Associations of General and Abdominal Obesity With Multiple Health Outcomes in Older Women:  The Iowa Women's Health Study FREE

EXCESS WEIGHT, often measured as increased body mass index (BMI; calculated as weight in kilograms divided by the square of height in meters) or increased weight for height, has long been considered a risk factor for many chronic diseases. In the past 2 decades, visceral or abdominal obesity, as reflected anthropometrically by an increased waist circumference or waist-hip ratio, has also emerged as an important predictor of risk of obesity-related diseases. However, as discussed in a recent review,¹ it is unclear which measure of abdominal obesity best characterizes risk. Recent guidelines addressing obesity² recommended waist circumference over waist-hip ratio, as waist circumference is simpler to measure and interpret and correlates well with visceral fat measured by computed tomography.¹ Yet, waist circumference also is highly correlated with BMI and thus reflects general, and abdominal, obesity.

Most reports associating abdominal obesity with disease have examined men or younger women, have focused on 1 or 2 diseases at a time, or have failed to consider the contribution of abdominal obesity independent of general obesity. Understanding the roles of different patterns of obesity in multiple diseases is important, given the steady increase in body weight during the past 2 decades in the United States.³

The main aim of the Iowa Women's Health Study was to assess the association of body fat distribution and disease incidence in a large cohort of older women. Previously reported findings^4- 10 from this cohort had limited follow-up. This report extends the follow-up and reports the joint and relative associations of BMI, waist circumference, and waist-hip ratio with multiple disease end points.

In January 1986, a mailed questionnaire was completed by 41,836 (42.7%) of 98,030 randomly selected women, between the ages of 55 and 69 years, who had a valid Iowa driver's license in 1985. Although driver's license information indicated that respondents were 3 months older than nonrespondents, had a lower BMI by 0.4, and were more likely to live in rural counties of Iowa, the associations of body weight with mortality and cancer incidence were similar in respondents and nonrespondents.¹¹ The baseline questionnaire included standard questions on educational level, smoking status and amount, usual alcohol intake during the past year, hormone replacement therapy status, and reproductive history. Participants were asked 3 questions about whether they participated in any leisure time exercise and, if so, the frequency of moderate- and heavy-intensity activities. These latter 2 questions were combined to create a 3-level activity score (low, medium, and high). A food frequency questionnaire was used to assess dietary intake.¹² Prevalent cancers were ascertained by asking the women whether they had ever been told by a physician that they had any form of cancer, excluding skin cancer. Prevalent heart disease was ascertained through self-reports of physician-diagnosed myocardial infarction, angina, or other heart disease. Prevalent hypertension and diabetes were similarly assessed.

Enclosed with the questionnaire were a paper tape measure and written instructions to have a friend measure circumferences of the waist (2.5 cm above the umbilicus) and hips (maximal protrusion). The waist-hip circumference ratio was calculated. Participants reported current height and current weight, from which the BMI was derived. The self-measured or self-reported anthropometric variables obtained by this protocol were valid (intraclass correlation coefficient with measures by trained technician, ≥0.84) and reliable (intraclass correlation of measures at 2 different periods, ≥0.85).¹³

Information on the vital status of the cohort was collected by several methods. Participant identifiers (name, address, social security number, birth date, and maiden name) were linked by computer to death certificates at the State Health Registry of Iowa for 1986 through 1996. To identify deaths outside of Iowa and nonfatal, noncancer end points, mailed follow-ups were undertaken in 1987 (91%), 1989 (90%), 1992 (83%), and 1997 (79%) (response rates are given in parentheses); the vital status of nonrespondents was identified by linkage with the National Death Index. We estimate that 99% of deaths in the cohort have been identified. Underlying causes of deaths were coded according to the International Classification of Diseases, Ninth Revision (ICD-9).¹⁴ For deaths occurring in Iowa, underlying causes were assigned by the Iowa Department of Health; for deaths outside of Iowa, underlying causes were assigned by an experienced nosologist following International Classification of Diseases, Ninth Revision (ICD-9) rules.

