OBJECTIVE: To compare levels of physical function, across levels of body mass index (BMI), among middle- to olderaged

Transcription

1 International Journal of Obesity (1998) 22, 958±965 ß 1998 Stockton Press All rights reserved 0307±0565/98 $ Lower levels of physical functioning are associated with higher body weight among middle-aged and older women EH Coakley 1,4, I Kawachi 1,3, JE Manson 1,2,5, FE Speizer 1,6, WC Willet 1,4,5 and GA Colditz 1,5 1 The Channing Laboratory and 2 Division of Preventive Medicine, Department of Medicine, Harvard Medical School and Brigham and Women's Hospital; and the Departments of 3 Health and Social Behavior, 4 Nutrition, 5 Epidemiology and 6 Environmental Health, Harvard School of Public Health, Boston, MA, USA OBJECTIVE: To compare levels of physical function, across levels of body mass index (BMI), among middle- to olderaged women. DESIGN: Cross-sectional study. Physical function, body weight and other covariates were measured in SUBJECTS: women aged 45±71 y, free of cardiovascular disease and cancer, participating in the Nurses' Health Study. MAIN OUTCOME MEASURES: The four physical function scores on the Medical Outcomes Study (MOS) Short Form- 36 (SF36) Health Survey: physical functioning, vitality, bodily pain and role limitations. RESULTS: After adjusting for age, race, smoking status, menopausal status, physical activity and alcohol consumption, there was a signi cant dose-response gradient between increasing levels of BMI in 1992 and reduced function. For example, women with a BMI between 30±34.9 kg=m 2 averaged: 9.0 point lower physical functioning score (95% Con dence interval (CI) 79.5, 78.5), 5.6 point lower vitality score (95% CI: 76.1, 75.1), and 7.0 point lower freedom from pain score (95% CI: 77.6, 76.4). These declines represent an approximate 10% loss of function compared to the reference category of women with BMIs ranging from 22.0±23.9 kg=m 2. For the same BMI comparison, heavier women were at 66% increased risk of limitations in ability to work or perform other roles (RR ˆ 1.66; 95% odds ratio (OR) CI: 1.56, 1.76). These ndings were replicated when the sample was restricted to women who had maintained their BMI over a ten year period. CONCLUSIONS: In addition to increasing risk of chronic health conditions, greater adiposity is associated with lower every day physical functioning, such as climbing stairs or other moderate activities, as well as lower feelings of wellbeing and greater burden of pain. Keywords: BMI; physical function; epidemiology Introduction Excess body weight is associated with increased risks of total mortality, as well as increased incidence of a variety of diseases including diabetes mellitus, hypertension, cardiovascular diseases, osteoarthritis and certain cancers. 1 The relation of adiposity to limitations in daily activities has been examined only to a limited degree. In a follow-up study of 2554 women aged 45±74 y taking part in the National Health and Nutrition Examination Survey (NHANES) I, 2 those in the highest tertile of the body mass index (BMI) (>27 kg=m 2 ) had a twofold increase in the risk of developing mobility disability compared to women in the lowest tertile. In that study, examples of mobility disability included any limitations in carrying out daily activities, such as walking across a room, climbing stairs or transferring from a bed, bath or chair. 2 Although other studies have also reported an Correspondence: Eugenie H. Coakley, Channing Laboratory, 181 Longwood Avenue, Boston, MA 02115, USA. Received 13 October 1997; revised 27 April 1998; accepted 12 May 1998 association between obesity and disability, 3±6 less is known about the impact of various levels of overweight on more comprehensive measures of daily functioning. If overweight is associated with subtle limitations in activities such as, performing chores like carrying groceries or accomplishing work tasks, as well as symptoms such as being tired or in pain, this may not be captured by many disability scales, but could nonetheless impose a considerable burden on the quality of life for individuals. In this crosssectional study of middle-aged women, we therefore examined the association between body weight and four indices of daily functioning, as assessed by the Medical Outcomes Study Short Form-36 (SF-36) Health Survey: 7 physical functioning, role limitations due to physical problems, vitality and bodily pain. Methods The Nurses' Health Study Cohort The Nurses' Health Study cohort was established in 1976, 8 when female registered nurses aged

