mobile phone and (2) watching television or movies for the 7 days prior to the interview. This measure was developed by the investigator prior to the trial. A trained research staff used the 7-day worksheet to assess the duration of these activities for the 7 days. The Social Support for Exercise Survey consists of 13 items assessing the level of perceived support from family and friends for behavior changes related to exercise [24]. Each item is scored separately for family and friends, and scores can range from 13 to 65 with higher scores indicating greater support. The Barriers to Being Active Quiz consists of 21 items assessing seven subscales: lack of time, lack of social influence, lack of energy, lack of willpower, fear of injury, lack of skill, and lack of resources. Each subscale can range from 0 to 9 and total scores can range from 0 to 63, with higher scores indicating more barriers to physical activity [25]. The Modified Self-Efficacy for Physical Activity Scale, consisting of six items (five original questions plus one extra question), was used to assess confidence in one’s ability to exercise, an important determinant of the stages of change for exercise behavior. Total scores can range from 6 to 30, with higher scores indicating greater self-efficacy for physical activity. Anthropometric measures included height, weight in kilograms, and waist and hip circumferences; BMI was calculated based on height and weight in kilograms at the screening/baseline visit. Participants were asked to change to a hospital gown and remove their shoes prior to the measurement.

    Statistical Analysis

    The k-means clustering method (hereafter referred to as k-means) [26] was applied to the accelerometer dataset. This method takes as input: (1) a set of data points with each data point corresponding to a single individual, (2) a subset of characteristics summarizing each data point, and (3) a number of desired clusters. In the terminology of machine learning, the subset of summarizing characteristics is known as the features of the data [27]. As output, this method separates the data points into distinct groups (ie, clusters) such that the data points within each group have similar characteristics and the data points between different groups have different characteristics.

    To apply k-means, we used the Lloyd algorithm [28] to perform the computations. To ensure accurate modeling, we repeated the Lloyd algorithm a total of 25 times with random initialization to find the most accurate clustering (as measured by the percentage of variance of the data explained by the identified cluster medians). To determine an appropriate number of desired clusters, we applied the elbow method [29]. The elbow method selects the number of clusters to be such that adding an additional cluster does not significantly reduce the within-group sum of squares. We applied k-means two times, and each time we used a different subset of summarizing characteristics. The two different subsets we used in our analysis are described subsequently. After applying k-means, chi-square or ANOVA tests were used to compare sociodemographic and clinical characteristics among these clustered groups. To visualize the clusters, we first computed the mean for each group selected by k-means of the corresponding data points. Then we applied Loess smoothing [30] in time to better visualize average temporal trends. Statistical analyses were performed in R 3.1.1 [31].

    Raw METs Data

    The k-means clustering was applied to raw METs data from each enrolled participant to evaluate if raw minute-level METs were able to classify participants by physical activity and time to do physical activity. All observations including day and night were included because participants engaged in activity at various time points. Thus, naively removing night data would lead to a loss of information. Specifically, the features for each individual consisted of a 10,080-dimensional vector comprised of consecutive (at the minute interval) METs observations for 7 days. Missing data occurred mainly during nighttime and hence were simply replaced by 1, which is the METs reading for a stationary individual.

    Figure 1. Equations 1-3.
    View this figure

    Normalized METs ≥3 Data

    We also explored how MVPA (METs≥3) were associated with sociodemographic data and clinical outcomes. Thus, k-means clustering was applied to a normalized version of the METs data from each enrolled participant, and the data were normalized as follows: suppose for each participant i (i=1,2,...,215) and time t (t=1,2,...,10,080), the raw METs record was di,t. We first converted the raw METs records into binary values (see Equation 1 in Figure 1). This binary conversion corresponds to whether the participant was having MVPA or not, which is an important indicator to characterize a person’s physical activity level. Next, we averaged the binary values for the 7 days to compute the MVPA frequency for a typical day for each individual. We ended up with 1440 features for each individual, which indicates the minute-level normalized METs for the day averaged over all days (see Equation 2 in Figure 1).

