Disclosure: The authors declared no conflict of interest.
Version of Record online: 12 MAR 2013
Copyright © 2012 The Obesity Society
Volume 21, Issue 1, pages E78–E85, January 2013
How to Cite
Østergaard, J.N., Grønbæk, M., Ängquist, L., Schnohr, P., Sørensen, T.I.A. and Heitmann, B.L. (2013), Combined influence of leisure-time physical activity and hip circumference on all-cause mortality. Obesity, 21: E78–E85. doi: 10.1002/oby.20062
Funding agencies: The Danish Heart Foundation supported this research.
- Issue online: 12 MAR 2013
- Version of Record online: 12 MAR 2013
- Accepted manuscript online: 3 OCT 2012 05:18PM EST
- Manuscript Accepted: 30 JUL 2012
- Manuscript Received: 3 NOV 2011
- The Danish Heart Foundation supported this research
Hip circumference has been shown to be inversely associated with mortality. Muscle atrophy in the gluteofemoral region may be a possible explanation and thus physical activity is likely to play an important role.
To estimate the combined effects of hip circumference and physical activity on mortality.
Design and Methods:
From the Copenhagen City Heart Study, 3,358 men and 4,350 women aged 21 to 93 years without pre-existing diagnosis of diabetes, stroke, ischemic heart disease, or cancer in 1991-1994 and with complete information on the variables of interest were included in the analyses. The participants were followed to 2009 in the Danish Civil Registration System, with 1.3% loss to follow-up and 2,513 deaths. Hazard ratios (HR) were estimated for combinations of physical activity and hip circumference.
Hip circumference was inversely associated with mortality irrespective of being physically active or not. However, being physically active seemed to counterbalance some of the adverse health effects of a small hip circumference; when comparing inactive to active, the excess mortality at the 25th percentile of hip circumference is 40% in men (HR = 1.40, 95% CI: 1.14-1.72) and 33% in women (HR = 1.33, CI: 1.10-1.62). These associations were observed after adjustment for waist circumference and weight change in the 6 months before the examination.
Less effects of physical activity were found in individuals with greater hip circumferences. A small hip circumference appears hazardous to survival. However, being physically active may counterbalance some of the hazardous effects of a small hip circumference.
Overweight and obesity are associated with increased risk of type 2 diabetes, cancer, cardiovascular diseases and all-cause mortality (1, 2). Yet, how to measure overweight and obesity is widely debated and fat distribution is likely to be more important to predict longevity than body mass index (BMI) or total body fat. Further, it has been proposed only to concentrate on waist circumference and to discard hip circumference, as this measure is a combined measure of bone structure, fat and muscle mass. However, recent studies found that a small hip circumference independent of the other measures of body size, shape andcomposition predicted morbidity and mortality (3-8). Metabolic disturbance due to muscle atrophy in the gluteofemoral region among those with narrow hips has been proposed as explanations (9).
Type 2 diabetes is associated with mortality, and numerous prospective population studies have shown that physical activity is likely to prevent the development of type 2 diabetes (10). Physical activity increases insulin sensitivity and improves glucose metabolism through mechanisms such as increased muscle mass, insulin-receptor upregulation in muscle, and thereby increased insulin and glucose delivery to muscle (10). Physical activity preserves or elevates muscle mass and physical activity may therefore be of particular importance for those with a small hip circumference and hence a potentially low gluteal muscle mass. Physical activity is therefore likely to modify the inverse association between hip circumference and mortality that has been recently studied (3, 5, 6). Thus, the aim of this study was to estimate the combined effects of hip circumference and leisure-time physical activity on all-cause mortality among adults.