2 30±55 y completed a mailed questionnaire, requesting information about risk factors for cancer and cardiovascular diseases, including current and past smoking habits, and past personal history of myocardial infarction, angina, cancer, diabetes, hypertension and high serum cholesterol levels. Since then, follow-up questionnaires have been mailed every two years to the entire cohort, to update information on a variety of health risk factors and the occurrence of major illnesses. Further details of the Nurses' Health Study have been described elsewhere. 8 The Medical Outcomes Study Short Form-36 (SF-36) Health Survey On the 1992 questionnaire, we included a slightly modi ed version of the Medical Outcomes Study (MOS) Short Form-36 (SF-36) Health Survey. 8 The SF-36 questionnaire was originally developed for the Rand Corporation's Health Insurance Experiment. 9 It is a self-administered, 36-item questionnaire that measures health-related function in eight domains: physical functioning; role limitations due to physical problems; role limitations due to emotional problems; vitality; bodily pain; social functioning; mental health; and general health perceptions. The instrument has been extensively validated within the Medical Outcomes Study, 10 and in other settings 11, and has high construct validity 11±13 and high internal consistency reliability 10,11,14±16 and test±retest reliability. 17 In the present study, we examined the relationship of BMI to four scales of the SF-36 pertaining to domains of physical function: (a) the physical functioning (PF) scale, as measured by the sum of ten Likert-scaled items, indicating ability to perform a variety of daily activities and tasks that require physical effort, such as climbing stairs and carrying groceries; (b) the role limitations due to physical problems (RP) scale, as measured by the sum of four Likertscaled items, indicating whether one's physical health impaired one's ability to perform work or other activities; (c) the bodily pain (BP) scale, as measured by the sum of two Likert-scaled items, indicating frequency of pain and its interference with daily activities; and (d) the vitality (VT) scale, as measured by the sum of four Likert-scale items, indicating levels of subjective well-being, energy and fatigue. We focused on the four subscales relating to physical, rather than mental, health because obesity has already been linked to major chronic disease in this cohort. 18,19 Our goal was to determine whether obesity could be linked to a broader de nition of physical health. Of these four domains, vitality is a more general measure that is also correlated with psychological well-being. However, because vitality is also strongly correlated with the physical health component of the SF-36, 7 we have included it as an outcome in these analyses. After summing the items, each scale is then standardized so that they each range from 0 (lowest level of functioning) to 100 (highest level). As a general guide, Ware et al 7 have suggested that a ve-point or greater decrement in any of the SF-36 scales, is clinically relevant. Predictor variables BMI (de ned as weight divided by squared height) was assessed from height (in m) as ascertained on the original 1976 questionnaire and body weight (in kg) as reported in the 1992 questionnaire. The validity of self-reported weight in this cohort has been established in a sub-study of 184 participants living in the greater Boston area. 20 Six to twelve months after completing the study questionnaire, participants were weighed in light clothing on a digital bathroom scale. The weights reported in the questionnaire were highly correlated with actual measurements (Spearman r ˆ 0.96), although the self-reported weight averaged 1.5 kg less. 20 In our analyses of BMI and functioning, we controlled for several potential confounding factors, including age, cigarette smoking, minority status, levels of physical activity, alcohol consumption and menopausal status. Using data from the 1990 and 1992 surveys, women were categorized as never, former or current smokers (1±14, 15±24 or 25 cigarettes=d). Women were classi ed as being either of Caucasian or non-caucasian ancestry. Recreational physical activity was determined using a reproducible, validated questionnaire 21 that assesses the frequency of eight common activities in which women engage. The physical activity battery enabled us to calculate an average weekly activity score that accounted for both the type and duration of activities. 22 Recreational physical activity was not included as a covariate in the PF model, since both measure similar domains. Alcohol intake (g per day) was assessed via food frequency questionnaire. In one set of models we also considered as covariates, self-reported diagnoses of diabetes, arthritis (combined rheumatoid arthritis and osteo-arthritis), high blood pressure and hypercholesterolaemia. Hormone replacement therapy was considered a covariate in an analysis limited to post-menopausal women. With the exception of alcohol intake, which was assessed on the 1990 questionnaire, all other predictor variables were assessed in Study population To be included in this study, women had to have completed the SF-36 survey on the 1992 questionnaire, and have information on age, weight and height, smoking and exercise (n ˆ ). This excluded 4726 deaths, women who completed a shorter 1992 survey which did not contain the SF-36 and 1417 who were missing other key data (usually body weight). The shorter survey was sent to women who had not responded to several mailings of the main 1992 questionnaire. The short version only 959