Cancer incidence was identified by computer linkage with the State Health Registry of Iowa, a National Cancer Institute–supported Surveillance, Epidemiology, and End Results cancer registry. Other incident disease end points were obtained during follow-up surveys as the first positive self-report of a physician diagnosis.

For this report, we examined mortality from all causes and from 4 mortality subsets: coronary heart disease (CHD), all other cardiovascular disease (CVD), cancer, and all other causes, for which the percentages of deaths were 18%, 13%, 47%, and 22%, respectively. We also examined the incidence of any cancer and common sites in women (breast, lung, colon, uterus, and ovary). Finally, we examined the incidence of self-reported high blood pressure, diabetes, and hip fracture, the latter 2 of which have been shown to be self-reported reasonably accurately in this cohort or elsewhere.^6,15- 16 Although self-reports of myocardial infarction and stroke were collected, we chose to present CHD- and CVD-related mortality instead, as potentially more valid end points.

Of the 41,836 women, we first excluded from consideration women who were premenopausal (n = 569); then, women who reported a baseline history of cancer, heart disease, or diabetes (n = 9426); and then, women who had missing data on BMI and waist-hip ratio (n = 139). This left 31,702 for analysis. For incidence analyses of high blood pressure, prevalent cases at baseline were excluded. For uterine and ovarian cancer incidence, women who had undergone a hysterectomy and a bilateral ovariectomy, respectively, were also excluded. For multivariable analyses, we also excluded women who were deemed to have invalid dietary data (n = 2167), namely, 30 or more items missing on the food frequency questionnaire or implausible energy intake (<2520 or ≥10,584 kJ/d).

Person-years of follow-up for computation of mortality rates were calculated as the time elapsed from completion of the baseline questionnaire to either the end of 1996 or death. For cancer incidence, person-years accumulated from baseline until (1) identified emigration from Iowa (approximately 0.5% annually); (2) identification of a (first) registered cancer diagnosis; (3) death; or (4) December 31, 1996, whichever occurred first. For other, self-reported incident end points, person-years were the sum of the known disease-free period plus half of the period between questionnaires, during which the diagnosis was first made; for women without incident end points, person-time continued until their last completed questionnaire.

Women were categorized according to quintiles (or tertiles when there were few end points) of the baseline anthropometric variables among the total cohort at risk. Age-adjusted relative risks (RRs), their 95% confidence intervals (CIs), and the P value for trend in RRs were calculated by Mantel-Haenszel methods.¹⁷ Multivariately adjusted RRs and their 95% CIs were computed by proportional hazards regression using SAS statistical software (program PHREG; SAS Institute Inc, Cary, NC). Covariates were those variables found to be important risk factors in previous Iowa Women's Health Study analyses.^{6- 7,9,18- 19} We computed multivariately adjusted RRs first for BMI, waist-hip ratio, and waist circumference separately. We then included BMI and waist-hip ratio simultaneously in a model to examine their independence. We did not include BMI and waist circumference in the same model because they were highly correlated. However, we did cross classify BMI and waist circumference according to recent obesity guidelines² and computed age-adjusted odds ratios for pooled end points using unconditional logistic regression. (Person-years were omitted from the tables but are available on request from one of us [A.R.F.].)

The cohort included 31,702 postmenopausal women who at baseline were free of cancer, heart disease, and diabetes. The Pearson product moment correlation of BMI with waist circumference was 0.82; BMI with waist-hip ratio, 0.40; and waist circumference with waist-hip ratio, 0.72. Body mass index and waist-hip ratio were associated with most lifestyle variables examined (Table 1). For example, among nondietary factors, BMI and waist-hip ratio were associated negatively with educational level, estrogen use, physical activity, alcohol intake, and multivitamin use. Body mass index, but not waist-hip ratio, was also associated negatively with ever smoking, whereas waist-hip ratio, but not BMI, was associated positively with age.

