Habitual physical activity patterns in a nationally representative sample of U.S. adults

Health care providers (HCPs) favorably rated empathic communication skills training in terms of clinical relevance, novelty, clarity, and facilitator effectiveness. Participants also reported an increase in their self-confidence to communicate empathically with lung cancer patients, from pre- to post-training.

Abstract

Physical inactivity is a leading determinant of noncommunicable diseases. Yet, many adults remain physically inactive. Physical activity guidelines do not account for the multidimensionality of physical activity, such as the type or variety of physical activity behaviors. This study identified patterns of physical activity across multiple dimensions (e.g., frequency, duration, and variety) using a nationally representative sample of adults. Sociodemographic characteristics, health behaviors, and clinical characteristics associated with each physical activity pattern were defined. Multivariate finite mixture modeling was used to identify patterns of physical activity among 2003–2004 and 2005–2006 adult National Health and Nutrition Examination Survey participants. Chi-square tests were used to identify sociodemographic differences within each physical activity cluster and test associations between the physical activity clusters with health behaviors and clinical characteristics. Five clusters of physical activity patterns were identified: (a) low frequency, short duration (n = 730, 13%); (b) low frequency, long duration (n = 392, 7%); (c) daily frequency, short duration (n = 3,011, 55%); (d) daily frequency, long duration (n = 373, 7%); and (e) high frequency, average duration (n = 964, 18%). Walking was the most common form of activity; highly active adults engaged in more varied types of activity. High-activity clusters were comprised of a greater proportion of younger, White, nonsmoking adult men reporting moderate alcohol use without mobility problems or chronic health conditions. Active females engaged in frequent short bouts of activity. Data-driven approaches are useful for identifying clusters of physical activity that encompass multiple dimensions of activity. These activity clusters vary across sociodemographic and clinical subgroups.

Implications.

Practice: Practitioners may consider encouraging individuals to engage in a variety of physical activity types.

Policy: Physical activity questions that also evaluate individual activities should be included in National Health and Nutrition Examination Survey and other population-level epidemiological surveys.

Research: Future research should consider estimating physical activity as a multidimensional phenomena and implement analytic approaches, such as finite mixture modeling, to incorporate these multiple dimensions.

Introduction

Habitual physical activity is a cornerstone of a healthy lifestyle that has been shown to prevent the development of coronary artery disease and reduce the risk of Type 2 diabetes, osteoporosis, obesity, depression, as well as some forms of cancer (i.e., breast and colon) [1–7]. National guidelines for aerobic physical activity recommend a weekly dose of at least 150 min of moderate-intensity physical activity, 75 min of vigorous-intensity physical activity, or an equivalent combination for adults [8]. Despite this collective evidence and these guidelines, 55%–64% of adults in the USA do not meet these recommendations for regular aerobic activity [9]. This preponderance of physical inactivity has been shown to account for up to 11% of all national health care costs annually [10, 11]. Moreover, 8% of all deaths in the USA have been attributed to physical inactivity, with this proportion rising to 10% in those aged 40–69 years [12]. To increase the proportion of American adults who are physically active and to reduce the morbidity, mortality, and economic burden associated with physical inactivity, targeted approaches to increasing activity levels across the lifespan are needed.

Health behaviors are resistant to change despite research-supported models for predicting health behaviors (such as physical activity) and interventions for modifying them [13, 14]. There are at least two poorly understood dimensions of habitual physical activity that may inform targeted intervention approaches to increase physical activity at the population level. The first pertains to the patterning of physical activity participation. Different work schedules and life commitments (i.e., caring for others), for example, may manifest in different patterns of physical activity. For example, some adults may accrue 150 min of moderate-to-vigorous physical activity per week by jogging 30 min, 5 days per week, whereas others may complete two vigorous 45 min boot-camp-style classes per week. Both are meeting the recommended aerobic activity guidelines per week but in quite different ways. Understanding patterns of physical activity participation, and the sociodemographic factors related to these different patterns, could help practitioners develop targeted interventions to these subgroups. Moreover, the extent to which one physical activity pattern (e.g., 30 min of jogging 5 days per week vs. 45 min of boot camp 2 days per week) associates with different levels of risk for negative clinical characteristics has yet to be determined.

