Author + information
- Received November 16, 2007
- Revision received February 19, 2008
- Accepted March 4, 2008
- Published online August 19, 2008.
- Rebecca P. Gelber, MD, DrPH⁎,⁎⁎,
- J. Michael Gaziano, MD, MPH⁎,†,¶,
- E. John Orav, PhD‡∥,
- JoAnn E. Manson, MD, DrPH†,§,
- Julie E. Buring, ScD⁎,†,§,# and
- Tobias Kurth, MD, ScD⁎,†,§,⁎ ()
- ↵⁎Reprint requests and correspondence:
Dr. Tobias Kurth, Brigham and Women's Hospital, Division of Aging, 1620 Tremont Street, Boston, Massachusetts 02120.
Objectives This study examined associations between anthropometric measures (body mass index, waist circumference, waist-to-hip ratio, waist-to-height ratio [WHtR]) and risk of incident cardiovascular disease (CVD) (including nonfatal myocardial infarction, nonfatal ischemic stroke, and cardiovascular death).
Background Controversy exists regarding the optimal approach to measure adiposity, and the utility of body mass index has been questioned.
Methods Participants included 16,332 men in the Physicians' Health Study (mean age 61 years in 1991) and 32,700 women in the Women's Health Study (mean age 61 years in 1999). We used Cox proportional hazards models to determine relative risks and 95% confidence intervals (CIs) for developing CVD according to self-reported anthropometric indexes.
Results A total of 1,505 CVD cases occurred in men and 414 occurred in women (median follow-up 14.2 and 5.5 years, respectively). Although WHtR demonstrated statistically the strongest associations with CVD and best model fit, CVD risk increased linearly and significantly with higher levels of all indexes. Adjusting for confounders, the relative risk for CVD was 0.58 (95% CI: 0.32 to 1.05) for men with the lowest WHtR (<0.45) and 2.36 (95% CI: 1.61 to 3.47) for the highest WHtR (≥0.69; vs. WHtR 0.49 to <0.53). Among women, the relative risk was 0.65 (95% CI: 0.33 to 1.31) for those with the lowest WHtR (<0.42) and 2.33 (95% CI: 1.66 to 3.28) for the highest WHtR (≥0.68; vs. WHtR 0.47 to <0.52).
Conclusions The WHtR demonstrated statistically the best model fit and strongest associations with CVD. However, compared with body mass index, differences in cardiovascular risk assessment using other indexes were small and likely not clinically consequential. Our findings emphasize that higher levels of adiposity, however measured, confer increased risk of CVD.
We face an epidemic of overweight and obesity that affects more than a billion adults worldwide (1). While multiple health organizations recommend using body mass index (BMI) for the identification of overweight and obese individuals (2,3), uncertainty exists regarding the optimal approach to measure adiposity, and the utility of BMI has been questioned (4,5). Using BMI as the current standard measure of adiposity can result in misclassification of risk among certain populations and may not adequately describe adiposity in relation to cardiovascular risk.
Anthropometric indexes of central adiposity and body composition, including the waist circumference (WC), waist-to-hip ratio (WHR), and waist-to-height ratio (WHtR), may more accurately identify groups at risk for the adverse health consequences of excess weight (2,6–8). However, the relative utility of various anthropometric measures in assessing cardiovascular risk remains unclear (9,10). Furthermore, although some studies have explored the associations between anthropometric indexes and cardiovascular risk beyond BMI, analytic approaches and results have been inconsistent and most studies have not directly compared multiple indexes (10).
We therefore examined the associations between various anthropometric parameters of adiposity and the risk of incident cardiovascular disease (CVD) in prospective cohorts of more than 49,000 men and women.
We used data from 2 prospective cohorts of health professionals in the U.S. to examine the associations between various anthropometric indexes and cardiovascular risk. Baseline and follow-up information was self-reported and collected through mailed questionnaires every 6 months for the first year and annually thereafter. Details of the study methods and results have been described elsewhere (11–15).
The PHS (Physicians' Health Study)
Study subjects included participants in the PHS, a completed randomized trial of aspirin and beta carotene in the primary prevention of CVD and cancer (11,12). This trial included 22,071 apparently healthy male physicians, ages 40 to 84 years in 1982, without a history of CVD or other major illnesses.
Analyses were limited to men who remained in the cohort at 9 years (n = 20,889), the time at which waist and hip circumferences were requested. We excluded men who did not return the 9-year questionnaire (n = 183); those missing information on waist (n = 2,097), hip (n = 2,135), or BMI (n = 623); those with WC or hip circumferences <20 or >70 in (n = 28); and men with a history of CVD, coronary revascularization procedures, or angina prior to the 9-year questionnaire (n = 1,780). There were 16,332 men left for our analyses.