A total of 2476 participants died. In the age-adjusted models, BMI and waist circumference showed U-shaped associations with all-cause mortality (Table 2). After adjustment for other risk factors, women with general obesity were not at increased risk of death: the relative risks for the fifth vs first quintiles were 0.91 (95% CI, 0.8-1.0) for BMI and 1.1 (95% CI, 1.0-1.3) for waist circumference. In contrast, there was a graded risk of mortality across quintiles of waist-hip ratio, with an age-adjusted RR of 1.5 (95% CI, 1.4-1.8) and a risk factor–adjusted relative risk of 1.2 (95% CI, 1.1-1.4) for a comparison of extreme quintiles. When BMI and waist-hip ratio were in the multivariate models together, the RRs were 0.79 (95% CI, 0.7-0.9) comparing extreme quintiles of BMI and 1.4 (95% CI, 1.2-1.6) for waist-hip ratio. Figure 1 shows that there was a positive age-adjusted association of mortality with waist-hip ratio for each stratum of BMI. The highest mortality stratum was that with the lowest BMI and highest waist-hip ratio. Elimination of ever smokers from Figure 1 reduced the 3 highest RRs (in the high waist-hip ratio and low BMI quintiles), but they remained elevated (RRs, 1.2-1.7).

Body mass index, waist-hip ratio, and waist circumference were associated positively with CHD-related mortality (438 deaths), with multivariately adjusted RRs for the fifth vs first quintiles of 1.7, 1.9, and 2.2, respectively (Table 2). When BMI and waist-hip ratio were in the same model, the RRs decreased to 1.3 (95% CI, 0.9-1.9) for BMI and to 1.8 (95% CI, 1.2-2.7) for waist-hip ratio but still suggested that general and abdominal obesity contribute to CHD. A stratified analysis also indicated that a higher BMI and waist-hip ratio were associated with greater CHD-related mortality among the total sample (not shown) and among the never smokers (Figure 2).

The waist-hip ratio was associated positively, BMI was associated negatively, and waist circumference was not associated with other CVD-related deaths (326 deaths) (Table 2). The RRs for the highest vs lowest quintiles of BMI and waist-hip ratio were 0.60 (95% CI, 0.4-0.9) and 1.7 (95% CI, 1.1-2.7), after adjustment for covariates and each other.

The number of participants who died of cancer was 1156. Body mass index, waist-hip ratio, and waist circumference were not significantly associated with cancer-related mortality after adjustment for other risk factors (Table 2).

Body mass index was associated negatively, waist-hip ratio was associated positively, and waist circumference was not associated with the residual group of deaths (556 deaths) (data not shown). The RRs for the highest vs the lowest quintiles of BMI and waist-hip ratio were 0.51 (95% CI, 0.4-0.7) and 1.5 (95% CI, 1.1-2.0), respectively, after adjustment for covariates and each other.

Cancer incidence (3738 events) tended to be associated positively with all 3 anthropometric variables, with RRs between 1.1 and 1.4 for the highest vs the lowest quintiles (Table 3).

Body mass index and waist circumference were moderately strongly associated with breast cancer incidence (n = 1299), with risk factor–adjusted RRs for extreme quintiles being 1.6 and 1.7, respectively (Table 3). The RR for waist-hip ratio was smaller (RR, 1.3) and attenuated further when BMI was in the model (RR, 1.1). However, those in the lowest combined quintiles of BMI and waist-hip ratio were at the lowest risk of breast cancer (Figure 3).

The associations of anthropometric variables with colon cancer (n = 462) (Table 3) mirrored those for breast cancer.

Body mass index was associated negatively with lung cancer incidence (n = 386), even after multivariate adjustment (Table 3). In contrast, the waist-hip ratio and waist circumference were not associated independently with lung cancer incidence. Elimination of ever smokers left only 52 persons with incident lung cancers, among whom the age-adjusted RRs across quartiles of BMI were 1.00, 0.58, 0.62, 0.34, and 0.56 (P = .09 for trend). There was no age-adjusted association of waist-hip ratio and waist circumference with lung cancer among never smokers.