The “Frequency, Intensity, Time, Type” (FITT) model of physical activity prescription underpins the current physical activity recommendations; however, the “type” of activity is largely overlooked [15]. Yet, the reinforcing value (i.e., motivating value [16]) of different physical activity types differ in that some types of activity are reinforcing to some but not to others. Adults who report a greater reinforcing value for a particular type or mode of exercise, are more likely to meet activity guidelines [17]. Thus, it is plausible that participating in a greater variety or number of different physical activities would be positively associated with meeting activity guidelines. Quantifying the association between the number of different activities participated in with meeting activity guidelines and the extent to which this variety of activities engaged in differs by demographic factors (i.e., age group, sex, and educational attainment) may help decipher if the number of activity types should be a consideration for the “type” exercise prescription guideline. The importance of considering a multidimensional physical activity measurement is underscored by evidence that a greater number of activity types, but not a higher frequency of activity, protects against all-cause mortality in older adults [18].

The goals of the current study were to use a data-driven approach to identify distinct patterns of physical activity across multiple dimensions. These findings will advance our knowledge about patterns of physical activity participation by including important but often overlooked dimensions, such as the variety or number of different activity types. A second goal was to define the sociodemographic, health behavior, and clinical characteristics associated with each physical activity pattern or cluster. These findings will extend existing evidence of unique physical activity clusters found among older adults living in northern Manhattan [19] to a national sample of adults. Ultimately, this work may lead to targeted interventions to enhance physical activity adherence and lower noncommunicable disease risk.

Materials and Methods

Database

The Centers for Disease Control and Prevention collect health, nutrition, and health behavior information from noninstitutionalized U.S. civilians annually through the National Health and Nutrition Examination Survey (NHANES). The National Center for Health Statistics (NCHS) Research Ethics Review Board approves all NHANES protocols [20]. Procedures have been described previously for recruiting participants, obtaining informed consent, and collecting data [20].

Sample

NHANES uses a multistage probability sampling design to obtain a representative sample of the U.S. population. NHANES data from cohorts 2003–2004 and 2005–2006 were used for these analyses. The NHANES 2003–2006 participants were queried about the number of different activity types they engaged in and participation in different activity types. NHANES participants were no longer queried about individual leisure-time activities in subsequent cohorts. Thus, the NHANES 2003–2006 cohorts were appropriate for these analyses because the aim was to identify patterns of physical activity across multiple dimensions. All participants over the age of 21 who were not pregnant were included in these analyses (N = 9,816).

Measurements

Physical activity was self-reported by participants during the household interview. Participants responded to the number of times they engaged in approximately 60 individual leisure-time activities per day, per week, or per month. Examples of activities included aerobics, bicycling, fishing, hiking, racquetball, rowing, sit-ups, stretching, treadmill, boxing, surfing, and yard work.

Activity responses were converted to represent the number of times participants engaged in the activity over the past 30 days. Participants also reported their level of exertion in these activities (moderate or vigorous) and the average number of minutes spent in these activities. Activities were excluded if they were reported for less than 10 min duration. Activities were coded as missing if they were reported for ≥12 hr per day. The 12 hr threshold was selected because participants were queried about “leisure time” physical activities. Oher physical activity surveys reasonably limit self-reported activity to 3 hr per day. However, it is reasonable to spend more than 3 hr per day in many “leisure time” activities, such as golf, fishing, hunting, kayaking, horseback riding, and downhill skiing.

Physical activity responses were linked to metabolic equivalence (MET) scores to estimate the energy expended during the activity (kilocalories per kilogram per hour). MET scores were assigned to each activity according to the intensity level reported for that activity [21]. Energy expenditure (kilocalories) was calculated using the formula (MET score) × (activity duration) × (frequency amount/activity) × (weight of the participant) for each participant’s activity. Participants were coded as meeting the American College of Sports Medicine guidelines for aerobic physical activity if the total amount of moderate and/or moderate-to-vigorous activity per week was ≥150 min [9]. The total energy expenditure estimated for each activity cluster in this study is comparable to the total energy expenditure estimated for all but the most active groups in similar physical activity clusters reported previously [19].

Other health behaviors

Smoking

Participants responded to the queries: “Have you smoked at least 100 cigarettes in your life?” (yes/no) and “Do you smoke cigarettes now?” (every day/some days/not at all). Current smokers reported smoking “every day” or “some days.” Former smokers reported “yes” to smoking at least 100 cigarettes in their life, but “not at all” to smoking cigarettes now. Nonsmokers reported “no” to smoking at least 100 cigarettes in their life and “not at all” to smoking cigarettes now.

Alcohol use was coded as none, moderate, or heavy based on participant responses to a series of queries on alcohol use. Nondrinkers responded “no” to drinking at least 12 alcoholic beverages during any 1 year or over their lifetime or reported “0 drinks” over the past 12 months. Heavy and moderate use drinkers were identified by calculating the number of drinks per week from participant responses to the number of days participants reported drinking per week, per month, or per year and the number of drinks participants reported drinking on those days. Gender-specific thresholds were used to identify moderate (≤7 drinks per week for women, ≤14 drinks per week for men) and heavy drinkers (>7 drinks per week for women, >14 drinks per week for men) [22].