    Finally, we normalized this vector for each individual by time to have unit Euclidean norm (see Equation 3 in Figure 1). This normalization ensured that the overall physical activity level of each participant was similar and that the clustering results then categorized participants using the time in day (ie, morning, noon, evening) information.


    Results

    Overall Participants’ Characteristics

    Overall, the mean age of participants was 52.4 (SD 11.2) years, 54.4% (117/215) were white, 48.8% (105/215) were single or divorced, and 73.0% (157/215) were well educated, reporting college- or graduate-level educations. In addition, 49.3% (106/215) had used a pedometer and 57.2% (123/215) had participated in a diet/weight loss plan prior to study enrollment. The majority of the sample (80.5%, 173/215) drove a car at least once per week.

    Clustering on Raw METs Data

    The k-mean clustering separated the participants into three groups (Figure 2). The elbow method indicated that separating the data into four groups did not reduce within-group sum of squares significantly. Therefore, we chose three clusters for this analysis (Multimedia Appendix 1). There were 65, 48, and 102 participants in groups 1, 2, and 3, respectively. We refer to these clusters as the “morning engaged,” “afternoon engaged,” and “unengaged” groups. Figure 1 shows the mean METs for each minute in a day by the three groups after Loess smoothing with span 0.1. The plot in Figure 2 indicates that the morning engaged group engaged in activity earlier than the afternoon engaged group and both groups had similar overall activity level, whereas the unengaged group did not engage in activity as much as the other two groups.

    Figure 2. A k-means cluster analysis of raw METs (metabolic equivalent of tasks) data with the Lloyd algorithm (N=215) for physical activity frequency during the day for each cluster.
    View this figure
    Table 1. Comparison of sociodemographic and clinical characteristics among the three clustered groups based on raw metabolic equivalent of tasks (METs) data (N=215).
    View this table

    In Table 1, sociodemographic self-reported questionnaires and physical activity measures were compared among the three groups. The unengaged group represented 47.4% of the sample (102/215). The unengaged group had significantly lower weekly total minutes of accelerometer-measured MVPA with 10 minutes criteria and mean daily steps than the other two groups (overall P=.006 and P<.001, respectively). Furthermore, the unengaged group had the highest depressive symptoms score compared with the afternoon engaged and the morning engaged groups (overall P value <.001).

    Clustering on Normalized METs ≥3 Data

    The k-mean clustering separated the participants into three groups (Figure 3). This number of clusters was also chosen by the elbow method, and it showed the within-group sum of squares corresponding to the different number of clusters, and using four clusters did not reduce the within-group sum of squares significantly (Multimedia Appendix 2). There were 46, 61, and 108 participants in groups 1, 2, and 3, respectively. We will refer to these groups as the MVPA morning and evening active, MVPA noon peak active, and MVPA evening peak active groups. The clusters were named as such because the MVPA morning and evening active group engaged in MVPA either in the morning or in the evening, the MVPA noon peak active group engaged in MVPA around noon, and the MVPA evening peak active group engaged in MVPA in the evening and at night. The mean normalized METs for each group are shown in Figure 3 after Loess smoothing with span 0.1. We can interpret the vertical axis as the frequency of participants in that group who engaged in MVPA at a particular time. The MVPA morning and evening active group had two peaks: one in the morning and one in the evening. The MVPA noon peak active group tended to engage in MVPA around noon and did slightly less in the evening. The evening peak active group tended to gradually increase MVPA toward evening and with a peak around 6 pm. As seen in Table 2, the MVPA evening peak group (n=108) had higher BMI (P=.03), waist circumference (P=.02), and hip circumference (P=.03) than the MVPA noon peak group (n=61).