3 960 ascertained disease status and a few key exposures, to minimize respondent burden. We then excluded all women who reported a diagnosis of cancer, angina, stroke or myocardial infarction prior to June 1992 (n ˆ ). Of the remaining women, 191 were excluded for having extreme values for exercise (>30 h=week) or alcohol consumption (>315 g ethanol=week). Thus, women were included in this analysis. Women who completed the shorter version of the 1992 survey, were somewhat older, heavier and more sedentary, on average, than those who completed the longer survey. However, there was substantial overlap (for example, equivalent inter-quartile ranges) in the distributions of these responses. This indicates that estimates based on the subgroup should not be greatly biased by the type of survey returned. Data analysis The SF-36 was considered usable, if respondents answered at least half of the items within each scale. When respondents were missing some (but less than half) of the individual items within a scale, we used the imputation method recommended by Ware et al, 7 to replace the missing items with the respondent's own mean of the other available items for that scale. After replacing the missing items, the scales were computed as previously described. This method is justi ed (that is, does not unduly underestimate variance) when the inter-item correlation is high and when item nonresponse is very low. 23 In the present study, only 7% of respondents have imputed scale scores. The primary analyses were cross-sectional. Multiple linear regression was used to examine the association between BMI and PF, VT and BP. Even though these scales are constrained to range between 0±100, the marginal distributions of these outcomes gave an indication they were approximately normally distributed, with a broad range of values represented. Logistic regression was used to examine the association between BMI and role limitations due to physical problems (RP), because the marginal distribution of this variable was heavily skewed toward the upper end of the scale and only a few values along the continuum of values were represented. The RP scale was thus dichotomized, based on having any role limitations vs none (that is, we modeled the odds of attaining less than a perfect score of 100). For each outcome (PF, VT, BP, RP), two regression models were developed based on different sets of covariates. For the primary analyses, all continuous covariates were categorized a priori into mutually exclusive categories. BMI was categorized using cut points recommended by the World Health Organizations (WHO) 24 < 21.0 kg=m 2, 21±23.9 kg=m 2, 24± 24.9 kg=m 2, 25±27.9 kg=m 2, 28±29.9 kg=m 2, 30± 34.9 kg=m 2, 35±39.9 kg=m 2 and 40.0 kg=m 2. A series of eight indicator variables were then de ned for each category (1 ˆ in category; 0 ˆ not). All but one of the indicator variables were then entered into the regression model. The excluded category was 21± 23.9 kg=m 2 and served as the reference category. The regression coef cients for the other BMI indicator variables could then be interpreted as the difference in outcome for that BMI category, compared to the relatively lean women in the referent group. By modelling the categorical, rather than the continuous, form of BMI, we did not have to assume a linear relationship between BMI and the functional outcomes. A similar type of categorization process was used for age and physical activity; the categories were based on quintiles of the corresponding empirical distributions. Tertiles were used for alcohol consumption, because many of the women in this cohort do not consume alcohol. Smoking status and indicator variables for menopausal status and minority status also were included in these models. A second set of models included the same covariates as in the primary analyses, but also included indicator variables corresponding to a ever having a physician diagnosis of the following medical conditions related to obesity: diabetes mellitus, high blood pressure, hypercholesterolemia and arthritis. Since one of the presumed pathways by which overweight affects physical function is through the development of chronic conditions such as arthritis, diabetes and hypertension, these `mechanistic models' allowed us to test the hypothesis that these conditions lie in the causal