Clinical characteristics

Body mass index (BMI) was calculated from measured heights and weights using the formula weight (kilogram)/height (meter)². Heights and weights were measured by trained technicians as per the NHANES protocol. Obesity was defined by a BMI ≥30; overweight by a BMI ≥25 and <30; and underweight by a BMI <18.5 [23]. An ideal BMI was defined as ≥18.5 and <25.

Mobility: Participants were coded as “any difficulty” if responding “yes” to queries about limitations in the work they can do because of a physical, mental, or emotional problem; needing special equipment to walk; limitations related to confusion; difficulty climbing 10 stairs or walking a quarter mile. Participants were coded as “no difficulty” if responding “no” to these queries.

Chronic Health Conditions including diabetes, heart failure, coronary artery disease, cancer, and stroke were self-reported by participants as a health care provider diagnosed condition. Responses were coded as “yes,” “no,” or “don’t know.” Sociodemographic characteristics including age (in years), gender (male/female), race/ethnicity (White/Mexican/Other [Black, Asian]), and education (less than high school/greater than high school) were self-reported.

Statistical Analyses

Multivariate finite mixture modeling (MFMM) was used to identify clusters of physical activity among 2003–2004 and 2005–2006 adult NHANES participants [19]. The four physical activity measures included in the cluster analyses were the duration, frequency, total energy expenditure, and the number of types. Normal distributions were assessed for the logarithms of the duration, frequency, and total energy expenditure of the physical activity. A Poisson distribution was used to model the number of activity types. A common variance–covariance matrix of the log-normal variables across clusters and a conditional independence of the Poisson variable within clusters were assumed for these analyses. Participants reporting no physical activity were grouped as a separate cluster. A two-cluster model was fit among participants reporting physical activity and the number of clusters was increased by one successively to determine the number of clusters that best fit the data (MPlus Software, version 7.31). These models were compared using the Bayesian information criterion (BIC), Akaike information criterion (AIC), and relative entropy to select the best model wherein there were different response probabilities across the physical activity measures in each cluster. AIC and BIC informed the optimal model selection in terms of fit and parsimony with lower AIC and BIC values indicating optimal fit [24]. Entropy was also used in model selection because it reflects the accuracy of cluster assignment [24]. Entropy ranges from 0 to 1 with higher values indicating greater accuracy [24]. The interpretability of the clusters was also considered for determining the best model for the data.

Walking habits were described separately for each activity cluster in terms of duration, frequency, and total energy expenditure. Means and standard deviations or proportions were used to describe the sociodemographic characteristics of the participants within each physical activity cluster for continuous and categorical variables, respectively. Chi-square tests were used to identify significant sociodemographic characteristic differences within each physical activity cluster. Chi-square tests were also used to assess associations between the physical activity clusters with health behaviors and clinical characteristics. Sensitivity analyses were conducted using multiple comparison pairwise chi-square tests with the Bonferroni correction. Statistical significance was set at the alpha level of 0.0033 (0.05/15).

Results

Sample characteristics

The analyses included data from 9,816 participants. Survey weights were applied to the sample to ensure national representativeness. The 2003–2004 and 2005-2005 cohorts were similar in age, gender, race/ethnicity, and educational attainment. The mean age for the 2003–2004 cohort was 50.2 (19.2) years and for the 2005–2006 cohort was 51.5 (19.4) years. Both cohorts were approximately 52% female, 48% male, 52% White, 21% Black, 20% Mexican American, 4% other, and 3% Hispanic. Most of the sample (71%) had a high school degree or higher.

Leisure-time aerobic physical activity clusters identified by multivariate finite mixture modeling

A five-cluster solution for leisure-time aerobic physical activity was determined among the 9,816 adult NHANES participants using a statistically principled data-driven approach based on the BIC (51,844.03), AIC (51,642.82), and optimal entropy (0.717) metrics. The largest number of participants were in the no activity (n = 4,346) and daily frequency/short-duration activity clusters (n =3,011). Participants in the daily frequency/short-duration activity cluster were likely to engage in two different activity types per month and six activity sessions per week lasting about 30 min per session. The next largest activity cluster was the high-frequency/average-duration cluster (n = 964). Participants in the high-frequency/average-duration cluster were likely to engage in 5 different activity types per month and 16 activity sessions per week (>2 per day on average) lasting about 50 min per session. Low-frequency/short-duration cluster participants were likely to engage in one activity type per month and one session per week lasting about 30 min (n = 730). The remainder of the participants were grouped into two similarly sized long-duration activity clusters where the activity sessions were approximately 2 hr. In these long-duration activity clusters, low-frequency participants engaged in one activity type per month and one activity session per week (n = 392). Daily frequency participants engaged in two different activity types per month and five activity sessions per week (n = 373). American College of Sports Medicine guidelines for physical activity were met by the majority of participants in all activity clusters, except for participants in the low frequency/short duration and no activity clusters (see Table 1).

