Comparison of Adiposity Measures as Risk Factors in Postmenopausal Women

Abstract

Context:

There is a continuing debate about which adiposity measure is the best risk factor.

Objectives:

This study compared the associations of 14 health outcomes with combinations of four adiposity measures: body mass index (BMI), waist to hip ratio (WHR), waist, and waist to height ratio.

Design:

Data were from the Women's Health Initiative, a prospective study of women enrolled from 1993–1998 with a median follow-up time of 8 yr. Regression models were used to test the association of adiposity measures with outcome after adjusting for a number of variables related to demographic characteristics and health behavior.

Setting:

The women were recruited from 40 clinical centers throughout the United States.

Participants:

The sample analyzed included 141,652 postmenopausal women age 50–79 yr who met the criteria for the Women's Health Initiative randomized control trials.

Main Outcome Measures:

Outcomes included death and eight medical conditions.

Results:

Adiposity measures were most strongly associated with diabetes, hypertension, joint replacement, and gallbladder disease; moderately associated with myocardial infarction, endometrial cancer, and death; and least strongly associated with colon cancer, stroke, and breast cancer. Associations were nearly identical for waist and waist to height ratio. For most outcomes, waist was a stronger individual risk factor than BMI or WHR. However, BMI and WHR were the most useful combination of adiposity measures for stratifying participants according to risk of hypertension or diabetes.

Conclusions:

The adiposity measure most useful for stratifying persons on the basis of risk depends on the outcome of interest. When the outcome is diabetes or hypertension in postmenopausal women, the best indication of risk is a combination BMI and WHR.

Adiposity is a well-established risk factor for health outcomes (1, 2). It is commonly measured by the body mass index (BMI), which is weight in kilograms divided by height squared in meters. BMI was defined in the mid-19th century by the Belgian mathematician Adolphe Quetelet (3) and was popularized as a measure of obesity by Ancel Keys in 1972 (4). Although BMI depends on factors other than the percentage of body fat such as muscle mass and frame size, it is associated with the amount of body fat and is easily obtained. Body fat distribution can be measured by the ratio of waist girth to hip girth [waist to hip ratio (WHR)]. WHR was first used as a risk factor in studies published in the early 1980s (5–7) and has continued to be used as a risk factor (8–10). After the introduction of WHR, waist circumference became used as an adiposity measure in numerous studies. Waist girth by itself measures both adiposity and body fat distribution. More recently, waist has been divided by height (11–13). There is, and has been for some time, a debate about which of these measures is the best risk factor. The present study provides new information for this debate by analyzing data from a large, carefully collected prospective database to evaluate the association of a number of health outcomes with these adiposity measures used singly and in combination.

Materials and Methods

Data were obtained from the Women's Health Initiative (WHI). The WHI study design has been described in detail (3–5, 14–16). In brief, it was a long-term national health study that focused on strategies for preventing heart disease, breast and colorectal cancer, and osteoporosis in postmenopausal women. Women between the ages of 50 and 79 yr were enrolled for an observational study or randomized controlled trials from 1993–1998 at 40 clinical centers throughout the United States. The three therapies tested in the randomized controlled trials were estrogen therapy for women without a uterus, estrogen plus progestin for women with a uterus, and diet modification. All participants signed informed consent forms. The institutional review boards at all participating institutions, including the coordinating center, subcontractors, and clinical centers, approved the study protocols and procedures. The median follow-up time was 8 yr. Data available for analysis included 161,748 WHI participants: 93,651 from the observational study, 16,590 from the randomized controlled trial (RCT) of estrogen plus progestin, 10,722 from the RCT of estrogen only, and 40,785 additional women who were in the diet study but not either of the hormone therapy studies.

Sample selection

We eliminated from the observational study 9554 women who would not have qualified for the RCT for any of the following reasons: platelets fewer than 75,000/mm³, hematocrit less than 32%, oral daily use of a glucocorticosteroid, BMI less than 18 kg/m², systolic blood pressure greater than 200 mm Hg, diastolic blood pressure greater than 105 mm Hg, breast cancer ever, other cancers in the last 10 yr, or stroke, transient ischemic attack, or myocardial infarction (MI) in the last 6 months. Another 1693 participants were eliminated from the analyses because they were missing information on any of the adiposity measures, 30 were eliminated because BMI was less than 18 kg/m², and 2864 were eliminated because of missing information on physical function, a key covariable. From the remaining 147,607 subjects, additional subjects were removed for an analysis of a specific healthcare outcome because they were missing the follow-up date for that outcome. The maximum sample size in an analysis was 145,234 (for hypertension or gallbladder disease), and the minimum sample size was 143,807 (for joint replacement). For the analysis of systolic blood pressure (not for any other analysis), we eliminated the 35,133 participants under treatment for hypertension because the blood pressure was altered by medication. Therefore, the sample size for this analysis was 109,372.