Methods and Procedures
Study population and follow-up
The Copenhagen City Heart Study began in 1976. An age-stratified sample of 19,329 men and women living in the area surrounding the Copenhagen University Hospital were randomly drawn from the Danish Civil Registration System and invited to participate. Individuals were drawn among ∼90,000 eligible inhabitants aged 20 years or older in 1976-1978, and they were re-invited in 1981-1983 and 1991-1994. In 1991-1994 a sample of 3,000 individuals was added. The participants were invited to the Copenhagen University Hospital and were asked to complete a self-administered questionnaire regarding various health-related issues, including questions about leisure-time physical activity. A physical examination was performed where height, weight, hip and waist circumference were objectively measured. Using a tape measure to the nearest 0.5 cm, hip circumference (cm) was measured as the largest circumference over the buttocks and waist circumference (cm) at the level of the umbilicus with the subject standing unclothed and with relaxed breathing. Weight was measured with light clothing and height without shoes, and from these the BMI was calculated as weight (kg) divided by the height (m) squared. During the examination, the staff checked the responses to the questionnaires with regard to potential nonresponse or inconsistent responses. A detailed description of the Copenhagen City Heart Study is published (11). The Danish ethics committee for the city of Copenhagen and Frederiksberg approved the study (no 100.2039/91).
Of the 10,135 persons participating in the Copenhagen City Heart Study in 1991-1994 (response rate of 61%), 1719 with previous diagnosed diabetes, stroke, ischemic heart disease, chronic obstructive pulmonary disease or cancer were excluded. Further, we excluded 1% with the most extreme hip circumference values within each sex; the extreme low hip circumference values excluded were 62.0−82.0 cm for men and 63.0−78.0 cm for women, and the extreme high values were 125.5−171.5 cm for men and 130.5−150.5 cm for women. Finally, 629 participants were excluded due to incomplete information on the exposure variables, leaving 7,708 participants: 3,358 men and 4,350 women (Figure 1). Using the unique person identification number in the Danish Civil Registration System, the participants were followed from date of entry into the study, until date of death, loss to follow-up, emigration, or until May17, 2009, whichever came first (12). These identification numbers encode date of birth, sex, and a record linkage with complete hospital discharge history for each individual. Because of emigration or disappearance, 103 persons (1.3%) were lost to follow-up before May 17, 2009.
Hip circumference was included continuously in the analyses (cm, restricted cubic spline (13)). However, we also constructed three groups from the quartiles of hip circumference at baseline within each sex: (1) narrow hips (first quartile within sex; men ≤95.0 cm; women ≤93.0 cm), (2) medium hips (second and third quartile within sex), (3) wide hips (fourth quartile within sex; men >104.0 cm; women >104.5 cm).
Level of leisure-time physical activity
Leisure-time physical activity was assessed as the mean weekly level during the previous year graded in four levels based on a questionnaire constructed by Saltin and Schnohr (14). The four levels were: (1) being almost entirely inactive or engaging in light physical activity <2 h per week (e.g., reading, watching television, cinemagoer); (2) engaging in light physical activity for 2-4 h per week (e.g. walking, cycling, light gardening, light physical exercise); 3) engaging in light physical activity for more than 4 h per week or more vigorous activity for 2-4 h per week (e.g., brisk walking, fast cycling, heavy gardening, sports that cause perspiration or exhaustion); and (4) engaging in highly vigorous physical activity for more than 4 h per week or regular heavy exercise or competitive sports several times per week. Level 1 was considered physically inactive and levels 2, 3, and 4 combined were considered being physically active. In additional sensitivity analyses, we used three levels of physical activity; level 1 was considered physically inactive, level 2 was considered to have a low level of physical activity and levels 3 and 4 combined were considered being moderate to highly physically active.
Potential confounders were identified to be: age (days, continuous), sex, waist circumference (cm; restricted cubic spline (13)), body mass index (kg m−2; restricted cubic spline (13)), weight change in the 6 months before the examination (kg m−2; restricted cubic spline (13)), alcohol intake (drinks/week; restricted cubic spline (13)), pack-years (calculated as years of smoking multiplied by packs (of 20 cigarettes) currently consumed and for ex-smokers their past average pack consumption; restricted cubic spline (13)), daily smoking habits (never-smoker; ex-smoker; smoker of 1-14 g day−1 and <14 g day−1), education (<8 years, 8-11 years, or at least 12 years), and marital status (married or cohabiting; divorced or widowed; and single).
