Obesity increases risk of declining physical activity over time in women: a prospective cohort study

Authors


  • Author contributions: JT and LT conceived the study. LT organized and supervised the study and analyzed the data. JL, BB, and JT carried out the study. JT had primary writing responsibilities, but all authors were involved in writing the paper and had final approval of the submitted and published versions.

  • Disclosure: The authors have no conflicts of interest to disclose.

Abstract

Objective

Research indicates that risk of obesity increases as physical activity (PA) decreases; however, the reciprocal effect has been rarely studied. The present investigation was conducted to determine the contribution of obesity on objectively measured PA over 20 months.

Design and Methods

A prospective cohort design with 254 middle-aged women was employed. Body fat percentage (BF%) was measured using Bod Pod, and obesity was defined as BF% ≥32%. PA was assessed objectively using 7-day accelerometry at baseline and ∼20 months later at follow-up.

Results

Of the 254 subjects, 124 were obese (49%) at baseline. Mean BF% was 32.1 ± 7.8 and average age was 41.7 ± 3.1 years. Mean weekly PA was 2.79 ± 0.85 million activity counts for all participants. Over the 20-month period, PA decreased significantly more in obese women (−8.1% ± 27.1%) than in nonobese women (0.3% ± 31.7%) after adjusting for confounders (F = 5.3, P = 0.022). Moderate plus vigorous PA levels also decreased more in obese women (−28.1 ± 73.6 min/week) than in nonobese women (−5.9 ± 66.8 min/week), after adjusting for covariates (F = 7.84; P = 0.0055).

Conclusions

It appears that obese women tend to reduce PA over time at a faster rate than nonobese women. Evidently, obesity is a risk factor for decreasing PA over time in middle-aged women.

Introduction

There is an inverse, dose-response association between physical activity (PA) and obesity. The PA Guidelines for Americans Committee Report suggests that of the 24 cross-sectional studies that have examined PA and obesity, 23 reported a significant, inverse relationship [1]. Furthermore, when participants have been segmented into multiple activity levels based on PA volume (e.g., MET-hours per week), incremental differences in BMI have been confirmed, supporting a dose-response connection [2].

It is possible that the relationship between PA and obesity is reciprocal. However, the cross-sectional nature of the aforementioned studies prohibits conclusions regarding temporality; that is, whether PA reduces obesity or vice versa. In contrast, prospective investigations are useful for examining the temporality of a given relationship, and several such studies exist supporting the role of PA on future weight change [3]. A recent meta-analysis comprised of 120,877 US men and women from three separate cohorts showed that, while baseline PA levels were not predictive of weight change, participants in the upper quintile for PA change gained 1.8 fewer pounds over a 4-year period, on average, when compared to the lowest quintile of PA change [11].

The current literature evaluating the impact of obesity on future PA levels is sparse, though available evidence seems to support an inverse association [12]. However, one significant limitation of the available literature regarding the role of obesity in PA change is that it has relied exclusively on self-reported PA levels. Furthermore, the majority of the research has relied on BMI to assess overweight/obesity status. These assessment methods, while typical for large cohort studies, are prone to measurement error and misclassification [16, 17]. To date, no investigation has examined the contribution of obesity on changes in future activity levels using high-quality measurement methods, including a well-validated body composition estimate and objectively measured PA.

Therefore, the primary purpose of the present investigation was to determine the contribution of obesity on objectively measured PA over 20 months in adult women. A secondary aim was to assess the influence of several potential confounding factors, including age, baseline PA, season of assessment, and days between assessments, on the obesity-PA association over time.

Methods

Subjects and procedures

Participants included 275 middle-aged women at baseline who were recruited from approximately 20 cities in two metropolitan areas in the Mountain West through emails, flyers, and newspaper advertisements. Women were excluded from the study if they were pregnant or planning to become pregnant, current smokers, or chronically ill. Participants made two visits to the research lab at the university for baseline assessments. During the first visit, participants signed an informed-consent document, reported demographic information, and had their body composition measured. Before leaving, participants were trained regarding how to wear an accelerometer-based PA monitor. Participants wore the activity monitors for 1 week and then returned them to the research lab. During the interim between the baseline and follow-up assessments, no attempt was made to influence participants' behavior. Approximately 20 months after the baseline measures, 254 participants revisited the research lab and repeated the body composition and PA assessments. A total of 21 participants (8%) were lost to follow-up. All statistical analyses and results are based on the 254 participants who completed both the baseline and follow-up assessments. The University's Institutional Review Board (IRB) approved all study procedures.