Participants missing values for some covariables that had large numbers of missing data were not eliminated. There were 8965 participants missing information about income, 7790 missing information about strenuous exercise, and 38,991 missing information about endometrial cancer. Participants missing information on income or strenuous exercise were given the mean score for that variable. Participants missing information about endometrial cancer as an outcome were coded as none because only 0.5% of the reporting participants had endometrial cancer. To assess the effect of imputing missing data, we included in the regression equation two indicator variables: one for missing income and one for missing strenuous exercise. Including these two variables did not influence either the statistical significance or strength of association of any of the adiposity measures.

Data elements

The WHI provided information needed to construct the four adiposity measures and the conditions tested for an association with these measures. Height without shoes was measured by using a wall-mounted stadiometer that measures in centimeters. The waist was measured for erect subjects at the narrowest part of the torso at the end of a normal expiration. Hips were measured in a horizontal plane at the site of maximum extension of the buttocks. Associations between outcomes and adiposity measures that involved waist or hips would be weakened by inconsistencies or errors in these measures.

Outcomes identified as occurring during the follow-up period were diabetes, MI, stroke, breast cancer, endometrial cancer, colon cancer, and death. Outcomes identified by reported history in the baseline questionnaire were diabetes, systolic blood pressure (for participants not treated for hypertension), gallbladder disease, and joint replacement. It was not known whether the joint replacement was knee, hip, or shoulder. The inclusion of diabetes at baseline (in addition to diabetes that developed during follow-up) had the advantage of including the measure of risk most proximally related to disease but the disadvantage of measuring a risk factor that may have been influenced by the disease. The history of gallbladder disease was also included in the analysis because it was more common than the development of new gallbladder disease.

The other data elements used in the analyses were demographic and health behavior covariables significantly associated with at least one of the medical outcomes. These included age, race, specific WHI study, region of the United States, education, income, Medicaid insurance, parity, years smoking, alcohol intake, minutes of strenuous exercise per week, current smoker, years of smoking, use of estrogen, and use of estrogen plus progesterone. The same covariables were included in all analyses.

Statistical analysis

The association of an outcome with an adiposity measure was adjusted for covariables using a regression technique appropriate for the outcome: logistic regression for binary medical conditions assessed at baseline, Cox proportional hazard regression for outcomes that occurred during follow-up, and linear regression for systolic blood pressure at baseline.

The strength of association of the adiposity measure with a condition was measured using the odds ratio from logistic regression, the hazard ratio from the Cox proportional hazard regression, or the change in systolic blood pressure from linear regression. The odds ratio and hazard ratio are closely related to measures of relative risk. Measures of strength of association were based on an increment of 1 sd of the adiposity measure.

Associations of adiposity measures with the medical conditions were all significant at the P < 0.0001 level, which is the lowest P value commonly reported. Because it was not meaningful to compare P values, the statistical significances of the associations were compared on the basis of χ² values, which determine P values. The C statistic is not presented because it is associated with all of the variables and not only the adiposity measures of interest. Confidence intervals are also not presented because they were extremely narrow due to the large sample size; the extremes of the 95% confidence interval for hazard ratios all varied less than 0.04 from the estimates except for endometrial cancer and colon cancer.

Interaction terms were used to test whether the association between an adiposity measure and an outcome was influenced by a demographic factor. We did not report the statistical significance of nonlinear terms in our analysis because it complicated the presentation and did not qualitatively influence the results.

A graphical representation was used to show how the risk of new diabetes or current hypertension varied using a combination of BMI and WHR or BMI and waist. To obtain the points used in this graph, BMI was divided into three levels: nonobese (≤25 kg/m²), overweight (between 25 and 30 kg/m²), and obese (≥30 kg/m²). Waist and WHR were also divided into three levels: the lower 20th percentile, 20–80th percentile, and the upper 20th percentile. These divisions by percentiles rather than preset values made the comparisons between the waist and WHR groups more meaningful.

An indicator variable was formed for each of the nine combinations of waist and BMI. Eight of the nine indicator variables were included in the Cox proportional hazard regression model with the covariables; the indicator variable for the reference group (nonobese women with waist in the lowest 20th percentile) was not included. The same approach was used to analyze nine groups defined by BMI and WHR. The hazard ratios comparing levels of an adiposity measure were of course not the same as the hazard ratio based on an increment of 1 sd of the adiposity measure.

Statistical analyses were performed using SAS version 9 (SAS Institute Inc., Cary, NC).

Results

Participants are characterized in Table 1. Most were between the ages of 56 and 69, white, not currently smoking, not participating in regular strenuous exercise, and from the WHI observational study or RCT of diet. Nearly 40% had a college education.