pathway between obesity and decrements in function. If this is indeed the case, then the independent effect of BMI on physical function would be substantially attenuated. Because our primary models are cross-sectional, we can only assume BMI predicts function, but cannot rule out that BMI is a consequence of function. To address this limitation, we re-ran our models on the subset of women who maintained their weight for ten years (1982±1992). Weight maintenance was de ned as maintaining a BMI within 1.5 BMI units (about 5 kg) over a period of ten years and maintaining twoyear weight changes within 1.0 BMI units (about 3 kg). Thus, in this subanalysis, weight cyclers and incremental weight gainers or losers were excluded. Finally, to address the potential effects of hormone replacement therapy (HRT) on functioning, we conducted a separate analysis on post-menopausal women, adding a variable indicating ever-use of HRT to the regression models. All analyses were performed using version 6.09 of SAS. 25 Results The mean age of the study population was 58.7 y (range 46±72 y). Comparison of the prevalence of chronic conditions across categories of BMI (Table 1) con rmed that overweight women reported higher

4 prevalence of hypertension, arthritis and diabetes. compared to leaner women, overweight women exercised less, but also smoked less and drank less alcohol. After adjusting for age, physical activity (not in PF model), alcohol consumption, minority status and menopausal status, BMI had a strong inverse relationship with all four measures of functioning (Table 2). In Table 2, the rst row for the PF, BP and VT scales contains the regression intercept, which can be interpreted as the average predicted scale score for the referent group of lean women (BMI ranging from 21.0±23.9 kg=m 2 ), adjusted for the other covariates. The subsequent rows contain the regression coef cients for other categories of BMI and can be interpreted as the relative increase or decrease in scale score for a given level of BMI, compared to the referent group. For example, the average PF for the referent roup was 95.6 and women with a BMI between 28.0±29.9 had an average 5.3 lower PF score, or approximately 5.5% lower functioning. The role function (RP) column contains the odds ratio (OR) of having a limitation for different levels of BMI, compared to the referent group, controlling for the other covariates. Women who were very lean (BMI < 21.0 kg=m 2 ), experienced equivalent physical functioning as the referent group. However, statistically signi cant lower functioning became apparent at higher levels of BMI, with statistically signi cant loss of function, even at BMI levels ranging from 25±27.9 kg=m 2. Women with BMIs ranging from 25±27.9 kg=m 2. Women with BMIs ranging from 28±29.9 kg=m 2 had an average ve point lower level of PF, three point lower VT and a ve point lower BP score. Declines worsened for higher BMI levels. For example, women with BMIs ranging from 35±39.9 kg=m 2 reported a 16-point lower 961 Table 1 Prevalence of health behaviors and diagnosed chronic conditions, according to level of body mass index (BMI) Overall (n ˆ ) 21.9 (n ˆ11331) 22^23.9 (n ˆ10 988) BMI category (kg=m 2 ) 24^24.9 (n ˆ 5553) 25^27.9 (n ˆ12749) 28^29.9 (n ˆ 5785) 30^34.9 (n ˆ 6900) 35^39.9 (n ˆ 2220) (n ˆ 984) Age (ave y) Physical activity (ave met-h=wk) Ethanol (ave g=wk) Current smoker (%) Post-menopause (%) Minority (%) Hypertension (%) High cholesterol (%) Diabetes (%) Arthritis (%) Table 2 Relative impact of body mass index (BMI) on physical functioning (PF), vitality (VT), Bodily Pain (BP) and Role Functioning (RP) Average (95% CI) loss of function for: OR (95% CI) of: Intercept (the average score for referent category) n PF 95.6 (95.1, 96.0) VT 57.0 (56.1, 57.3) BP 76.1 (75.4, 76.8) RP n limited ˆ n not limited ˆ BMI (kg=m 2 ) (0.0, 0.8) (0.3, 1.2) (0.6, 1.7) (0.89, 1.0) 22 ± hreferenti hreferenti hreferenti hreferenti 24 ± (71.4, 70.3) (70.9, 0.2) (71.6, 70.3) (0.98, 1.13) 25 ± (72.7, 71.8) (72.4, 71.5) (72.0, 71.9) (1.17, 1.30) 28 ± (75.8, 74.8) 73.9, 72.8) (75.8, 74.5) (1.40, 1.59) 30 ± (79.5, 78.5) (76.1, 75.1) (77.6, 76.4) (1.56, 1.76) 35 ± , , 77.5) (711.4, 79.6) (1.89, 2.28) > (727.5, 725.3) (713.5, 711.2) (717.0, 714.4) (2.69, 3.54) R 2 14% 10% 5% ± Results for PF, VT and BP are based on multiple linear regression. The intercept values represent the average Short Form-36 (SF-36) Health survey 7 scores for the referent group. For each BMI category, the regression coef cients represent the average change in SF-36 score relative to the referent group. Results (odds ratio, OR) for role limitations due to RP are based on multiple logistic regression. All models control for age, race, smoking, alcohol and menopause status. VT, BP and role models, also control for physical activity. 95% CI ˆ 95% con dence intervals.