Physical activity

To assess PA, participants were asked to wear the Actigraph (ActiGraph LLC, Pensacola, FL) model 7164 accelerometer for seven consecutive days. Monitors were attached to custom-fitted elastic belts and issued to each participant with instructions to wear the monitor over the left hip. Participants were asked to wear the accelerometer at all times, with the exception of water-based activities, such as swimming and bathing, since the monitors were not waterproof. Wear-time requirements included a minimum of 12 h per day [19], and participants who did not meet this requirement were asked to re-wear the device on the same day(s) of the week as the day(s) in question. Mean wear-time from 7 am to 10 pm (a 15-h period) across the 7 days was 13.9 h (93% wear-time compliance). Total activity counts were used as an index of PA (i.e., total movement), as employed in a number of investigations [20]. The difference between total activity counts at baseline and follow-up was used to index change in PA, expressed in absolute counts as well as a percentage.

Time spent in light, moderate, and vigorous intensity PA was also assessed using the accelerometer. Based on pilot data and previous research [22, 26, 27], a cut-point of 1000-2999 activity counts/min was used to assess light-intensity PA. Moderate-intensity PA, such as walking or active housework, was assessed using a cut-point of 3000-4999 counts/min, and ≥5000 counts was used to represent vigorous PA, such as jogging and aerobic dancing.

Body fat percentage

Baseline body fat percentage (BF%) was estimated objectively via air displacement plethysmography using the Bod Pod (COSMED USA, Inc. Concord, CA). Bod Pod measurements are repeatable. Using a sample of 100 women from the present study, a test-retest reliability assessment was performed with complete repositioning. Results showed an intraclass correlation of 0.999 (P < 0.0001) between the two measurements [28]. Similarly, concurrent validity has been shown by comparing BF% estimates in female adults using the Bod Pod to dual-energy X-ray absorptiometry findings, resulting in an intraclass correlation of 0.97 (P < 0.0001) [29].

Participants were asked to void and change into a standard, lab-issued swimsuit and swim cap, after which duplicate BF% measures were taken. Participants were instructed to remain still and breathe normally while seated in the capsule. If necessary, additional measures were taken until two estimates were within one percentage point of each other, after which the average of these measures was used to indicate BF%. For 75% of the subjects, only two Bod Pod measurements were taken. Obesity was defined as BF% ≥32%, as recommended by the American Council on Exericise [30].

Season of assessment

Previous research has shown that objectively measured PA tends to vary according to the season of the year [31, 32]. In short, there is a trend towards lower levels of PA and greater levels of sedentary behavior during the winter months when compared to other months. Consequently, season of assessment was measured and controlled in the present study. Given the climate of the Mountain West, fall was defined as September, October, and November and winter included the months December, January, and February. Spring was defined as March, April, and May, and summer included the months June, July, and August.

Data analysis

A power analysis was conducted using the PASS 6.0 statistical software (NCSS, Kaysville, UT) to determine the number of participants needed for a 2 × 2 contingency table using χ2 to detect a small effect size (Cohen's W) of 0.20 with alpha at the 0.05 level. The 254 subjects employed in the present study resulted in power of 0.89. Likewise, another analysis was performed to identify the number of subjects needed to detect a small effect size of 0.20 when comparing the means of two groups (obese and nonobese) using one-way ANOVA. Findings indicated that the 254 participants of this investigation resulted in power of > 0.99. In short, the sample size was more than adequate and statistical power was very good to excellent (>0.80) for the analyses of this investigation.

Means and standard deviations were calculated for all baseline and follow-up descriptive characteristics. Change in PA and intensity of PA were calculated by subtracting baseline values from follow-up values. To calculate percent change in PA, the difference in PA counts was divided by the original activity counts. Changes in PA and PA intensity levels were compared across obese and nonobese categories using the GLM procedure. Partial correlations were also calculated using the GLM procedure in order to examine the influence of specific potential confounding factors on the obesity and PA associations. These factors included age, baseline PA or baseline PA intensity level, season of assessment, and the number of days between assessments, considered individually and collectively. Mean PA and PA intensity levels, adjusted for differences in the potential confounders, were computed using the least-squares means procedure. To determine the extent to which a substantial decrease in PA (≥20%), treated as a categorical variable, differed by obesity status, estimates of relative risk were calculated using incidence data, comparing those with a substantial PA decrease to those without a substantial PA decrease. All statistical analyses were performed using SAS software (version 9.3, SAS Institute, Inc., Cary, NC). Alpha was set at 0.05 to determine significance.