Body mass index and waist circumference were strong predictors of uterine cancer incidence (n = 298), with approximately 4-fold higher incidence for the fifth vs the first quintile but little association across the middle quintiles (Table 3). The association of waist-hip ratio with uterine cancer incidence was eliminated with adjustment for, or stratification by, BMI (Table 3 and Figure 4).

Body mass index was not associated, and waist circumference and waist-hip ratio were positively but not monotonically associated, with ovarian cancer incidence (n = 141) (Table 3). The risk factor–adjusted RR comparing extreme quintiles of waist-hip ratio was 2.0 (95% CI, 1.1-3.7), and this value was not changed after adding BMI to the model.

Body mass index, waist-hip ratio, and waist circumference were strong risk factors for incidence of self-reported diabetes (n = 1578) (Table 4). Among the 3 anthropometric variables, waist circumference displayed the greatest RRs. As shown in Figure 5, BMI and waist-hip ratio appeared to contribute independently to diabetes incidence; the age-adjusted RR of diabetes for the highest combined quintiles vs the lowest was exceptionally high, 29 (95% CI, 18-46).

All 3 anthropometric variables were associated positively with incidence of self-reported high blood pressure (n = 4077). Relative risks were approximately 2.0 for highest vs lowest quintiles of each variable (Table 4), suggesting that both general and abdominal obesity are risk factors. The age-adjusted RR for highest vs lowest combined quintiles of BMI and waist-hip ratio (Figure 6) was 2.8 (95% CI, 2.4-3.3).

Body mass index was associated inversely with occurrence of self-reported hip fracture (n = 484), whereas waist-hip ratio was associated slightly positively and waist circumference was not at all associated with hip fracture (Table 4). In the multivariate model including both anthropometric variables, the RRs comparing extreme quintiles were 0.49 (95% CI, 0.4-0.7) for BMI and 1.6 (95% CI, 1.1-2.2) for waist-hip ratio. Figure 7 depicts the contrasting age-adjusted joint associations of BMI and waist-hip ratio with hip fracture.

Recently, an expert panel proposed joint BMI (6 classes from underweight [BMI, <18.5] to obesity class 3 [BMI, ≥40]) and waist circumference (≤88 or >88 cm) cut points to stratify patients for risk of diabetes, hypertension, and CVD.² The left half of Table 5 summarizes the expert consensus panel's categories of risk for the different body size groupings, and the right half shows estimated risk based on Iowa Women's Health Study data. For this analysis, we excluded participants with hypertension at baseline and used 2 pooled end points, namely, (1) occurrence of incident diabetes, incident hypertension, or any CVD-related death (including CHD); or (2) occurrence of incident diabetes, hypertension, hip fracture, or cancer or all-cause death. Because waist circumference and BMI were highly correlated, there were few women in either the high BMI and low waist circumference category or the low BMI and high waist circumference category (Table 5). The age-adjusted odds ratios from our cohort for the composite end point of diabetes, hypertension, or death from CVD seemed to correspond to the consensus risk categories. However, BMI was not predictive of the broader end point (also including cancer incidence, hip fracture incidence, and all deaths) in women with waist circumferences of 88 cm or less. Adjustment of the data for other covariates, including smoking, did not materially alter these odds ratio estimates.

Our main aim was to examine the joint risk and RR of general obesity vs abdominal obesity on multiple health outcomes in a cohort of older women initially free of cancer, heart disease, and diabetes and followed up for 11 to 12 years. Our hypothesis was that the relative contributions of BMI (primarily a marker of general obesity), waist-hip ratio (primarily a marker of abdominal obesity), and waist circumference (a marker of general and abdominal obesity) would vary among diseases. Understanding the various patterns of obesity that lead to different diseases may offer clues to their etiologies and help guide clinical decision making.