We calculated hazard ratios (HR) with 95% confidence intervals (CI) of all-cause mortality using Cox's proportional hazards regression model. The age of the participant was used as the underlying time scale in the regression analyses, which implies that the individuals contribute to the estimation of the hazards at the ages they were under observation; thus, they entered the analyses with the age at the time of baseline examination, which means that older individuals at examination will have a delayed entry compared with young individuals. Further, as age has a major impact on anthropometry we performed sensitivity analyses among those below 65 years censoring at age 65 and among those at least 65 years, excluding risk time before age 65.
Sex specific analyses were performed in four models: (1) age-adjusted, (2) multiadjusted (age + potential non-anthropometric confounders), (3) multi-adjusted + general fatness (age + non-anthropometric confounders + BMI), and (4) multi-adjusted + waist circumference (age + non-anthropometric confounders + waist circumference). Finally the same models were performed on the entire dataset including sex as a confounder. The reference group was physically active with a hip circumference of 100 cm. We used a combined reference group to ensure that the HRs for the different sizes of hip circumference combined with physical activity were estimated with reference to a common baseline risk. To examine whether physical activity modified the association between hip circumference and mortality, we performed a Wald test. Further, we also performed the analyses with three levels of physical activity. To address the problem of reverse causality, we adjusted the analyses for weight change in the 6 months before the examination and excluded participants with diagnosed disease as described earlier. Additionally, we performed analyses where cases that occurred during the first 5 years of follow-up were excluded. Finally, we performed analyses where we adjusted for pack-years instead of daily smoking.
Our statistical model was based on introducing hip circumference and potential continuous confounders as restricted cubic spline covariates (13, 15). Using restricted cubic splines facilitated fitting nonlinear effects on the outcome, because the spline set-up allows for changes to the fitted effect-model at certain a priori defined knot-points. The cubic instance indicates that there are cubic functions of the underlying covariate involved in these effect-adjustments, and the restricted prefix reflects that these cubic functions are defined in such a way that the fitted effect below the first knot and above the last knot will be forced to be linear. Compared with the procedure of categorizing continuous variables, this approach will of course fit effect-shapes in a somewhat less relaxed manner, but at the same time still allow for quite flexible modeling based on smoothly varying effects as compared to constant effects within groups. Moreover, the number of needed model-parameters will generally be reduced compared with reasonably narrowly defined groups. We have used four knot-points whose positions are empirically defined according to Table 2.3 (p.23) in Harrell (15). Using four knots lead to construction of three corresponding covariates: the first one corresponds to the standard linear covariate and the remaining ones constitute the nonlinear part. As mentioned earlier we excluded 1% with the most extreme hip circumference values within each sex to avoid potentially heavy influence of atypical individuals on the linear part below the first knot and above the last knot.
For every Cox model, evaluation of the proportional hazard assumption was done by visual inspections of Nelson-Aalen plots and tested using Schoenfeld; separately for each covariate. All analyses were performed using Stata version 11 (StataCorp, 2009).
Those who had medium or wide hips tended to be older and less frequently smokers than subjects with narrow hips (Table 1). In 1991-1994, 363 men (11%) and 476 women (11%) were physically inactive and of these, 81 men (22%) and 108 women (23%) had narrow hips below the 25 percentile. Subjects who were physically active tended to cohabit more, be better educated, smoke less, and have a slightly lower BMI and waist circumference than those who were inactive, regardless of sex. Further, physically active women tended to be younger than inactive women. During 100,156 person years of follow-up, 2,513 participants (33%) died. We observed no violations of the proportional hazards assumption.