Table 1.

Habitual physical activity patterns reported by U.S. adults (National Health and Nutrition Examination Survey 2003–2006)^a

	No activity	Low frequency Short duration	Low frequency Long duration	Daily frequency Short duration	Daily frequency Long duration	High frequency Average duration
N = 9,816
Number of participants	4,346	730	392	3,011	373	964
Average minutes per session (SD)	0 (0)	29.52 (18.26)	123.47 (76.38)	36.56 (22.62)	141.88 (87.77)	47.80 (29.57)
Average frequency per week (SD)	0 (0)	0.94 (0.70)	0.86 (0.63)	6.00 (4.43)	5.11 (3.77)	15.65 (11.57)
Average total kilocalories (103) per week (SD)	0 (0)	0.16 (0.16)	0.60 (0.59)	1.22 (1.20)	4.44 (4.36)	5.26 (5.17)
Number of activity types (per month)	0	1.19	1.30	1.96	2.16	4.90
Met aerobic PA guidelines (%)	0	12.6	89.3	98.4	100	100
Walking only: N = 2,997
Number of participants	n/a	319	98	1,854	141	585
Average minutes per session (SD)	n/a	26.8 (13.0)	73.6 (41.2)	34.1 (19.6)	129.0 (128.0)	40.0 (29.7)
Average frequency per week (SD)	n/a	0.84 (0.47)	0.69 (0.30)	4.43 (3.70)	4.00 (3.06)	5.45 (5.47)
Average total kilocalories (103) per week (SD)	n/a	0.10 (0.06)	0.24 (0.18)	0.64 (0.60)	3.15 (4.54)	1.10 (1.62)
Met aerobic PA guidelines (%)	n/a	14.1	66.3	86.0	83.0	86.3

SD standard deviation.

^aThe six groups in Table 1 include a no activity group. Participants not reporting any physical activity (PA) were excluded from the multivariate finite mixture modeling.

Walking patterns were characterized within each activity cluster, except the no activity cluster. Walking patterns from the largest to smallest clusters were as follows. Daily frequency/short-duration participants reported four walking sessions per week and 30 min for each walking session (n = 1,854). High-frequency/average-duration participants reported five walking sessions per week and 40 min for each walking session (n = 585). Low-frequency/short-duration participants reported walking once per week for about 30 min (n = 319). In the long-duration activity clusters, daily frequency participants reported four walking sessions per week and 2 hr for each walking session (n = 141): Low-frequency participants reported walking once per week and for about 1 hr (n = 98). American College of Sports Medicine guidelines were met by walking for most participants in the daily frequency and high-frequency activity clusters (e.g., 83%–86%) but to a lesser extent in the low-frequency activity clusters (e.g., 14%–66%; see Table 1).

Summary of physical activities in the 2003–2006 NHANES cohorts

Forty-seven different activity types were reported by participants. Walking was the most common type of activity reported (n = 2,997). Gymnastics and surfing were the least commonly reported activities (n = 2; see Table 2).

Table 2.

Types of activities reported by U.S. adults (National Health and Nutrition Examination Survey 2003–2006)