General obesity appeared to be a poor predictor of total mortality in these older women, but abdominal obesity, as measured by waist-hip ratio, showed a positive, monotonic, and independent association with total mortality. As reported after 5 years of follow-up,⁹ the women with the highest risk of death in this cohort were those with a low BMI and a high waist-hip ratio (Figure 1). This was also true, although attenuated, among the never smokers. Thus, a high waist-hip ratio appears to be the best anthropometric predictor of death in this cohort. Our results, nevertheless, should not be interpreted to mean that a high BMI is not a potential health hazard, especially when BMI becomes elevated early in life²⁰; a low BMI in some older adults may reflect ill health that we could not control for by excluding prevalent cancer, heart disease, and diabetes.

Abdominal fat measures (waist-hip ratio or waist circumference) appeared to show a monotonic association with CHD-related deaths, but only the highest quintile of BMI carried an increased risk of CHD-related mortality. This is generally consistent with previous prospective findings,^21- 25 suggesting that abdominal obesity, in addition to a high BMI, is an important risk factor for CHD-related mortality or incidence. Abdominal obesity appears to be a particularly good indicator of the adverse metabolic risk factor profile (ie, hyperinsulinemia, dyslipidemia, hypertension, and impaired fibrinolytic capacity),^26- 27 which increases the risk of CHD. In contrast with CHD, other cardiovascular-related deaths were positively associated with waist-hip ratio and negatively associated with BMI, when adjusted for each other. Cancer-related mortality was only weakly associated with measures of obesity.

The finding that "other causes" of death showed a positive association with waist-hip ratio but a negative association with measures of general obesity is difficult to interpret. The component causes of the other deaths were a heterogeneous mix, including injury and poisoning, chronic pulmonary disease, and acute conditions that developed during follow-up. There were too few events to analyze every cause of death separately. However, risk of death from injury or poisoning, deaths that cannot easily be attributed to biological effects of obesity, appears to be elevated, albeit nonsignificantly, among women with greater body size. This may reflect residual effects of lower socioeconomic status on injury- and poisoning-associated mortality, although we did adjust for educational level. Future research should address whether general or abdominal obesity contributes to other preventable diseases or, conversely, whether the underlying diseases alter body composition.

The incidence of self-reported diabetes was strongly associated with all of the measures of obesity. Women in the highest quintiles of BMI and waist-hip ratio had a dramatic RR of 29 compared with women in the lowest combined quintiles. These findings are consistent with those of previous prospective studies,^28- 32 and emphasize the role of general and abdominal obesity in causing insulin resistance^26,33 and, ultimately, diabetes. However, even women in the lowest quintile of BMI had a markedly elevated diabetes risk if they also had a high waist-hip ratio (Figure 5), suggesting the potentially pivotal role of visceral adipose tissue in insulin resistance.²⁶ Weight loss improves insulin sensitivity,³⁴ and efforts are under way to determine whether weight control and other lifestyle modifications can prevent the onset of diabetes.³⁵

The incidence of reported hypertension was slightly more strongly associated with general obesity than abdominal obesity, although those at highest risk (RR, approximately 3.0) of hypertension were in the highest quintiles of BMI and waist-hip ratio. Previous studies^33,36 have also suggested a somewhat stronger association of high blood pressure with generalized obesity than with abdominal obesity, with insulin resistance as the potential mechanism most commonly evoked.³⁷

Women with a higher BMI reported a decreased occurrence of hip fracture than did women with a lower BMI, similar to other prospective studies³⁸ and in line with increased estrogen levels in obese women.^39- 40 In contrast, hip fracture incidence was not related to waist circumference, and hip fracture incidence increased with higher waist-hip ratio adjusted for BMI. Possible explanations are that women with a high waist-hip ratio are more prone to falling, have more osteoporosis, or have less fat padding to protect against hip fracture on falling. Unfortunately, we have no data in this cohort on bone density or rates of falling.