Results

Participants' descriptive characteristics are presented in Table 1. At baseline, mean (±SD) BF% was 32.1 ± 7.8% and average age was 41.7 ± 3.1 years. A total of 124 participants (49%) were categorized as obese (BF% ≥32%). Mean baseline PA estimates consisted of 2.79 ± 0.85 million accelerometer counts for all participants, with significantly lower baseline PA levels among obese (2.66 ± 0.72 million counts) when compared to nonobese (2.92 ± 0.95 million counts) women (F = 6.25; P = 0.0131). This baseline PA difference remained significant after adjusting for age and season of assessment (F = 6.39; P = 0.0121).

Table 1. Descriptive information for all participants (n = 254)
VariablesMeanSD25th Percentile50th Percentile75th Percentile
  1. Values represent measurements at baseline, unless indicated otherwise.
  2. For change in PA (counts/week) and change in PA (%), negative values reflect decreases in PA over time.
Age (years)41.73.1394244
Body mass index23.53.320.923.225.8
Body fat percentage32.17.826.232.138.2
Total PA (counts/week)2,789,711849,8902,130,8572,759,8103,264,164
Days between assessments59166555581626
Change in PA (counts/week)−162,726762,181−574,994−145,108247,710
Change in PA (%)−3.329.6−20.2−5.810.5

Change in PA varied considerably with a mean decrease of 0.16 ± 0.76 million counts after 20 months. This decline in counts was equivalent to an overall reduction of 3.3% ± 29.6% in PA from baseline to follow-up. Likewise, participants reduced time spent in light (−40.0 ± 259.8 min/week), moderate (−13.0 ± 55.3 min/week), and vigorous (−3.4 ± 42.9 min/week) intensity PA, on average.

Changes in both total accelerometer counts and count-based percentages for obese and nonobese participants are displayed in Table 2. Obese women decreased PA levels by 6.3% ± 27.1%, whereas nonobese women decreased by 0.2% ± 31.7%. This unadjusted difference in PA change was not statistically significant (F = 2.7; P = 0.101). However, after adjusting for baseline PA, obese women had a significantly greater decrease in PA than nonobese women (F = 6.2; P = 0.014). This difference remained significant after adjusting for all of the potential confounding factors, including baseline PA, age, the season in which PA was assessed, and the number of days between assessments (F = 5.5; P = 0.020). Specifically, the fully adjusted PA levels decreased in obese women (−8.1% ± 27.1%) and slightly increased in nonobese women (0.3% ± 31.7%).

Table 2. Absolute and percentage change in total accelerometer counts over time in obese and non-obese women
 Obese n = 130Non-obese n = 124  
Outcome: PA change (%)MeanSDMeanSDFP
  1. Percent change in PA reflects the change in objectively measured activity counts from baseline to follow-up approximately 20 months later, expressed as a percentage.
  2. Means on the same row as a potential confounder are adjusted for differences in that covariate.
  3. DBA, days between assessments; PA, physical activity.
Variable Controlled:      
None−6.3%27.1%−0.2%31.7%2.70.1009
Baseline PA−7.6% 1.2% 6.20.0135
Baseline PA and age−7.5% 1.1% 5.60.0190
Baseline PA and season−8.2% 0.5% 6.00.0148
Baseline PA and DBA−7.7% 1.3% 6.30.0127
All of the covariates above−8.1% 0.3% 5.50.0200
Outcome: Total accelerometer count change
None−231,242717,379−90,894803,1202.20.1427
Baseline PA−269,701 −50,575 5.80.0172
Baseline PA and age−267,642 −52,733 5.40.0210
Baseline PA and season−287,510 −72,864 5.60.0184
Baseline PA and DBA−271,774 −48,402 6.00.0153
All of the covariates above−288,688 −74,302 5.50.0202

Similar trends were seen when comparing total accelerometer count changes (instead of percent changes), among obese and nonobese participants. Unadjusted, total count changes did not differ significantly between obese and nonobese women (F = 2.2; P = 0.1427); however, obese women had significantly larger declines in accelerometer counts than nonobese women after adjusting for baseline PA (F = 5.2; P = 0.0172), as well as after adjusting for all confounders (F = 5.5; P = 0.0202).