Activity	n	Average minutes per session (SD)	Average frequency per week (SD)	Total kcal (103)/week (SD)
Walking	2,997	40.43 (47.15)	4.10 (4.09)	0.77 (1.43)
Bicycling	933	40.85 (37.82)	2.60 (3.11)	0.93 (3.11)
Dance	740	58.45 (67.71)	1.58 (2.31)	0.55 (1.07)
Treadmill	730	29.04 (14.90)	2.92 (2.59)	0.70 (0.90)
Weight lifting	683	41.28 (28.74)	2.87 (2.11)	0.55 (0.81)
Aerobics	633	38.77 (22.02)	2.96 (2.53)	0.93 (1.08)
Stretching	556	16.50 (10.41)	4.61 (4.04)	0.25 (0.30)
Running	515	32.43 (20.58)	2.49 (2.80)	1.05 (1.25)
Hiking	503	61.10 (66.50)	2.32 (3.19)	1.14 (2.15)
Basketball	479	76.63 (59.93)	1.76 (2.64)	1.50 (2.64)
Swimming	474	46.65 (34.08)	1.89 (2.30)	0.86 (1.27)
Jogging	466	36.36 (42.94)	2.55 (2.27)	0.84 (1.87)
Stair climbing	424	22.66 (28.46)	6.54 (12.37)	1.41 (3.22)
Golf	352	196.60 (74.28)	1.28 (1.51)	1.22 (1.53)
Fishing	266	201.96 (115.20)	1.04 (1.15)	1.06 (1.31)
Bowling	184	140.83 (50.76)	1.04 (1.98)	0.60 (1.31)
Yoga	173	34.45 (21.35)	2.34 (5.51)	0.25 (0.40)
Soccer	162	72.04 (51.52)	1.47 (1.64)	1.26 (1.42)
Football	116	93.02 (60.13)	1.13 (1.32)	1.21 (2.39)
Tennis	110	76.90 (33.91)	1.59 (3.07)	1.03 (1.52)
Softball	84	107.37 (75.62)	1.39 (1.21)	1.12 (1.78)
Baseball	80	62.98 (46.66)	1.62 (3.95)	0.71 (1.05)
Volleyball	77	75.94 (48.42)	1.14 (1.59)	0.79 (0.98)
Hunting	75	229.04 (146.78)	1.21 (1.32)	2.19 (3.38)
Kayaking	68	87.37 (69.51)	0.75 (0.81)	0.49 (0.65)
Other	63	50.48 (44.22)	2.66 (2.08)	1.06 (1.13)
Frisbee	62	40.29 (42.31)	1.38 (1.79)	0.27 (0.48)
Martial arts	61	57.17 (41.67)	3.15 (2.79)	1.69 (2.03)
Rowing	59	32.80 (48.96)	2.08 (3.44)	0.58 (0.99)
Skating	57	78.82 (59.94)	1.06 (1.22)	0.77 (1.15)
Rollerblading	53	60.05 (49.02)	0.92 (1.06)	0.48 (0.67)
Wrestling	51	48.41 (30.39)	2.55 (2.57)	1.41 (1.85)
Yard work	41	133.56 (120.58)	1.86 (1.60)	1.76 (2.19)
Racquetball	39	71.59 (33.93)	1.39 (1.17)	1.40 (1.71)
Boxing	37	41.33 (26.28)	1.99 (1.96)	1.10 (1.35)
Skiing—cross country	36	28.73 (23.03)	2.57 (1.59)	0.89 (0.75)
Horseback riding	34	102.10 (57.59)	1.24 (1.61)	0.68 (1.15)
Skiing—downhill	26	158.75 (147.59)	0.79 (0.91)	1.27 (1.53)
Gardening	25	85.41 (67.96)	3.42 (2.72)	1.43 (1.74)
Hockey	14	94.81 (38.49)	0.87 (0.60)	0.93 (0.85)
Rope jumping	9	13.37 (4.66)	2.31 (2.09)	0.44 (0.45)
Sit-ups	6	17.56 (7.00)	3.42 (0.95)	0.52 (0.46)
Trampoline jumping	6	46.16 (49.76)	3.85 (1.79)	0.64 (1.04)
Skateboarding	4	184.66 (87.09)	4.26 (2.60)	6.32 (5.55)
Pushups	3	25.83 (19.16)	4.67 (2.07)	0.53 (0.34)
Cheerleading and gymnastics	2	125.00 (145.52)	0.70 (0.33)	0.87 (1.18)
Surfing	2	54 (31.75)	0.58 (0.49)	0.21(0.25)

SD standard deviation.

Sociodemographic characteristics and habitual physical activity patterns

Participants’ sociodemographic characteristics varied significantly within each activity cluster. White men, >49 years of age, with more than a high school education, comprised the largest group in the high-frequency/average-duration and daily frequency/long-duration activity clusters. White women with more than a high school education comprised the largest group in the no activity, low-frequency/short-duration, and daily frequency short-duration activity clusters. The proportion of young adults, 21–29 years of age, increased across the increasing activity clusters (e.g., no activity cluster = 11%; high-frequency/average-duration activity cluster = 27%).

Males comprised a greater percentage of the long-duration activity clusters than females: 77% daily frequency/long duration and 64% low frequency/long duration for males versus 23% daily frequency/long duration and 36% low frequency/long duration. Females comprised a greater percentage of the short-duration activity clusters than males: 58% daily frequency/short duration and 57% low frequency/short duration for females versus 42% daily frequency/short duration and 43% low frequency/short duration for males.