|Narrow hipsa||Medium hipsa||Wide hipsa|
|Physical activity||Physical activity||Physical activity|
|Activeb||Inactiveb||Activeb||In activeb||Activeb||In activeb|
|Age (years)||56 (43-67)||53 (36-65)||54 (45-65)||56 (44-67)||55 (44-67)||58 (48-67)||58 (47-67)|
|Hip circumference||100 (95-104)||92 (90-94)||93 (90-94)||100 (98-102)||99 (97-101)||108 (106-111)||109 (106-112)|
|Waist circumference||93 (86-101)||84 (79-89)||86 (82-91)||93 (88-98)||95 (89-100)||104 (99-111)||108 (103-116)|
|Body mass indexc||25.5 (23.3-28.0)||22.7 (21.2-24.2)||22.7 (21.1-24.6)||25.5 (23.9-27.2)||25.6 (23.5-27.4)||29.4 (27.5-31.7)||30.4 (28.0-32.8)|
|Alcohol (drinks per week)||10 (4-20)||10 (4-20)||8 (2-24)||10 (5-20)||10 (3-24)||10 (4-21)||7 (2-20)|
|Low level of education (%)||30||25||40||29||33||36||36|
|Age (years)||60 (47-70)||53 (38-68)||61 (45-72)||60 (48-70)||64 (52-72)||62 (54-71)||65 (55-73)|
|Hip circumference||98 (93-105)||90 (88-92)||90 (88-92)||99 (96-101)||99 (96-102)||110 (107-114)||111 (107-116)|
|Waist circumference||80 (73-89)||71 (67-74)||72 (68-77)||79 (75-86)||83 (77-89)||93 (87-100)||97 (90-105)|
|Body mass indexc||24.3 (21.9-27.5)||20.8 (19.7-22.3)||21.1 (19.8-23.0)||24.2 (22.7-26.1)||24.7 (23.0-26.2)||29.7 (27.6-32.3)||30.6 (28.1-34.6)|
|Alcohol (drinks per week)||4 (0-8)||4 (1-9)||3 (0-8)||4 (0-9)||3 (0-9)||3 (0-7)||1 (0-6)|
|Low level of education (%)||33||24||39||32||47||39||50|
Combined effects of hip circumference and leisure-time physical activity
Figures 2 and 3 show the mortality for combinations of hip circumference and leisure-time physical activity for men and women, respectively. Hip circumference was inversely associated with all-cause mortality before and after adjustment for potential confounders, regardless of physical activity or sex (Figures 2A, 2B, 3A, and 3B). Figures 2C and 3C show the association between hip circumference and mortality independent of general fatness, whereas Figures 2D and 3D show the association between hip circumference and mortality independent of waist circumference.
Physical activity modified the association between hip circumference for a given waist circumference and mortality, as the estimated inverse-like effect was different according to whether individuals reported being physically active or not among men and the same tendency was seen, although nonsignificant, among women (men: P = 0.03; women: P = 0.20; combinedMen&Women: P = 0.02) (Figures 2D and 3D). For both men and women the inverse association levelled off around a hip circumference of 100 cm and hip sizes above this level had almost the same mortality independent of physical activity. The excess mortality at the 25th percentile of hip circumference was 40% (HR = 1.40, CI: 1.14-1.72) and 33% (HR = 1.33, CI: 1.10-1.62) higher among inactive men (95.0 cm) and women (93.0 cm) compared with men and women being physically active, respectively. On the other hand, the mortality at the 75th percentile of hip circumference was 6% (HR = 1.06, CI: 0.85-1.32) and 10% (HR = 1.09, CI: 0.90-1.35) higher among inactive men and women compared with men (104.0 cm) and women (104.5 cm) being physically active, respectively. Additionally, the sensitivity analyses with three levels of physical activity showed that among those with a small hip circumference having a low level of physical activity or being moderate to highly physically active did not essentially change mortality patterns compared to the overall combined measure “being physically active,” regardless of sex (results not shown). However, men being moderate to highly physically active with a large hip circumference seemed to have lower mortality compared to those being physically inactive or engaging in a low level of physical activity (results not shown).
Normal weight participants
Subgroup analyses of 1,493 normal weight men (472 deaths) and 2,483 normal weight women (659 deaths) showed the same tendency as shown in Figures 2 and 3. The excess mortality at the 25th percentile of hip circumference among normal weight participants was 33% (HR = 1.33, CI: 0.93-1.90) and 41% (HR = 1.41, CI: 1.06-1.88) higher in the physically inactive men (92.0 cm) and women (90.0 cm) compared with active men and women, respectively. On the other hand, the mortality at the 75th percentile of hip circumference was 12% (HR = 1.20, CI: 0.79-1.60) and 15% (HR = 1.15, CI: 0.86-1.53) higher among inactive men (98.5 cm) and women (97.5 cm) compared with active men and women, respectively.