    Figure 3. A k-means cluster analysis of normalized METs (metabolic equivalent of tasks) ≥3 data (N=215). MVPA: moderate-to-vigorous physical activity.
    View this figure
    Table 2. Comparison of sociodemographic and clinical characteristics among three clustered groups based on normalized METs ≥3 data (N=215).
    View this table

    Discussion

    Principal Results

    This study is the first to identify clusters of women aged between 25 and 69 years based on seven consecutive days of accelerometer-measured METs and MVPA (≥3 METs). This first cluster analysis successfully identified three groups based on accelerometer-measured METs. It appears that only the difference between the afternoon engaged and the morning engaged groups is timing of activity throughout the day. However, the unengaged group (representing 47.4% of the sample) had a much lower activity level than the other two groups.

    We found that the unengaged group was more likely to have a college or graduate degree compared to the afternoon engaged and morning engaged groups. In the cluster analysis study of self-reported physical activity involving 3324 individuals in France, Omorou et al [32] also reported that individuals with high socioeconomic status were less likely to engage in occupational physical activity and active transportation compared to those with low socioeconomic status. In contrast, there is an inverse association between leisure physical activity and socioeconomic status [33]. In other words, individuals with high socioeconomic status had greater leisure physical activity time than those with low socioeconomic status.

    Furthermore, consistent with previous study findings [34,35], the unengaged group had a significantly higher depressive symptom score than the other two groups. It is estimated that approximately 20% to 25% of female adults have significantly elevated depressive symptoms (eg, CES-D score ≥16 points) [36]. This inverse association between depressive symptoms and physical activity levels has been well established. More than a dozen RCTs have examined the effect of physical activity on depressive symptoms, and some studies demonstrated physical activity could prevent or mitigate mild-to-moderate depressive symptoms [37]. The unengaged group may respond to a physical activity intervention differently compared with the afternoon active and morning active groups. However, this assumption needs to be confirmed in further studies.

    The second cluster analysis based on MVPA (normalized 3 ≥METs data) also showed three distinct groups (MVPA morning and evening peak, MVPA noon peak, and MVPA evening peak). A two-peak pattern of MVPA (7-8 am and 5-6 pm) in the MVPA morning and evening peak group might be explained by active commuting. The MVPA noon peak group appeared to have the greatest duration of MVPA compared with the other two groups. Moreover, this MVPA noon peak group had significantly lower metabolic risks (BMI, hip and waist circumferences) than the MVPA evening peak group. In a recent large epidemiologic study, the investigators also reported that bouts of 10 minutes or more of MVPA (as per current guidelines) and even bouts of less than 10 minutes were associated with lower levels of adiposity and a lower risk of metabolic syndrome in older adults [38]. The other studies found that bouts of at least 10 minutes of MVPA were a stronger predictor for metabolic risks than bouts of less than10 minutes of MVPA [39,40]. Given the benefit of MVPA regardless of its duration, less emphasis on bouts of at least 10 minutes of MVPA might help encourage physically inactive women to engage in MVPA throughout the day.

    Strengths and Limitations of the Study

    The strengths of this study were that we were able to use seven consecutive days of accelerometer-measured physical activity data instead of depending on participant recall to collect the vast majority of types of activities (active transportation, occupational and leisure activity), and to identify physical activity patterns that were specific to certain times of the day. In addition, the participants were not able to view their steps taken and intensity of physical activity during the data collection period; thus, this blinding function helped prevent participants from modifying their daily activity. Despite these strengths, some limitations need to be taken into account. First, the findings of this study might not be generalizable to men or children. Men tend to be more active than women are across their life spans. Second, in general, individuals with high levels of depressive symptoms are less likely to be enrolled in clinical studies compared to those with low symptoms. The proportion of the unengaged group could be larger than this data. Lastly, the accelerometer used in the mPED trial was not able to capture activities such as swimming, bicycling, and weight lifting. However, women who engaged in these activities in the mPED trial were relatively low in this sample [7].

    Despite the use of objectively measured physical activity, the sample size was relatively small in this study. Thus, these identified cluster groups need to be cross-validated using a large national dataset such as the National Health and Nutrition Examination Survey.

    Conclusions

    Classifying physically inactive individuals into more precise activity patterns could assist in tailoring the timing, frequency, duration, and intensity of physical activity interventions for women. For example, recommending bouts of physical activity before noon to the unengaged group or MVPA evening peak group may lead to an increase in their activity levels. Future research should consider examining how different types of baseline physical activity cluster groups will respond to different types of physical activity interventions.