When changes in time spent in different PA intensities (light, moderate, and vigorous) were compared between obese and nonobese participants, no differences were seen in unadjusted PA changes for light (F = 1.78; P = 0.1836), moderate (F = 1.72; P = 0.1915), or vigorous (F = 0.22; P = 0.6404) PA. These differences remained insignificant after adjusting for potential confounders, though all intensity categories showed a trend toward greater PA reductions among obese women (data not shown). However, when time spent in moderate and vigorous PA was combined (MVPA), obese women had significantly larger reductions than nonobese women, after adjusting for baseline MVPA (F = 7.48; P = 0.0067), as well as all potential confounders (F = 7.84; P = 0.0055). Specifically, fully adjusted MVPA levels decreased by 28.1 ± 73.6 min/week among obese women and decreased by 5.9 ± 66.8 min/week among nonobese women, on average.

In addition, nonobese women were 39% less likely to decrease PA substantially (≥20%) over the 20-month period when compared to obese women (RR = 0.61, 95% CI = 0.40-0.95), as shown in Table 3. Risk was further reduced after adjusting for differences in baseline PA (RR = 0.54, 95% CI = 0.34-0.85), and remained significant after controlling for age, season, and time between assessments in the model, as well as baseline PA (RR = 0.56, 95% CI = 0.36-0.88).

Table 3. Risk of decreased physical activity (≥ 20%) over time in obese compared to nonobese women
 Obese n = 130Nonobese n = 124
Outcome: Decreased PAaRRRR95% CI
  1. aParticipants were assigned to the “Decreased PA” category if their accelerometer counts decreased by 20% or more from baseline to follow-up, placing them in the lowest quartile for change in PA. Women in the lowest quartile for change in PA (n = 65) were compared to all other participants (n = 189).
  2. DBA, number of days between assessments; PA, physical activity; RR, relative risk.
  3. Relative risks on the same row as a potential confounder are adjusted for differences in that covariate.
Variable controlled:   
None1.00.610.40–0.95
Baseline PA1.00.540.34–0.85
Baseline PA and age1.00.550.35–0.87
Baseline PA and season1.00.550.35–0.86
Baseline PA and DBA1.00.540.34–0.85
All of the covariates above1.00.560.36–0.88

Discussion

Results from the current study suggest that nonobese women have significantly smaller reductions in PA over approximately 1.5 years when compared to obese women. According to adjusted accelerometer counts, obese women reduced their PA levels by ∼8%, whereas nonobese women maintained their PA levels, on average. It appears that this difference in PA change is because of, at least in part, greater reductions in MVPA among obese compared to nonobese women. In addition, we found that nonobese women have a 44% lower risk of substantially decreasing (≥20%) PA when compared to obese women, after adjusting for covariates. These findings seem consistent with the limited number of longitudinal studies suggesting that obesity increases risk of reducing PA levels over time ([12]).

One such study by Peterson et al. ([12]) compared self-reported leisure-time PA changes over a 10-year period in a cohort of almost 6000 Copenhagen adults (2443 men and 3403 women). To assess the impact of weight status on PA, participants were divided into quintiles based on BMI, after which the odds of being classified as physically inactive (<2 hrs/week) were compared between BMI quintiles. Those in the highest BMI group had an increased adjusted odds of becoming inactive (OR = 1.87 for women; OR = 1.48 for men), suggesting a nearly 50% (in men) and 90% (in women) higher likelihood of becoming physically inactive when compared to participants with a BMI near the median. Peterson and colleagues concluded that BMI is a strong predictor of becoming physically inactive 10 years later, even after controlling for potential confounders.

At least two other studies have also used differences in baseline BMI levels to demonstrate that overweight individuals are more likely to reduce PA over time ([13, 14]). Bak and colleagues ([13]) examined the effects of leisure-time and occupational PA, measured using a self-administered questionnaire, on later obesity in 1143 young obese men and 1278 nonobese controls, and also examined the reciprocal effect of obesity on later PA levels. Results showed that baseline PA was not a significant predictor of nonobese individuals becoming obese or of obese individuals maintaining obesity over the 10-year study period. However, elevated baseline BMI levels were predictive of increased physically inactivity over time, suggested by an increasing trend (P < 0.02) in the odds for physical inactivity at follow-up when controls were categorized according to baseline BMI.