The WHS (Women's Health Study)
Study subjects also included participants in the WHS, a completed randomized trial of aspirin and vitamin E in the primary prevention of CVD and cancer (13–15). This trial included 39,876 female health professionals, age ≥45 years in 1993 without a history of CVD or other major illnesses.
Analyses were limited to women who remained in the cohort at 6 years (n = 39,135), the time at which waist and hip circumferences were requested. We excluded women who did not return the 6-year questionnaire (n = 2,070); those missing information on waist (n = 3,099), hip (n = 3,103), or BMI (n = 758); those with WC <20 in (n = 4); and women with a history of CVD, coronary revascularization procedures, or angina prior to the 6-year questionnaire (n = 492). There were 32,700 women left for our analyses.
We calculated BMI from self-reported weight (kg) divided by the square of the height (m). To evaluate BMI over the range of “normal” and “overweight,” we categorized BMI as <20.0, 20.0 to <22.5, 22.5 to <25.0, 25.0 to <27.5, 27.5 to <30.0, 30.0 to <35.0, and ≥35.0 kg/m2. We also assessed BMI as a continuous variable (per standard deviation unit). We used baseline height and weight reported with the 9- (PHS) or 6-year (WHS) questionnaire to calculate BMI.
Participants were asked to report circumferences using a paper tape measure supplied with the questionnaire. Instructions requested that participants: 1) measure their WC at the level of the umbilicus; 2) measure their hip circumference as the largest circumference between the umbilicus and the thigh; and 3) record the measurements to the nearest one-quarter inch. The WHR was calculated by dividing the WC by the hip circumference, and WHtR was calculated by dividing the WC by the baseline height. We evaluated WC, WHR, and WHtR in 7 categories defined by the percentile distributions of participants in the corresponding 7 BMI categories and as a continuous variable, per standard deviation unit. Self-reports of anthropometric measures have been validated in other health professional cohorts (16).
From the PHS, covariates assessed at 9 years included age (5-year categories), physical activity (rarely/never, 1 to 2, 3 to 4, 5 to 7 days/week), and history of cancer, diabetes, elevated cholesterol (≥240 mg/dl or history of cholesterol-lowering medication use), or hypertension (self-reported systolic blood pressure [BP] ≥140 mm Hg, diastolic BP ≥90 mm Hg, or antihypertensive medication use). Other covariates were obtained at the most recent prior questionnaire: smoking at 5 years (never, past, current), alcohol consumption at 7 years (rarely/never, 1 to 3 drinks/month, 1 to 6 drinks/week, ≥1 drink/day), and, from baseline, parental history of myocardial infarction before the age of 60 years.
From the WHS, covariates at 6 years included age (5-year categories) and history of cancer, history of diabetes, elevated cholesterol (≥240 mg/dl, cholesterol-lowering medication use, or elevated cholesterol diagnosed by a clinician), or hypertension (self-reported systolic BP ≥140 mm Hg, diastolic BP ≥90 mm Hg, hypertension diagnosed by a clinician, or history of antihypertensive treatment). We measured smoking (never, past, current) at baseline, updating the variable through year 6. Other covariates were obtained at the most recent prior questionnaire: post-menopausal hormone use (never, past, current) at 5 years, alcohol consumption at 4 years (rare/never, 1 to 3 drinks/month, 1 to 6 drinks/week, ≥1 drink/day), and, from baseline, physical activity (≤1, 2 to 3, ≥4 times/week), parental history of myocardial infarction before the age of 60 years, highest level of education (less than a bachelor's degree, bachelor's degree, master's/doctoral degree), and race (white, black, other). Dietary information was obtained at baseline from a 161-item standardized food frequency questionnaire (17), and nutrient intake was adjusted for total energy intake using the residual method (18). Dietary variables included cereal fiber, folate, glycemic load, trans fat, polyunsaturated-to-saturated fat ratio, and omega-3 fatty acids (all in quintiles) (19).
We defined incident major CVD as first nonfatal myocardial infarction, nonfatal ischemic stroke, or fatal CVD (defined as fatal myocardial infarction, fatal ischemic stroke, sudden death, or any deaths related to ischemic heart disease; International Classification of Diseases-Ninth Revision [ICD-9] codes 410 to 414, 430 to 438, and 798).