Mexican American participants decreased across the increasing activity clusters (e.g., no activity cluster = 11%; high-frequency/average-duration activity cluster = 4%). White participants increased across the increasing activity clusters (e.g., no activity cluster = 66%; high-frequency/average-duration activity cluster = 76%). Adults with less than a high school education decreased across the increasing activity clusters (e.g., no activity cluster = 29%; high-frequency/average-duration activity cluster = 7%). The proportion of adults with more than a high school education increased across the increasing activity clusters (e.g., no activity cluster = 71%; high-frequency/average-duration activity cluster = 93%; see Table 3).

Table 3.

Sociodemographic characteristics within habitual physical activity patterns for U.S. Adults (National Health and Nutrition Examination Survey 2003–2006)^a

	Total sample	No activity	Low frequency Short duration	Low frequency Long duration	Daily frequency Short duration	Daily frequency Long duration	High frequency Average duration
Number of participants	9,816	4,346	730	392	3,011	373	964
Met aerobic PA guidelines (%)	30	0	12.6	89.3	98.4	100	100
Age (years)***		a b c d e	a f	b g	c f h	d i	e g h i
21–29	15%	11%	18%	17%	14%	21%	27%
30–39	20%	19%	25%	26%	19%	21%	24%
40–49	22%	21%	26%	22%	24%	22%	22%
50–59	18%	18%	17%	18%	19%	17%	16%
60–69	11%	13%	8%	8%	12%	10%	7%
70+	12%	18%	7%	9%	11%	10%	4%
Gender***		a b c d	e f g	a e h i	b h j k	c f i j l	d g k l
Female	52%	53%	57%	36%	58%	23%	45%
Male	48%	47%	43%	64%	42%	77%	55%
Race***		a b c			a	b	c
Black	11%	13%	10%	9%	10%	9%	12%
Mexican American	8%	11%	9%	8%	6%	5%	4%
Other	5%	6%	5%	5%	6%	7%	4%
Other Hispanic	3%	4%	3%	4%	3%	4%	4%
White	72%	66%	73%	74%	76%	75%	76%
Education***		a b c d e	a f	b g	c h	d i	e f g h i
<High school	18%	29%	13%	13%	12%	16%	7%
High school or greater	82%	71%	87%	87%	88%	84%	93%

^aThe six groups in Table 2 include a no activity group. Participants not reporting any physical activity (PA) were excluded from the multivariate finite mixture modeling.

***p < 0.001.

^{a b c d e f g h I j k l}Corresponding letters indicate a statistically significant p value <0.0033 (0.05/15) of the multiple comparison pairwise chi-square test using the Bonferroni correction.

Health behaviors, clinical characteristics, and physical activity patterns

Participant health behaviors and clinical characteristics varied significantly within each activity cluster. Nonsmokers and moderate alcohol users, as well as adults with no mobility problems, comprised a greater percentage of all activity clusters (vs. the no activity cluster). Participants without chronic health conditions, specifically diabetes, heart failure, coronary heart disease, and stroke, comprised a greater percentage of all activity clusters. The percentage of participants with mobility problems decreased across increasing activity clusters (e.g., 29% = no activity, 4% = high frequency/average duration). The percentage of participants with diabetes also decreased across increasing activity clusters (e.g., 11% no activity, 4% high frequency/average duration; see Table 4).

Table 4.

Health behaviors and clinical characteristics associated with habitual physical activity patterns for US Adults (National Health and Nutrition Examination Survey 2003–2006)^a