5 962 physical functioning (PF), an eight-point lower VT score and an 11-point lower BP score (Table 2). The BP scale is scored in such a way that a high score indicates no pain, while lower scores indicate more severe and limiting pain. For these more extremely overweight women, these decrements of the three scales represent an average 14±16% lower average function, compared to the reference group. The magnitude of these differences in function are both statistically and clinically signi cant. 6 When we examined the role functioning=physical (RP) scale with logistic regression, women with BMIs ranging from 28±29.9 kg=m 2 and 1.49 times the odds of reporting a work limitation due to physical health, compared to the leaner women (95% con dence interval, CI: 1.40±1.59). The odds of a role limitation was 2.1 times higher among those with a BMI of 35±39.9 kg=m 2 (95% CI: 1.89±2.28). Stepwise regression was used to determine the relative importance of BMI in explaining the variation in each health outcome. BMI, exercise, age, alcohol consumption, smoking status, menopause status, minority status and the eight chronic illness variables, were allowed to compete for entry into the models. BMI turned out to be the single most important predictor of physical functioning, bodily pain and role functioning. BMI was the second most important predictor of vitality after physical activity. The other key variables in the stepwise regressions were age for the PF and VT models, and arthritis for RP and BP models. When we re-analysed the models including a set of indicator variables for diagnoses of hypertension, hypercholesterolemia, diabetes and arthritis, inclusion of these variables only somewhat attenuated the impact of BMI on the PF and BP scales (data not shown). No attenuation was found for the VT and RP scales. The attenuation was most noticeable for the higher BMI categories and reduced the effect of BMI by an average of 2±4 points (for PF and BP scales). For example, without the inclusion of chronic conditions as covariates, the average decrement in physical function (PF) for women with BMI between 30± 34.9 kg=m 2 was 9.0 points; with the inclusion of chronic conditions, the average decrement was still a signi cant 7.2 points. All four comorbid conditions had statistically signi cant impact on the four dimensions of physical health, independent of BMI (data not shown). Among the four conditions, diagnosis of arthritis had the single biggest impact on function, resulting in average decrements of eight points for PF, ve points for VT and 13 points for BP. Other covariates were also signi cant predictors of function. As noted above, higher physical activity was a predictor of better function. Cigarette smoking had a statistically signi cant negative effect on PF, VT, BP and RF. Age had a negative effect on physical function, pain and role function, but a positive effect on vitality. Table 3 presents the additive effects of these exposures combined with BMI. We used our main regression models to contrast lean women (BMI 21.0± 23.9 kg=m 2 ), who never smoked, and moderately active (exercised 25±29 MET-hours per week), to moderately overweight women (BMI 28± 29.9 kg=m 2 ), who were regular smokers and sedentary (Table 3). We developed scenarios for two agegroups, 45±51 y and 62±66 y (the rst and fourth age quintiles), and adjusted for menopausal status, minority status and alcohol intake. The format for Table 3 is the same as for Table 2, except that the second and subsequent rows represent the change in scale score for a combination of the listed covariates, predicted from the regression models. Moderately overweight smokers experienced an 8.7 point lower PF. When these women were also sedentary, the reduction in VT was 10.7 points and 7.3 Table 3 The association between combinations of health habits and physical function (PF), vitality (VT), bodily pain (BP) and role functioning=physical (RP) Average (95% CI) loss of function for: PF VT BP OR (95% CI) of: Regression intercept (the average score for the referent category) RP Age 45 ±51 y: BMI 28±29.9 kg=m (75.8, 74.8) (73.9, 72.8) (75.8, 74.5) (1.40, 1.59) BMI 28±29.9 kg=m 2 and rarely exercises n=a BMI 28±29.9 kg=m 2, rarely exercises and smokes 78.7 a (79.5, 77.98) Age 62 ±66 y: BMI 28±29.9 kg=m (714.4, 713.0) (78.8, 77.4) (711.7, 79.8) 1.6 (2.3, 0.9) BMI 28±29.9 kg=m 2 and rarely exercises n=a 73.2 (74.0, 72.3) BMI 28±29.9 kg=m 2, rarely exercises and smokes a ( 7 5.8) (718.0, 716.0) (76.9, 74.8) (77.6, 76.0) 77.3 (78.4, 76.3) 77.8 (78.7, 77.0) 79.5 (710.5, 78.5) (711.3, 78.9) Based on multiple linear (PF, VT, BP) and logistic (RP) regression models presented in Table 2. n=a ˆ exercise not in PF model. a Effect of BMI and smoking only. 95% CI ˆ 95% con dence interval; OR ˆ odds ratio; BMI ˆ body mass index. (1.71, 2.03) 2.05 (1.83, 2.30) 2.11 (1.93, 2.32) 2.64 (2.39, 2.94) 2.92 (2.56, 3.32)