Subgroup analyses of 1,838 overweight men (703 deaths) and 1,869 overweight women (668 deaths) showed the same tendency as shown in Figures 2 and 3. The excess mortality at the 25th percentile of hip circumference among overweight participants was 25% (HR = 1.25, CI: 0.97-1.62) and 24% (HR = 1.24, CI: 0.95-1.60) higher in the physically inactive men (99.5 cm) and women (101.0 cm) compared with active men and women, respectively. On the other hand, the mortality at the 75th percentile of hip circumference was not different among inactive men (107.0 cm) and women (111.0 cm) compared with active men and women, respectively (men: HR = 0.99, CI: 0.74-1.32; women: HR = 1.05, CI: 0.80-1.39).
Further the analyses with adjustment for pack-years instead of current daily smoking did not alter the estimates essentially, but the P values for interaction was lower (men: P = 0.02; women: P = 0.06) (Table2). Finally, the sensitivity analyses regarding age showed the same tendencies for those below and above 65 years of age, although the interaction only was significant among those above 65 years (Table 2).
|Percentiles of hip circumference||Pa|
|Adjusted for smoking pack-years instead of daily smoking|
|Men (n = 3,069, deaths = 1126)b|
|Women (n = 4,042, deaths = 1,275)b|
|<65 years (n = 5,046, deaths = 274)d|
|65+ years (n = 5,149, deaths = 2,239)d|
Our study suggests that physical activity modified the association between hip circumference and mortality, in such way that physically inactive men and women with narrow hips had ∼35% higher risk of early mortality than those who were physically active with narrow hips. Several studies have found that a larger hip circumference is related to longevity (3-7). Our results are in accordance with these studies but show, in addition, that physical activity can modify the mortality associated with hip circumference and that being physically active is especially important for those with a small hip circumference. Cross-sectional studies have shown that a small hip circumference seems related to low HDL-cholesterol, high prevalence of diabetes and hypertension (16, 17). Furthermore, exercise training is an effective therapeutic intervention to increase insulin action in skeletal muscle (18). Our analyses showed that physical activity seemed protective for those with narrow hips, potentially depending on a more favorable body composition among the active, including a larger muscle mass in the gluteofemoral region, and on beneficial effects from physical activity on lipoprotein metabolism, inflammatory markers, endothelial function, and insulin resistance (19-21). Previous studies have found that there is probably a critical low fat free mass below which risk of morbidity and early mortality increase exponentially (22, 23). Furthermore, the lower-body muscle mass may be of particular importance for insulin regulation (24).
It is also interesting that physical activity did not seem to add further benefit for those with a hip circumference above 100 cm, which could depend on the larger amounts of lower-body fat or gluteal muscle mass being associated with the greater hip size. This has been found to have favorable effects on cardiovascular risk factors and hence on the incidence of cardiovascular disease and mortality (3-7).
The previously shown inverse association between hip circumference and mortality could also be explained by genetic differences in the skeletal muscle, the ability to store fat in the gluteofemoral region and the size of the pelvic bone (25-27). Any alteration in the skeletal muscle mass is likely to influence the hip circumference (28). Furthermore, the ability to store fat in the gluteofemoral region may be beneficial through a faster clearance of free fatty acids from the blood (29-31). Finally, a small bone structure is seen among persons born with low birth weight and low birth weight has been shown to be associated with high mortality (32). However, our results indicate that these potentially genetic effects on hip size and mortality seem to be attenuated by physical activity.
We did not include all the anthropometric measures at once, as we have to consider the substitution aspect. If the analysis did not allow BMI or waist to differ according to different values of hip circumference, the result would not be interpretable. To be sure that our final model was not an indirect measure of waist circumference, we chose to include waist circumference and height; thus allowing BMI to vary. The importance of adjusting for waist circumference was especially seen among those with a hip circumference above 100 cm as the seemingly elevated mortality was eliminated among women after adjusting for waist circumference and even lowered among men. Subgroup analyses of either normal weight or overweight participants showed essentially similar results. Thus, the modifying effect of physical activity on the association between hip circumference and all-cause mortality seemed independent of both waist circumference and general fatness.