    Acknowledgments

    This project was supported by a grant (R01HL104147) from the National Heart, Lung, and Blood Institute; by the American Heart Association; and by a grant (K24NR015812) from the National Institute of Nursing Research. MZ and AA were supported in part by a grant (CMMI-1450963) from the National Science Foundation, and MZ and KG were supported in part by funding from Fujitsu Research Labs, the UC Center for Information Technology Research in the Interest of Society (CITRIS), and the Philippine-California Advanced Research Institutes (PCARI). The study sponsors had no role in the study design; collection, analysis, or interpretation of data; writing the report; or the decision to submit the report for publication.

    Conflicts of Interest

    None declared.

    Multimedia Appendix 1

    Result of the Elbow Method for the raw METs data.

    PNG File, 89KB

    Multimedia Appendix 2

    Result of the Elbow Method for the normalized METs ≥3 data.

    PNG File, 77KB

    References

    1. Rossi A, Dikareva A, Bacon SL, Daskalopoulou SS. The impact of physical activity on mortality in patients with high blood pressure: a systematic review. J Hypertens 2012 Jul;30(7):1277-1288. [CrossRef] [Medline]
    2. Ford ES, Li C, Zhao G, Pearson WS, Tsai J, Churilla JR. Sedentary behavior, physical activity, and concentrations of insulin among US adults. Metabolism 2010 Sep;59(9):1268-1275. [CrossRef] [Medline]
    3. Mammen G, Faulkner G. Physical activity and the prevention of depression: a systematic review of prospective studies. Am J Prev Med 2013 Nov;45(5):649-657. [CrossRef] [Medline]
    4. US Department of Health and Human Services. 2008 Physical Activity Guidelines for Americans. 2008.   URL: https://health.gov/paguidelines/pdf/paguide.pdf [accessed 2017-12-12] [WebCite Cache]
    5. Evenson KR, Wen F, Metzger JS, Herring AH. Physical activity and sedentary behavior patterns using accelerometry from a national sample of United States adults. Int J Behav Nutr Phys Act 2015 Feb 15;12:20 [FREE Full text] [CrossRef] [Medline]
    6. US Department of Health and Human Services. Federal Register. 2015. Announcement of intent to establish the 2018 Physical Activity Guidelines Advisory Committee and solicitation of nominations for appointment to the committee membership   URL: https:/​/www.​federalregister.gov/​documents/​2015/​12/​18/​2015-31837/​announcement-of-intent-to-establish-the-2018-physical-activity-guidelines-advisory-committee-and [accessed 2017-12-12] [WebCite Cache]
    7. Fukuoka Y, Haskell W, Vittinghoff E. New insights into discrepancies between self-reported and accelerometer-measured moderate to vigorous physical activity among women - the mPED trial. BMC Public Health 2016 Aug 11;16(1):761 [FREE Full text] [CrossRef] [Medline]
    8. Troiano RP, Berrigan D, Dodd KW, Mâsse LC, Tilert T, McDowell M. Physical activity in the United States measured by accelerometer. Med Sci Sports Exerc 2008 Jan;40(1):181-188. [CrossRef] [Medline]
    9. Azevedo MR, Araújo CLP, Reichert FF, Siqueira FV, da Silva MC, Hallal PC. Gender differences in leisure-time physical activity. Int J Public Health 2007;52(1):8-15 [FREE Full text] [Medline]
    10. Sun F, Norman IJ, While AE. Physical activity in older people: a systematic review. BMC Public Health 2013;13:449 [FREE Full text] [CrossRef] [Medline]
    11. Everitt BS, Landau S, Leese M, Stahl D. Cluster Analysis, 5th Ed. London: John Wiley & Sons, Ltd; 2011.