Overall, the incidence of any cancer was positively, albeit weakly, associated with both general and abdominal obesity. Previous prospective studies have reported that general obesity increases the incidence of postmenopausal breast cancer,^41- 42 colon cancer,^43- 44 and uterine cancer,^45- 46 but few have examined abdominal obesity. In our study, general obesity appeared to be more strongly associated with breast cancer, colon cancer, and uterine cancer than did abdominal obesity. Increased adipose conversion of androgens to estrogens⁴⁷ is usually cited as a mechanism for obesity increasing the risk of these estrogen-dependent cancers. Abdominal fat distribution does not appear to further increase total estrogen concentrations in obese postmenopausal women, but increases androgen levels and lowers sex hormone–binding globulin levels.³⁹ This may explain why abdominal obesity is not associated, beyond general obesity, with breast, colon, and uterine cancer.

Others^48- 49 have reported little association of ovarian cancer with general obesity. In contrast, as reported¹⁰ after 7 years of follow-up, ovarian cancer was associated positively with abdominal obesity in this cohort. Because a high waist-hip ratio is associated with hyperinsulinemia,³³ hyperandrogenemia,³⁹ and polycystic ovaries,⁵⁰ 1 or more of these 3 factors may increase ovarian cancer risk.

Associations of anthropometric variables with lung cancer may be confounded by smoking, despite statistical adjustment. Analysis for never smokers suggested that, although not quite statistically significant, women with a low BMI independently could be at increased risk of lung cancer. Reasons for such an association, if real, are unclear.

Cross classifying BMI by waist-hip ratio permitted sufficient numbers of events to differentiate patterns of obesity with disease occurrence. Yet, recent guidelines for the identification, evaluation, and treatment of obesity² incorporate BMI and waist circumference in the assessment, not waist-hip ratio. Because BMI and waist circumference are highly correlated, some of the recommended categories for risk stratification—namely, high BMI and low waist circumference or low BMI and high waist circumference—have too few women to be useful. Nevertheless, our data do corroborate that use of BMI and waist circumference may better differentiate risk of diabetes, hypertension, and CVD than BMI alone. However, BMI was not valuable in predicting risk of the pooled end point, including cancer, hip fracture, and all deaths, when waist circumference was 88 cm or less (Table 5).

The potential drawbacks of this study warrant consideration. The covariates and some nonfatal end points were self-reported, but we chose nonfatal end points that are self-reported fairly accurately.^15- 16,51 We did not have data on blood pressure, lipid levels, or other potentially valuable physiologic measures. Misclassification of these self-reported variables may have led to bias, although it is difficult to imagine that bias would account for different patterns of associations for different anthropometric indexes. Anthropometric indexes were self-measured, but we have documented their validity.¹³ It is possible that relations between body size and disease differ according to the duration of obesity or the age of the study population. Furthermore, even though we excluded women with known heart disease, cancer, and diabetes at baseline, other or subclinical diseases could have caused some of these older women to have lost weight before baseline. However, it was previously shown that excluding women with weight loss since the age of 50 years did not alter mortality results appreciably.⁹ Finally, these results apply to older white women and may not generalize to other ethnic groups, to younger women, or to men.

Clinical guidelines using BMI and waist circumference² indeed identify older women at risk of diabetes, hypertension, and CVD. The high correlation between BMI and waist circumference, however, leads to categories that actually contain few women. Waist-hip circumference further identifies women at increased risk of death and several other important illnesses; it could be considered for additional risk stratification.

Accepted for publication December 16, 1999.

This study was supported by grant R01 CA39742 and training grant T32 CA09607 (Dr Mink) from the National Cancer Institute, Bethesda, Md.

We thank Laura Kemmis for manuscript preparation.

Reprints: Aaron R. Folsom, MD, Division of Epidemiology, School of Public Health, University of Minnesota, 1300 S Second St, Suite 300, Minneapolis, MN 55454-1015 (e-mail: folsom@epi.umn.edu).