6 points for BP. The OR of having any physical role limitation was 2.05 (95% CI: 1.83±2.30). Among older women (62±66 y), the decrements for these scenarios were similar (Table 3). There were women in the study group (38%), who maintained their weight over the preceding ten years (1982±1992), regardless of the level of their weight in For these women, their 1992 BMI is re ective of their BMI for the past decade. Among these women, the association between level of BMI and current functioning, show the same pattern as for the cohort as a whole (Figure 1). Statistically signi cant declines began at BMI levels of 25± 27.9 kg=m 2. Among those whose long-term BMI was between 30±34.9 kg=m 2, an average 10.9 point reduction in physical functioning was observed, as well as a 6.1 point reduction in vitality, and an OR of 1.82 (95% CI: 1.61±2.10) for having a role limitation due to physical problems. The burden of pain was much greater for long-term obese women, an 8.9 point negative effect. In the separate analysis of post-menopausal women, hormone replacement therapy did not confound the relationship between BMI and the four domains of functioning. Finally, to describe the clinical signi cance of the observed reductions in physical functioning, we developed a content-based interpretation of the PF Figure 1 Difference in function levels by body mass index (BMI) category of weight maintainers. PF ˆ physical function, VT ˆ vitality, BP ˆ bodily pain. j ˆ PF scale; j ˆ VT scale; j ˆ BP scale. scale following the methods described by Ware et al. 8 This was done by cross-tabulating the predicted PF score from the regression (Table 2) with the actual responses to individual items that make up the scale. The PF scale is made up of ten items eliciting the ability to perform the following: vigourous activity, moderate activity, lifting=carrying groceries, climbing several ights of stairs, bending=kneeling=stooping, walking more than one mile, walking several blocks, walking one block and bathing=dressing oneself. Table 4 presents the percentage of respondents who were limited in four of these activities at various levels of predicted PF scores. (PF was chosen because BMI was most highly associated with this scale. The values in Table 4 were computed by (1) assigning a predicted physical functioning score to each woman based on the primary regression model; (2) these predicted values were categorized, and (3) the percentage of women in these categories who actually reported having a limitation for each of the four items was calculated.) For example, 8% of women with a predicted PF score of 95 actually reported being limited in their ability to engage in moderate activity (such as vacuuming or playing golf), whereas 18% of women who scored 85±89 reported this limitation. The information presented in Table 4 can now be used to interpret the results in Table 2 and Table 3. For example in Table 2, women with a BMI between 28±29.9 kg=m 2 had an average predicted PF score of 90.3 (the sum of the intercept and the coef cient for BMI in that category). From Table 4, we can determine that 12% of the women with PF scores in this range were limited in moderate activities; nearly 20% were limited in their ability to climb a few ights of stairs, or to bend, stoop or kneel. To take a further example, women aged 45±51 y with BMI in the range 28±29.9 kg=m 2 and who smoked, had an average PF score of 86.9 (the sum of the intercept and regression coef cient for BMI 25±29 kg=m 2 ; Table 3). According to Table 4, >10% of such women reported limitations carrying groceries and nearly 30% reported limitations climbing several ights of stairs or bending=kneeling=stooping. 963 Table 4 Content-based interpretation of changes in physical functioning (PF) A PF Score of: Predicted PF Score n Percent of respondents who reported a limitation for: Moderate activity Lift=carry groceries Climb several flights of stairs Bend, kneel, stoop 95 ± % 6% 10% 12% 90 ± % 9% 17% 18% 85 ± % 13% 28% 29% 80 ± % 16% 38% 40% 75 ± % 21% 53% 55% 70 ± % 27% 67% 67% 65 ± % 30% 74% 72% 60 ± % 40% 86% 84% % 47% 87% 84% Moderate activity, lift=carry groceries, climbing stairs, bending=stooping=kneeling are four of the ten items that comprise the PF scale of the Short Form-36 (SF-36) Health Survey. 8 Predicted PF scores are from the multivariate linear regression with PF as the outcome (Table 2); see also footnote.