    12. Lee P, Yu Y, McDowell I, Leung G, Lam T. A cluster analysis of patterns of objectively measured physical activity in Hong Kong. Public Health Nutr 2013 Aug;16(8):1436-1444. [CrossRef] [Medline]
    13. Fukuoka Y, Gay C, Haskell W, Arai S, Vittinghoff E. Identifying factors associated with dropout during prerandomization run-in period from an mHealth physical activity education study: the mPED trial. JMIR Mhealth Uhealth 2015 Apr 13;3(2):e34 [FREE Full text] [CrossRef] [Medline]
    14. Fukuoka Y, Komatsu J, Suarez L, Vittinghoff E, Haskell W, Noorishad T, et al. The mPED randomized controlled clinical trial: applying mobile persuasive technologies to increase physical activity in sedentary women protocol. BMC Public Health 2011 Dec 14;11:933 [FREE Full text] [CrossRef] [Medline]
    15. Taylor-Piliae RE, Norton LC, Haskell WL, Mahbouda MH, Fair JM, Iribarren C, et al. Validation of a new brief physical activity survey among men and women aged 60-69 years. Am J Epidemiol 2006 Sep 15;164(6):598-606. [CrossRef] [Medline]
    16. Borson S, Scanlan J, Brush M, Vitaliano P, Dokmak A. The mini-cog: a cognitive 'vital signs' measure for dementia screening in multi-lingual elderly. Int J Geriatr Psychiatry 2000 Nov;15(11):1021-1027. [Medline]
    17. Borson S, Scanlan JM, Chen P, Ganguli M. The Mini-Cog as a screen for dementia: validation in a population-based sample. J Am Geriatr Soc 2003 Oct;51(10):1451-1454. [Medline]
    18. Oshima Y, Kawaguchi K, Tanaka S, Ohkawara K, Hikihara Y, Ishikawa-Takata K, et al. Classifying household and locomotive activities using a triaxial accelerometer. Gait Posture 2010 Mar;31(3):370-374. [CrossRef] [Medline]
    19. Ohkawara K, Oshima Y, Hikihara Y, Ishikawa-Takata K, Tabata I, Tanaka S. Real-time estimation of daily physical activity intensity by a triaxial accelerometer and a gravity-removal classification algorithm. Br J Nutr 2011 Jun;105(11):1681-1691. [CrossRef] [Medline]
    20. Ainsworth BE, Haskell WL, Leon AS, Jacobs DR, Montoye HJ, Sallis JF, et al. Compendium of physical activities: classification of energy costs of human physical activities. Med Sci Sports Exerc 1993 Jan;25(1):71-80. [Medline]
    21. Ainsworth BE, Haskell WL, Whitt MC, Irwin ML, Swartz AM, Strath SJ, et al. Compendium of physical activities: an update of activity codes and MET intensities. Med Sci Sports Exerc 2000 Sep;32(9 Suppl):S498-S504. [Medline]
    22. Radloff LS. The CES-D Scale: a self-report depression scale for research in the general population. Appl Psychol Meas 1977 Jun 01;1(3):385-401. [CrossRef]
    23. Ware J, Kosinski M, Keller SD. A 12-Item Short-Form Health Survey: construction of scales and preliminary tests of reliability and validity. Med Care 1996 Mar;34(3):220-233. [Medline]
    24. Sallis JF, Grossman RM, Pinski RB, Patterson TL, Nader PR. The development of scales to measure social support for diet and exercise behaviors. Prev Med 1987 Nov;16(6):825-836. [Medline]
    25. Barriers to Being Active Quiz. Atlanta, GA: US Department of Health & Human Services   URL: http://www.cdc.gov/diabetes/ndep/pdfs/8-road-to-health-barriers-quiz-508.pdf [accessed 2015-11-04] [WebCite Cache]
    26. MacQueen J. Some methods for classification and analysis of multivariate observations. Proc Fifth Berkeley Symp on Math Statist and Prob 1967;1:281-297 [FREE Full text]
    27. Bishop C. Pattern Recognition and Machine Learning. New York: Springer; 2006.
    28. Lloyd S. Least squares quantization in PCM. IEEE Trans Inform Theory 1982 Mar;28(2):129-137 [FREE Full text] [CrossRef]