Confirmation of all end points required review of available medical records by an end points committee of physicians who used standardized criteria. Myocardial infarctions were confirmed by elevated plasma levels of cardiac enzymes or diagnostic electrocardiograms. Fatal myocardial infarction was confirmed based on autopsy reports, symptoms, circumstances of death, and history of coronary heart disease. Nonfatal stroke was defined as a focal neurological deficit of vascular mechanism and sudden onset that lasted more than 24 h. Fatal stroke was documented with evidence of a cerebrovascular mechanism using all available information, including death certificates and medical records. Brain imaging and clinical information were used to distinguish between types of stroke. The interobserver agreement on the classification of major stroke subtypes was excellent in both the PHS (κ = 0.81) (20) and WHS (κ = 0.96) (21).
Deaths were identified through systematic searches of the National Death Index. We identified cohort members who died before March 31, 2006, and obtained their death certificates from state agencies. Trained nosologists classified causes of death according to ICD-9 codes in conjunction with the “Automated Classification of Medical Entities Decision Tables” to select the underlying cause of death. The reliability of the National Death Index for epidemiologic purposes among female health professionals has been previously validated (98% sensitivity, ∼100% specificity) (22).
We compared participant characteristics according to BMI categories by chi-square tests for categorical variables and analysis of variance for continuous variables. We calculated Pearson correlations (r) among the indexes. For each index, we used Cox proportional hazards models to compute hazard ratios as the measure of the relative risks (RRs) and 95% confidence intervals (CIs) for CVD. Person-years of follow-up were calculated as the time from exposure assessment to development of the first end point of interest, censoring, or end of follow-up (March 31, 2006), whichever occurred first.
We considered 3 multivariable-adjusted models. Model 1 adjusted for potential confounders. In the PHS, these included age, physical activity, smoking, alcohol consumption, and parental history of myocardial infarction before the age of 60 years. In the WHS, these included age, physical activity, smoking, alcohol consumption, parental history of myocardial infarction before the age of 60 years, post-menopausal hormone use, race, education, and the dietary factors.
Anthropometric indexes were evaluated by: 1) comparing the strengths of the associations with CVD; and 2) comparing model fit as assessed by log-likelihoods. All log-likelihood comparisons involved models with constant sample sizes. For indexes demonstrating the strongest associations and best model fit, we examined a second model adjusting for the confounders in model 1 plus BMI. We compared nested models with and without BMI using the likelihood ratio test (LRT). In these models, the RRs for the indexes reflect associations with CVD beyond those conveyed by BMI.
We examined a third model adjusting for the variables in model 1 plus possible mediators of the association between adiposity and CVD (diabetes, elevated cholesterol, and hypertension). The RRs from this model reflect associations between the indexes and CVD beyond those mediated through the overt development of these intermediates.
In subanalyses, we excluded those with a history of smoking or cancer, those who developed the outcome or were censored early during follow-up (first 4 years in the PHS, 2 years in the WHS, given the shorter duration of follow-up in the WHS), and those with an absolute weight change ≥5% over the year preceding exposure assessment, to reduce bias due to potential confounding by smoking and pre-existing disease (23). In other analyses, we excluded persons with BMI ≥35 kg/m2, given potential misclassification of risk at the highest BMI.
In stratified and joint models, we explored interactions between anthropometric indexes and BMI, age, physical activity, or smoking status. To assess for statistically significant effect modification, we used the LRT contrasting age-adjusted models with and without interaction terms of interest.
We assessed for linear and curvilinear trends in the RR across categories of indexes by including the relevant indexes in models as continuous or quadratic variables, respectively, assigning median values to each category, and comparing models using the LRT. Two-sided p values were reported in all analyses. The p values <0.05 were considered statistically significant. All data analyses were performed using SAS software version 9.1 (SAS Institute, Cary, North Carolina).
Participant characteristics are shown according to baseline BMI categories in Tables 1 and 2.⇓ Men with higher BMI were younger; had higher WC, WHR, and WHtR indexes; and were more likely to have a history of hypertension, diabetes, and high cholesterol. They were also more likely to smoke, consumed less alcohol, and were less physically active. Leaner women were more likely to be current smokers; to use post-menopausal hormones; had lower levels of trans fat and omega-3 fatty acid intake; and had higher levels of polyunsaturated-to-saturated fat, glycemic load, folic acid, and cereal fiber intake.
The BMI most strongly correlated with WC (r = 0.78 for men, 0.82 for women) and WHtR (r = 0.80, 0.84), as well as weight (r = 0.86, 0.93).