7 964 Discussion The overall ndings of this study, consistent with previous reports, indicate that being overweight is associated with functional limitations. Moreover, the impact of BMI on physical functioning appears to be independent of the diagnosis of obesity-related chronic diseases such as hypertension, diabetes mellitus and arthritis, and is present even at moderate overweight. The magnitude of functional limitations among overweight individuals is striking, especially when combined with other risk behaviors such as cigarette smoking or being sedentary. For example, the nine-point decrement in physical functioning among moderately overweight (BMI in the range 28.0±29.9 kg=m 2 ) and cigarette-smoking, middleaged women, indicates that a substantial proportion of them are limited in performing even minor daily activities such as bending, kneeling or stooping. In her study, Stewart et al 10 found that a nine-point decrement in the physical functioning (PF) scale was comparable to having arthritis or back problems. Stewart et al 10 also found that a 7±10 point decrement in the pain scale (BP) was comparable to having angina or back problems. Thus, for middle- to olderaged women, the functional limitations and pain associated with having a BMI>29.0 kg=m 2 were comparable to those experienced due to arthritis. The effects are additive; on average, someone with both a high BMI and arthritis would experience a double reduction on these scales. The major limitation of this study is its crosssectional design. The direction of causality is not always clear cut. For example, poor physical functioning may have preceded the development of obesity. It is possible that poor health resulted in sedentary behavior, which subsequently led to weight gain. However, we found similar or even somewhat greater, limitations among women who had been consistently obese ten years prior to the measurement of current function (although we did not have data on prior function). The selected nature of the Nurses' Health Study cohort, raises the issue of the generalizability of our ndings. We compared the responses of (Nurses' Health Study participants on the four scales (PF, VT, BP, RP) to a nationally representative sample of women included in the National Opinion Research Center's (NORC) General Social Survey (GSS; unpublished data±personal communication with Dr Benjamin Amick). The mean vitality (VT) score, adjusted for age, were quite similar in the two samples (62.8 in the Nurses' Health Study, compared to 63.9 in the GSS). The mean BP scores were also similar (76.5 in the Nurses' Health Study, compared to 74.7 in the GSS). Women in the Nurses' Health Study reported somewhat higher PF scores (89.4) compared to women in the GSS (85.3), which may re ect a degree of occupational selection. On the other hand, the Nurses' Health Study members reported lower average role functioning=physical (RP) scores (82.0) compared to women in the GSS (87.1), which may re ect the relatively demanding nature of the nursing occupation, such that a given degree of physical impairment results in a greater impact on role performance. Conclusion Although more de nitive conclusions must await the availability of longitudinal data, the ndings of the present study suggest that being overweight in middle life may be associated with limitations in common, day-to-day activities, and perceived levels of energy and fatigue (vitality). This nding is particularly relevant because 50% of this sample had a BMI 25 kg=m 2, and nearly 30% had a BMI 28.0 kg=m 2. These gures are comparable to national estimates for women. 26 If higher body weights are associated with these more immediate and tangible health effects, these ndings may be useful in promoting the 1996 USDA Dietary Guidelines, 27 as an incentive to maintain or improve weight levels. Acknowledgements This study was funded in part by NIH grants CA40356, DK46798, and AG12806, and by the Boston Obesity=Nutrition Center, Epidemiology Core (DK46200). The authors acknowledge the helpful comments of Dr Meir Stampfer on an earlier draft of this manuscript. References 1 Pi-Sunyer FX. Medical hazards of obesity. Ann Intern Med 1993; 119: 655 ± Launer LJH, Harris T, Rumpel C, Madans J. Body mass index, weight change, and risk of mobility disability in middle-aged and older women. JAMA 1994; 271: 1093± Verbrugge LM, Gates DM, Ike RW. Risk factors for disability among US adults with arthritis. J Clin Epidemiol 1991; 44: 167 ± Pinsky JL, Branch LG, Jette AM, et al. Framingham Disability Study: relationship of disability to cardiovascular risk factors among persons free of diagnosed cardiovascular disease. Am J Epidemiol 1985; 122: 644 ± Rissanen A, Heliovaara M, Knekt P, Reunanen A, Aromaa A, Maatela J. Risk of disability and mortality due to overweight in a Finnish population. BMJ 1990; 301: 835± Harris T, Kovar MG, Suzman R, Kleinman JC, Feldman JJ. Longitudinal study of physical ability in the oldest-old. Am J Public Health 1989; 79: 698 ± Ware JE Jr, Snow KK, Kosinski M, Gandek B. SF-36 Health Survey. Manual and Interpretation Guide. The Health Institute, New England Medical Center: Boston, Colditz GA. The Nurses' Health Study: ndings during 10 years of follow-up of a cohort of US women. Curr Probl Obstet Gynecol Fertil 1990; 13: 129± Brook RH, Ware JE, Davies-Avery A, et al. Conceptualization and Measurement of Health for Adults in the Health Insurance Study. Vol. III. Overview. Rand Corp: Santa Monica, 1979.