	Total sample	No activity	Low frequency Short duration	Low frequency Long duration	Daily frequency Short duration	Daily frequency Long duration	High frequency Average duration
Number of participants	9,816	4,346	730	392	3,011	373	964
Smokes cigarettes***		a b c	a		b d	d e	c e
Current	25%	31%	24%	25%	19%	33%	19%
Former	25%	25%	23%	26%	27%	23%	25%
Never	50%	44%	54%	49%	54%	44%	56%
Alcohol use***		a b c d e	a f	b	c g	d g h	e f h
Heavy	8%	9%	8%	11%	7%	14%	9%
Moderate	62%	52%	64%	66%	66%	64%	76%
None	29%	39%	28%	23%	27%	22%	15%
Mobility problems***		a b c d e	a f	b g	c h i	d h	e f g i
Any difficulty	17%	29%	12%	9%	12%	7%	4%
No difficulty	83%	71%	88%	91%	88%	93%	96%
Body mass index***		a b			a		b
Ideal	31%	28%	28%	32%	34%	29%	34%
Obese	33%	38%	36%	28%	30%	36%	28%
Overweight	34%	32%	34%	38%	34%	34%	36%
Underweight	2%	2%	2%	2%	2%	1%	1%
Diabetes***		a b c d	a e		b e	c	d
Yes	8%	11%	5%	7%	7%	8%	4%
Borderline	1%	2%	2%	2%	1%	1%	1%
No	91%	87%	93%	92%	92%	91%	95%
Don’t know	0%	0%	0%	0%	0%	0%	0%
Congestive heart failure***		a b			a		b
Yes	3%	5%	2%	2%	2%	2%	1%
No	97%	95%	98%	98%	98%	98%	99%
Don’t know	0%	0%	0%	0%	0%	0%	0%
Coronary heart disease*
Yes	4%	5%	3%	3%	3%	2%	3%
No	96%	95%	96%	97%	96%	97%	97%
Don’t know	0%	1%	0%	0%	0%	0%	0%
Cancer***		a	b		b c		a c
Yes	9%	9%	7%	8%	10%	9%	5%
No	91%	91%	93%	92%	90%	91%	95%
Don’t know	0%	0%	0%	0%	0%	0%	0%
Stroke***		a b		a	c		b c
Yes	3%	4%	3%	2%	2%	1%	1%
No	97%	95%	97%	98%	98%	99%	99%
Don’t know	0%	0%	0%	0%	0%	0%	0%

^aThe six groups in Table 4 include a no activity group. Participants not reporting any physical activity (PA) were excluded from the multivariate finite mixture modeling.

*p < 0.05, ***p < 0.001.

^{a b c d e f g h i}Corresponding letters indicate a statistically significant p value < 0.0033 (0.05/15) of the multiple comparison pairwise chi-square test using the Bonferroni correction.

Discussion

This study identified five distinct patterns of aerobic activity participation, with walking being the most common activity, in a nationally representative sample of adults. High-activity clusters were comprised of a greater proportion of adults who were younger, male, White, and with at least a high school education. Of note was that highly active adults engaged in a greater variety or number of activity types. Additionally, a greater proportion of females engaged in high/daily frequency and short-duration activity bouts. In terms of health behaviors, nonsmokers and moderate alcohol users, as well as adults with no mobility problems, comprised a greater percentage of all activity clusters (vs. the no activity cluster). Participants without chronic health conditions, specifically diabetes, heart failure, coronary heart disease, and stroke, comprised a greater percentage of all activity clusters. Nonetheless, the small percentage of participants with chronic conditions across all activity clusters suggests that adults with chronic conditions are incorporating physical activity recommendations into their lifestyle. In sum, these data provide a more nuanced understanding about the sociodemographic, health behavior, and clinical characteristics of habitual physical activity participation.

Consistent with previous data, almost half (44%) of this sample engaged in no aerobic physical activity in the last month [25]. The no physical activity subgroup in our analyses was predominantly comprised of White women. All, or almost all, of the adults in the high-frequency/average-duration, the daily frequency/long-duration, and the daily frequency/short-duration clusters met the aerobic physical activity guidelines; combined, these clusters represented 44% of the sample. A key characterizing factor of these active clusters was that the participants reported engaging in at least two different activities in the last month; the most active participants reported participating in five. Previous reports from NHANES data showed that 43% of the sample participated in one or two activities [26]; the current data extend this work to suggest that engaging in more (>2) different types of aerobic activities was characteristic of high levels of activity. Meanwhile, walking was reported as the most common form of activity, reaffirming previous work showing the popularity of walking as a physical activity type. However, within all but one of the active clusters, when adults reported walking as their only form of activity, a lower proportion was found to meet the aerobic activity guidelines; this has been reported in previous studies also [27, 28].

Another new finding revealed by these data was that, across the different activity clusters, a considerably higher proportion of females than males were a part of the daily activity/short-duration cluster (58% vs. 42%, respectively). Within this cluster, participants reported being active for an average of 36.56 min on 6 days of the week, and 62% engaged in only walking. A plethora of literature demonstrates that following puberty, the decline in physical activity participation is more pronounced in females than males [29–31]; thus, that the daily frequency/short-duration cluster should be so highly populated by females is meaningful. There are several possible reasons why the daily frequency/short-duration activity pattern may be more attainable to females as compared to males. Females have less leisure time; American women report an average of 13.2 hr of household labor per week as compared to 6.6 hr per week for men [32]. Females are more likely than males to report barriers to being active and reduced control over their decision to engage in physical activity [33, 34]. Thus, shorter duration bouts of activity may be more achievable to females who are more likely to experience scheduling and social barriers to activity. Given that 98.4% of adults in this cluster met the aerobic physical activity guidelines and evidence that repeated short bouts of activity throughout the day reduces all-cause mortality [35], physical activity programs targeting inactive females should emphasize the attainability and benefits of engaging in shorter, more frequent bouts of activity.