    29. Thorndike RL. Who belongs in the family? Psychometrika 1953 Dec;18(4):267-276. [CrossRef]
    30. Cleveland WS. LOWESS: A program for smoothing scatterplots by robust locally weighted regression. Am Stat 1981 Feb;35(1):54-55. [CrossRef]
    31. R Development Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing; 2011:A.
    32. Omorou AY, Coste J, Escalon H, Vuillemin A. Patterns of physical activity and sedentary behaviour in the general population in France: cluster analysis with personal and socioeconomic correlates. J Public Health (Oxf) 2016 Sep;38(3):483-492. [CrossRef] [Medline]
    33. Ford ES, Merritt RK, Heath GW, Powell KE, Washburn RA, Kriska A, et al. Physical activity behaviors in lower and higher socioeconomic status populations. Am J Epidemiol 1991 Jun 15;133(12):1246-1256. [Medline]
    34. Waller K, Kaprio J, Korhonen T, Tuulio-Henriksson A, Kujala UM. Persistent leisure-time physical activity in adulthood and use of antidepressants: a follow-up study among twins. J Affect Disord 2016 Aug;200:172-177. [CrossRef] [Medline]
    35. van Uffelen JG, van Gellecum YR, Burton NW, Peeters G, Heesch KC, Brown WJ. Sitting-time, physical activity, and depressive symptoms in mid-aged women. Am J Prev Med 2013 Sep;45(3):276-281. [CrossRef] [Medline]
    36. Bromberger JT, Harlow S, Avis N, Kravitz HM, Cordal A. Racial/ethnic differences in the prevalence of depressive symptoms among middle-aged women: The Study of Women's Health Across the Nation (SWAN). Am J Public Health 2004 Aug;94(8):1378-1385. [Medline]
    37. Mead GE, Morley W, Campbell P, Greig CA, McMurdo M, Lawlor DA. Exercise for depression. Cochrane Database Syst Rev 2008 Oct 08(4):CD004366. [CrossRef] [Medline]
    38. Jefferis BJ, Parsons TJ, Sartini C, Ash S, Lennon LT, Wannamethee SG, et al. Does duration of physical activity bouts matter for adiposity and metabolic syndrome? A cross-sectional study of older British men. Int J Behav Nutr Phys Act 2016 Mar 15;13:36 [FREE Full text] [CrossRef] [Medline]
    39. Strath SJ, Holleman RG, Ronis DL, Swartz AM, Richardson CR. Objective physical activity accumulation in bouts and nonbouts and relation to markers of obesity in US adults. Prev Chronic Dis 2008 Oct;5(4):A131 [FREE Full text] [Medline]
    40. Loprinzi PD, Cardinal BJ. Association between biologic outcomes and objectively measured physical activity accumulated in ≥ 10-minute bouts and <10-minute bouts. Am J Health Promot 2013;27(3):143-151. [CrossRef] [Medline]


    Abbreviations

    BMI: body mass index
    CES-D: Center for Epidemiological Studies Depression Scale
    HDL: high-density lipoprotein
    LDL: low-density lipoprotein
    MET: metabolic equivalent of task
    mPED: Mobile Phone Based Physical Activity Education
    MVPA: moderate-to-vigorous physical activity
    RCT: randomized controlled trial
    SF-12: 12-item Short-Form Health Survey


    Edited by G Eysenbach; submitted 09.10.17; peer-reviewed by G Norman; comments to author 03.11.17; revised version received 23.11.17; accepted 24.11.17; published 01.02.18

    ©Yoshimi Fukuoka, Mo Zhou, Eric Vittinghoff, William Haskell, Ken Goldberg, Anil Aswani. Originally published in JMIR Public Health and Surveillance (http://publichealth.jmir.org), 01.02.2018.

    This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Public Health and Surveillance, is properly cited. The complete bibliographic information, a link to the original publication on http://publichealth.jmir.org, as well as this copyright and license information must be included.