Strengths and limitation
Although the size of the study population, the prospective population-based design, and the follow-up time are strengths of the present study, potential limitations comprise confounding, reverse causality and misclassification. Preclinical or diagnosed disease present at the time of examination may have influenced hip circumference and physical activity, either biologically or as a reason to improve health. This kind of reverse causality could have biased our results. However, the results were essentially similar before and after we excluded all individuals with pre-existing diagnosed diabetes, stroke, ischemic heart disease, or cancer in 1991-1994 or adjusted for weight change in the 6 months before examination. Further, similar results were seen when cases that occurred during the first 5 years of follow-up were excluded.
We dichotomized the level of physical activity to perform the analyses of the combined effects, which resulted in some loss of information. Furthermore, level of physical activity obtained through a questionnaire could be subject to misclassification and is likely to have weakened the associations seen. The measure of leisure-time physical activity used in this study has been shown to be a strong predictor for cardiovascular disease and mortality, and has been validated relating to maximal oxygen uptake, showing a concurrent increase in level of leisure-time activity and in maximal oxygen uptake (33). As studies within this cohort and numerous other studies have shown that the largest risk reduction appears between those being physically inactive to just a low level of physical activity, we dichotomized physical activity into those reporting to be physically inactive and the three other levels were combined into those being physically active (14). As we chose to dichotomize this measure of physical activity, we interpret our new physical activity variable as whether the participants were physically active or not, although it may be argued that this is more a measure of sedentary behavior. However, the participants have answered a question about leisure-time physical activity, not sedentary behavior. We can assume that level 2, 3, and 4 engaged in a greater volume of leisure-time-related physical activity than level 1. As sedentary behavior is a measure of accumulated sedentary time during the day, some of those categorized as physically active in our study could be highly sedentary (34, 35). Therefore, we cannot make any inferences about the impact of sedentary behavior from this study.
Adjustment for daily smoking does not account for duration of smoking; however, adjustment for pack-years did not alter the association essentially. Furthermore, although age has a major impact on anthropometry similar tendencies were seen for those below and above 65 years of age.
In this study, 61% of the invited cohort participated, and a decision not to participate could be associated with being physically inactive and has earlier been shown to be associated with obesity (36). However, obese inactive individuals would probably be categorized in the group with wide hips and such selective nonparticipation would tend to attenuate the association seen, suggesting even larger differences had they participated.
We think our results reflect the importance of muscle mass in the gluteofemoral region for longevity. We found that narrow hip circumference did add to waist circumference and general fatness in prediction of all-cause mortality in both men and women, with approximately 35% higher risk among those being physically inactive. Therefore, information on hip circumference in combination with physical activity status may be used to identify individuals with increased mortality risk. We were able to adjust for differences in waist circumference and general fatness in our analyses, but unfortunately, we were not able to solely adjust for muscle mass or even more explicitly for muscle or fat mass in the gluteofemoral region to investigate this hypothesis further.
In conclusion, a small hip circumference seems associated with early mortality, but the excess mortality was attenuated by being physically active. Consequently, being physically active may be of particular importance for prevention of early mortality among those with small hip circumference.
The work is part of the activities in the Danish Obesity Research Center (DanORC, www.danorc.dk). We thank the participants of the Copenhagen City Heart Study for their continued participation.
- 2Obesity and cardiovascular disease: pathophysiology, evaluation, and effect of weight loss: an update of the 1997 American heart association scientific statement on obesity and heart disease from the obesity committee of the council on nutrition, physical activity, and metabolism. Circulation 2006; 113: 898-918., , , et al.
- 6Hip Hip Hurrah! Hip size inversely related to heart disease and total mortality. Obes Rev 2010;12:478-481., .
- 11The Copenhagen City Heart Study: a book of tables with data from the first examination (1976-78) and a five-year follow-up (1981-83). Scand J Soc Med 1989; 170: 1-160..
- 15Regression modeling strategies. With applications to linear models, logistic regression and survival analysis. New York: Springer-Verlag, 2001..
- 33Habitual physical activity: aerobic power and blood lipids. In: Pernow ESB, ed. Muscle metabolism during exercise. New York: Plenum Press, 1979, pp 469-481., , , et al.