8 10 Stewart AL, Green eld S, Hays RD, Wells K, Rogers WH, Berry SD, McGlynn EA, Ware JE Jr. Functional status and well-being of patients with chronic conditions. JAMA 1989; 262: 907 ± Brazier JE, Harper R, Jones NM, O'Cathain A, Thomas KJ, Usherwood T, Westlake L. Validating the SF-36 Health Survey Questionnaire: new outcome measure for primary care. BMJ 1992; 305: 160± Kravitz RL, Green eld S, Rogers W, Manning WG Jr, Zubkoff M, Nelson EC, Tarlov AR, Ware JE. Differences in the mix of patients among medical specialities and systems of care. JAMA 1992; 267: 1617± Anderson J St C, Sullivan F, Underwood TP. The Medical Outcomes Study Instrument (MOSI) Ðuse of a new health status measure in Britain. Family Practice 1990; 7: 205± McHorney CA, Ware JE, Lu RJF, Sherbourne CD. The MOS 36-Item Short-Form Health Survey (SF-36): III. Tests of data quality, scaling assumptions, and reliability across diverse patient groups. Med Care 1994; 32(1): 40± McHorney CA, Ware JE, Rogers W, Raczek AE, Lu JF. The validity and relative precision of MOS Short- and Long-Form Health Status Scales and Dartmouth Coop Charts. Med Care 1992; 30: 253± Ware JE, Sherbourne CD. The MOS 36-Item Short Form Health Survey (SF-36). I. Conceptual framework and item selection. Med Care 1992; 30: 473 ± Kurtin PS, Davies AR, Meyer KB, DeGiacomo JM, Kantz ME. Patient-based health status measures in outpatient-based health status measures in outpatient dialysis and early experiences developing an outcomes assessment Program. Med Care 1992; 30 (Suppl): MS136± MS Colditz GA, Willett WC, Rotnitsky A, Manson JE. Weight gain as a risk factor for clinical diabetes mellitus in women. Ann Intern Med 1995; 122: 481± Willett WC, Manson JE, Stampfer JM, Colditz GA, Rosner B, Speizer FE, Hennekens CH. Weight, weight change, and coronary heart disease in women. JAMA 1995; 273: 461± Rimm EB, Stampfer MJ, Colditz GA, Chute CG, Litin LB, Willett WC. Validity of self-reported waist and hip circumferences in men and women. Epidemiology 1990; 1: 466 ± Wolf AM, Hunter DJ, Colditz GA, Manson JE, Stampfer MJ, Corsano KA, Rosner B, Kriska A, Willett WC. Reproducibility and validity of a self-administered physical activity questionnaire. Int J Epidemiol 1994; 23: 991± Ainsworth BE, Haskell WL, Leon AS, Jacobs Dr Jr, Montoye HJ, Sallis JF, Paffenbarger RS Jr. Compendium of physical activities: classi cation of energy costs of human physical activity. Med Sci Sports Exerc 1993; 25: 71± Little RDA, Rubin DB. Statistical Analysis with Missing Data. John Wiley and Sons: New York, World Health Organization. Physical Status: The Use and Interpretation of Anthropometry. World Health Organization: Genera, SAS Institute, Inc. SAS User's Guide: Statistics (6th edn). SAS Institute, Inc.: Cary, Kuszmarski RJ, Flegal KM, Campbell SM, Johnson CL. Increasing prevalence of overweight among US adults. JAMA 1994; 272: 205 ± US Department of Agriculture, US Department of Health and Human Services. Nutrition and Your Health: Dietary Guidelines for Americans. US Government Printing Of ce: Washington, DC,