Another key finding from this study was that adults with poor health behaviors (i.e., current smoker) and chronic health conditions (i.e., diabetes) were more prevalent in the inactive cluster. These data add to evidence showing how negative health behaviors tend to co-occur; for example, 70% of adults who engage in one negative health behavior also engage in a second, while one in five engage in three or more [36, 37]. Of specific relevance to physical activity, mid-life smoking has been shown to predict lower levels of physical activity [38], while adults with diabetes and coronary heart disease are significantly less likely to be active. Increased physical activity levels have enormous positive effects on improving other health behaviors and reducing comorbid effects of chronic conditions. For example, increased physical activity has been associated with reduced tobacco use and even increased odds of tobacco cessation [39–41]. Moreover, adults with diabetes and/or coronary heart disease can delay progression or worsening of symptoms by even incremental increases in activity [42]. Data from the current study converge with guidelines to promote walking interventions in these high-risk populations [43].

Data from the current study should be interpreted with consideration of some methodological limitations, including the use of cross-sectional, self-reported data from 2003 to 2006. First, these cross-sectional data cannot identify causal relationships. Second, self-reported physical activity may overreport activity [44]. In this study, the high-frequency/average-duration cluster reported approximately 770 min of activity per week. This high level of physical activity may be attributed to the inclusion of leisure-time physical activities, such as golf, fishing, hunting, or kayaking, that are not often accounted for in other physical activity surveys. Others have reported approximately 500 min of leisure-time physical activity per week in a highly active cluster of older adults with a mean age of 69 years [19]. Although accelerometers estimate physical activity, accelerometers underestimate certain types of activities, such as cycling, swimming, and weight lifting. Cycling was the second most common activity type reported in NHANES 2003–2006 participants (see Table 2). Third, these data were collected approximately 15 years ago. It is possible that there have been changes in specific aspects of physical activity, such as the prevalence or types of leisure-time activities. However, self-reported adherence to physical activity guidelines has been reported as unchanged over the past decade [45]. Specifically, older adults and adults with lower educational levels have remained less likely to meet physical activity guidelines [45]. These findings are similar to our findings whereby older adults and adults achieving lower educational levels are more likely to be in the no activity subgroup versus an active cluster. Finally, this study did not consider other factors known to relate to physical activity levels, including sleep duration [46] and depressive symptoms [47].

In sum, a greater variety of physical activities associates with higher-activity clusters and with meeting aerobic physical activity recommendations. This finding highlights the importance of assessing multiple dimensions of physical activity. Queries evaluating the number and type of individual activities add nuanced information that may allow providers to tailor comprehensive physical activity recommendations to patients in the future. Population-based surveys, such as NHANES, should include physical activity queries that evaluate individual activities to continue to advance our understanding of these physical activity dimensions for health and well-being.

Funding

This study was funded by the National Institute of Nursing Research (K99NR017416), grant support for Dr. Freda Patterson: National Institute on Minority Health and Health Disparities of the National Institutes of Health (R01MD012734) (FP), Institutional Development Award (IDeA) Center of Biomedical Research Excellence from the National Institute of General Medical Sciences of the National Institutes of Health (P20GM113125). Support for Dr. Susan Malone: National Heart, Lung, and Blood Institute (T32HL7953).

Compliance with Ethical Standards

Conflicts of Interest: Susan Malone, Freda Patterson, Laura Grunin, Gail Melkus, Barbara Riegel, Naresh Punjabi, Gary Yu, Jacek Urbanek, Ciprian Crainiceanu, and Allan Pack declare that they have no conflicts of interest.

Authors’ Contributions: F.P. and S.K.M wrote the manuscript with support from all authors. G.Y. conceived the idea and supervised the analyses of the data. L.G. performed the computations. J.U. provided the code for the analyses of these data. G.D.M., B.R., and N.P. contributed to the final version of the manuscript. A.P., B.R., and C.C. focused our research ideas culminating in the decision to access these data for these analyses.

Ethical Approval: For this retrospective study, formal consent is not required. This article does not contain any studies with animals performed by any of the authors.

Informed Consent: The de-identified data used in these secondary analyses were from participants who provided informed consent using consents and protocols approved by the National Center for Health Statistics Research Ethics Review Board.