Mortensen et al. ([14]) also used baseline BMI to predict physical inactivity (0 h/week of leisure PA), measured using a single question, in a cohort of men and women (n = 4595) from the University of North Carolina Alumni Heart Study. Unlike the Peterson and Bak studies, changes in BMI and lifestyle factors were tracked in yearly and biyearly waves, thereby reducing the likelihood of significant changes in baseline predictor variables over the study's duration. Study results suggested baseline inactivity did not predict changes in BMI; however, baseline BMI was found to increase risk of becoming physically inactive. Specifically, results showed significantly increased odds of becoming inactive (OR = 1.04-1.13 depending on age) per 1 kg/m2 increase in baseline BMI.

More recently, Lakerveld et al. ([15]) categorized 4841 Australian men and women according to abdominal obesity status (measured via waist circumference) to compare changes in PA, measured via interview-administered surveys, over 5 years. Results suggested that adults with abdominal obesity had approximately 1.4 times higher odds (OR = 1.40 in men, OR = 1.44 in women) of reducing PA levels when compared to individuals with a normal waist circumference.

The current study was different from those described above in that it used a high-quality measure of body composition to index obesity, and assessed changes in PA objectively using accelerometry, rather than questionnaire. Furthermore, the study design consisted of a relatively short follow-up period (20 months) in order to increase the likelihood that the baseline PA measure was representative of typical levels throughout the study duration.

Other strengths of this study were the direct measurement of incidence data from baseline to follow-up allowing risk to be measured directly, rather than employing the odds ratio, which does not assess risk, only association. Moreover, statistical control for differences in baseline PA, age, season, and time between baseline and follow-up assessments was performed.

Adjusting for differences in baseline PA levels resulted in a stronger association between obesity and PA change over time. In short, if all participants had the same PA level at baseline, the difference in average PA decrease between obese and nonobese women would have been even greater. This makes sense in theory, as obese women tended to have lower baseline PA levels, and therefore, had less room to further decrease PA over the study duration.

The study was not without weaknesses, however. One limitation was that the study design was prospective in nature, making cause-and-effect conclusions unsuitable. In addition, the current sample lacked demographic diversity because of the fairly homogenous nature of the Mountain West from which the sample was recruited (i.e., ∼90% of the sample identified themselves as non-Hispanic White). Therefore, generalizability of the results may be limited to individuals with similar characteristics.

Although cohort studies with very large sample sizes are typically impressive, the measurement methods of these investigations are usually less refined and precise, and measurement error is often greater. For example, BMI has been shown to misclassify obesity at a high rate, especially as age increases ([16, 17]). This common tendency to misclassify obesity because of elevated BF% despite healthy weights has even led researchers to coin the term “normal weight obesity” ([33]).

To date, no studies investigating the extent to which obesity contributes to decreased PA over time have indexed obesity using BF% instead of BMI, and none have measured PA employing an objective measure, such as accelerometry. The present study was designed to have good statistical power (>0.80) and to employ high-quality measurement methods. Consequently, several important and significant findings were revealed.

In summary, obese participants, those with BF% ≥32% at baseline, decreased their PA significantly more than women with BF% levels < 32%, after adjusting for potential confounders (F = 5.5, P = 0.020). Specifically, obese women experienced an 8.1% decrease in total PA, including a 28.1 min/week reduction in MVPA; whereas nonobese women largely maintained their total PA, with only a 5.9 min/week decrease in MVPA over the 20-month study, on average. A similar trend was evident when examining women with substantial decreases in PA, in that nonobese women had almost half the risk (RR = 0.56) of decreasing PA by 20% or more when compared to obese women, after adjusting for covariates.

While it is likely that the primary mechanism for PA's influence on obesity is related to energy expenditure, the mechanisms for obesity's influence on PA are less clear. Obesity may increase barriers to PA through increased discomfort from osteoarthritis and functional limitations, as well as increased perceived exertion for a given activity. Additional research is needed to help elucidate these underlying mechanisms. Doing so will allow future PA promotion to target these barriers and potentially improve their success among obese individuals.

In conclusion, these findings indicate that obesity is a risk factor for decreasing PA over time in middle-aged women, suggesting that obesity and physical inactivity are reciprocal in nature. Our results appear consistent with trends throughout the United States, which show an increasing prevalence of obesity among adults ([33]), with a concomitantly low percentage of Americans meeting the recommended PA guidelines when measured objectively ([34]). Indeed, it seems likely that the large proportion of inactive adults is contributing to the ongoing rise in obesity. In addition, the current findings support the possibility that the large proportion of obese adults in USA may be further diminishing PA levels. If the cyclical nature of the obesity-PA association is confirmed, targeting obesity as a potential mechanism may enhance the effectiveness of future efforts to promote PA.

Ancillary