A total of 1,505 cases of major CVD occurred in men after a median (standard deviation) follow-up of 14.2 (3.4) years (606 nonfatal myocardial infarctions, 604 nonfatal ischemic strokes, and 295 fatal CVD deaths), and 414 cases (174 nonfatal myocardial infarctions, 182 nonfatal ischemic strokes, and 58 fatal CVD cases) occurred in women after 5.5 (0.9) years. For all indexes, higher values were associated linearly with increasing risk of CVD, among both men and women and in both age- and multivariable-adjusted models (Tables 3 to 6).⇓⇓⇓⇓ Overall, associations with CVD did not vary substantially among the indexes, although associations were somewhat weaker for WHR, particularly among men (Fig. 1). We also found similar measures of model fit among the various indexes.
Among men, WHtR demonstrated the strongest gradient in the association with CVD, followed by WC, BMI, and WHR (Fig. 1A, Table 3). Measures of model fit were similar overall, with the best fit (lowest log-likelihood) seen using the WHtR (Tables 3 and 4). Results were generally similar for women. Both WHtR and WC demonstrated the strongest gradient in the association with CVD, with weaker associations for BMI and WHR (Fig. 1B, Table 5). Measures of model fit were overall similar, with the best fit seen using the WHtR (Tables 5 and 6). We found similar results examining indexes as continuous variables (data not shown).
Adding BMI to models with WHtR did not significantly improve model fit (p = 0.18 for men, 0.46 for women). Although the RRs for WHtR were attenuated after additionally adjusting for BMI, higher WHtR remained associated with increased risk of CVD (Tables 3 and 5). By contrast, adding WHtR to models with BMI did improve model fit and substantially attenuated the associations (Tables 3 and 5).
Additionally, adjusting for the potential intermediates in the association between adiposity and CVD generally attenuated the RRs for the various indexes; however, the overall associations remained comparable and statistically significant (Tables 4 and 6). We found similar results excluding participants with BMI ≥35 kg/m2 (data not shown).
Higher WHtR remained associated with increased risk of CVD after excluding persons with a history of smoking or cancer, unstable weight during the year prior to WHtR assessment, or with ≤4 (PHS) or ≤2 (WHS) years of follow-up (data not shown). By contrast, higher BMI was less strongly associated with increasing risk of CVD in this subgroup with a history of smoking or cancer or unstable weight. The RR (95% CI) was 1.43 (0.91 to 2.26) among obese (BMI ≥30 kg/m2) men and 1.68 (0.90 to 3.14) among obese women, compared with persons with BMI <25 kg/m2.
Examining the joint effects of BMI and WHtR, higher WHtR appeared strongly associated with CVD risk within BMI categories (Figs. 2A and 2B). In analyses stratified by BMI (</≥ 25 kg/m2), we did not find evidence for effect modification by BMI (LRT: p = 0.74 for men, 0.88 for women).
Higher WHtR and BMI were associated linearly with increased CVD risk among both younger and older men and women (Tables 7 and 8).⇓⇓ Although associations for both WHtR and BMI appeared stronger among younger participants, this finding was not statistically significant. Examining the joint effects of age and WHtR, those age ≥60 years consistently had the highest RR of CVD across all WHtR categories compared with younger persons (Figs. 3A and 3B). We also did not find evidence for effect modification by physical activity (LRT: p = 0.25 for men, 0.51 for women) or smoking status (LRT: p = 0.99 for men, 0.96 for women).
In these prospective cohorts of men and women, we found linear associations between higher adiposity measures and risk of incident CVD. Although all indexes demonstrated generally similar associations with CVD and measures of model fit, the WHtR most consistently showed the strongest associations and statistically best model fit. Overall, however, we did not find substantial or likely clinically meaningful differences between BMI and WHtR. These results were similar for both men and women.
Our findings remained materially unaltered after accounting for multiple confounders and potential intermediates in the pathway between adiposity and CVD, suggesting that the increased risk conferred by higher measures of adiposity may not solely be mediated by the development of diabetes, hypertension, or high cholesterol. Associations between indexes and CVD risk did not vary substantially by age, physical activity, or smoking status. Furthermore, associations between WHtR and CVD were attenuated but remained significant after adjusting for BMI, suggesting that much, but not all, of the risk conferred by a higher WHtR is reflected by BMI.
Previous studies on anthropometric indexes and cardiovascular risk have shown conflicting results. Both increased WC and WHR have been associated with higher coronary risk (7,8,24). Studies have been inconsistent, however, when comparing the various indexes.
Although WC may correlate more strongly with visceral fat than the WHR (9), particularly among the elderly (10), WC has not been a consistently stronger predictor of cardiovascular risk (25). In a cohort of male health professionals, WHR was particularly associated with coronary risk among the elderly, whereas BMI was more strongly associated among younger individuals (7). Conversely, in the Nurses' Health Study, both WHR and WC were associated with coronary risk after adjustment for BMI, with stronger associations among younger women (8). Furthermore, even though some suggest that WHR is comparably or more strongly associated with CVD compared with WC and BMI (6,26), others have shown stronger associations for WC (27–29). The WHtR has been more strongly associated with cardiovascular risk factors, such as hypertension, hyperglycemia, hypertriglyceridemia, and the metabolic syndrome, than BMI or WC has in selected populations, primarily among Asian populations (30–36).
Multiple biologic mechanisms have been implicated in mediating the adverse health effects of excess adiposity; however, the exact pathways are unknown (10). Visceral fat may be more sensitive to lipolysis, compared with subcutaneous fat, thereby preferentially increasing circulating free fatty acid levels (10). Other proposed mechanisms involve secretion of adipokines, which may differ by fat storage site (10).
Although our study has several strengths, including the prospective design, large sample size and number of outcome events, long duration of follow-up, measurement of multiple potential confounders, and confirmation of CVD events after medical record review, certain limitations should be considered. First, we used self-reported information on exposures, which may contribute to misclassification. However, validation studies of other health professional cohorts have demonstrated the reliability of self-reported information on anthropometric indexes and cardiovascular risk factors (16,37). Moreover, due to the prospective design, we expect that such potential misclassification would likely have yielded underestimates of effect.
Second, residual confounding may have occurred with self-reported information on comorbid conditions and potential mediators. However, we analyzed multiple biologically relevant confounders, and such residual confounding would not be expected to substantially alter the observed associations. Third, our study was limited to female health professionals and mostly Caucasian male physicians in the U.S., which may limit the generalizability of our results to other populations. However, our study population's homogeneity in income, educational attainment, and access to medical care may reduce confounding due to socioeconomic factors.
We found that higher measures of both overall and central adiposity confer greater risk of subsequent CVD in both men and women, regardless of the index chosen. Current clinical guidelines using BMI to define overweight and obesity may miss identifying persons at “normal” BMI levels with increased CVD risk related to central fat distribution. However, although the WHtR and central fat distribution may particularly reflect CVD risk, we found differences between BMI and WHtR in association with CVD and model fit to be small and likely not clinically consequential. Given its ease of measurement and current standard use in the classification of overweight and obesity, BMI may remain the most clinically practical measure of adiposity. Our findings emphasize that higher levels of adiposity, however measured, confer overall greater risk of CVD.
The authors are indebted to the participants in the Physician's Health Study and the Women's Health Study for their outstanding commitment and cooperation and to the entire Physicians' Health Study and Women's Health Study staff for their expert and unfailing assistance.
This work was supported by grants from the National Cancer Institute (#CA-34944, #CA-40360, #CA-047988, and #CA-097193), the National Heart, Lung, and Blood Institute (#HL-26490, #HL-34595, #HL-043851, and #HL-080467), and the National Institute on Aging (#5 T32 AG000158-18).
- Abbreviations and Acronyms
- body mass index
- blood pressure
- confidence interval
- cardiovascular disease
- likelihood ratio test
- relative risk
- waist circumference
- waist-to-hip ratio
- waist-to-height ratio
- Received November 16, 2007.
- Revision received February 19, 2008.
- Accepted March 4, 2008.
- American College of Cardiology Foundation
- World Health Organization
- Walker S.P.,
- Rimm E.B.,
- Ascherio A.,
- Kawachi I.,
- Stampfer M.J.,
- Willett W.C.
- Rimm E.B.,
- Stampfer M.J.,
- Giovannucci E.,
- et al.
- Snijder M.B.,
- van Dam R.M.,
- Visser M.,
- Seidell J.C.
- Willett W.C.,
- Sampson L.,
- Browne M.L.,
- et al.
- Willett W.,
- Stampfer M.J.
- Liu S.,
- Willett W.C.,
- Stampfer M.J.,
- et al.
- Berger K.,
- Kase C.S.,
- Buring J.E.
- Atiya M.,
- Kurth T.,
- Berger K.,
- Buring J.E.,
- Kase C.S.
- Rich-Edwards J.W.,
- Corsano K.A.,
- Stampfer M.J.
- Baik I.,
- Ascherio A.,
- Rimm E.B.,
- et al.
- Wang Y.,
- Rimm E.B.,
- Stampfer M.J.,
- Willett W.C.,
- Hu F.B.
- Colditz G.A.,
- Martin P.,
- Stampfer M.J.,
- et al.