Table 1.

Subject characteristics at baseline

Characteristics	Percentage (n = 147,607)
Age (yr)
49–55	17.82
56–69	60.90
70–81	21.28
Race
Caucasian	82.77
Non-Caucasian	17.23
Region of country
Northeast	23.12
South	25.48
Midwest	21.80
West	29.60
Education level
High school graduate or less	23.06
Post high school	37.85
College graduate or higher	39.09
Income
Unknown	6.54
<35,000	37.98
35,000–75,000	37.98
≥75,000	17.50
Medicaid insurance
Yes	1.31
No	98.69
Parity
No term pregnancies	12.18
1 or 2	33.66
3 or more	54.16
Current smoker
Yes	7.00
No	93.00
Smoke years
None	52.90
<10	11.38
10–19	10.27
≥20	25.45
Alcohol intake
<1 drink/month	42.01
<1 drink/wk to <7 drinks/wk	46.35
≥7 drinks/wk	11.65
Strenuous activity per week (min)
0	76.35
≤30	4.70
31–89	4.64
≥90	14.31
Use of estrogen
Yes	17.74
No	82.26
Use of estrogen plus progestin
Yes	17.20
No	82.80
WHI dataset
Observation study	55.07
RCT estrogen alone	7.03
RCT estrogen plus progestin	10.91
RCT diet (not in other RCT)	26.99

The distributions of the adiposity measures are shown in Table 2. In this sample, 30.3% of the participants would be defined as obese, BMI 30 kg/m² or more, and 34.8% would be defined as normal weight, BMI 25 kg/m² or less.

Table 2.

Value of obesity measure corresponding to a given percentile

Percentile	Obesity measures (n = 147,607)
	BMI (kg/m2)	Waist (cm)	WHR	WHtR
Minimum value	18.00	62.0	0.43	0.35
10th	21.67	70.8	0.72	0.44
20th	23.15	74.8	0.75	0.46
50th	26.93	84.5	0.80	0.52
80th	32.27	97.7	0.87	0.60
90th	35.74	105.0	0.91	0.65
Maximum value	60.91	125.0	1.47	0.85

All correlations between two adiposity measures were statistically significant at the P < 0.0001 level. The greatest correlation was between waist and waist to height ratio (WHtR), 0.97. The correlation between BMI and waist was also quite high, 0.85. The correlation between waist and WHR was lower, 0.71, and the correlation between BMI and WHR was much lower, 0.39.

As shown in Table 3, the adiposity measures had the highest strength of association measures and χ² values for diabetes, hypertension, joint replacement, and gallbladder disease. For example, when the value of waist was increased by 1 sd, the number of cases of new diabetes increased by 97%. The adiposity measures were less strongly associated with MI, endometrial cancer, and death and weakly but significantly associated with colon cancer, stroke, and breast cancer.

Table 3.

Associations of outcomes with risk-adjusted measures of obesity and body fat distribution in 147,607 participants from the WHI dataset

Outcome	% with outcome	BMI	Waist	WHtR	WHR	Waist added to BMIa	WHR added to BMIa
Diabetes at baseline	3.8	1.66 (1750)	2.20 (3202)	2.18 (3105)	2.08 (3129)	2.98 (1788)	1.93 (2364)
New diabetes during follow-up	6.3	1.65 (3491)	1.96 (4608)	1.97 (4695)	1.64 (3984)	1.99 (1395)	1.52 (2204)
Hypertension at baseline	29.2	1.64 (5990)	1.71 (6951)	1.73 (7014)	1.51 (4205)	1.49 (1193)	1.32 (1752)
SBP (untreated)b (n = 109,373)	100	3.09 (3461)	3.11 (3608)	3.26 (3854)	2.22 (1942)	1.86 (395)	1.33 (624)
MI during follow-up	2.3	1.21 (120)	1.30 (224)	1.31 (234)	1.27 (208)	1.42 (112)	1.22 (132)
Stroke during follow-up	1.9	1.11 (30)	1.17 (60)	1.17 (63)	1.16 (66)	1.23 (34)	1.14 (45)
Joint replaced at baseline	3.5	1.62 (1316)	1.59 (1019)	1.57 (966)	1.21 (170)	1.15 (32)	1.04 (6.1)
Cholecystectomy at baseline	14.7	1.50 (3101)	1.59 (3663)	1.60 (3701)	1.37 (1725)	1.44 (684)	1.22 (624)
Cholecystectomy during follow-up	7.4	1.23 (494)	1.26 (587)	1.26 (575)	1.18 (315)	1.21 (110)	1.12 (116)
Breast cancer during follow-up	4.1	1.10 (51)	1.10 (51)	1.08 (34)	1.05 (15)	1.06 (4.5)	1.02 (2.0)
Endometrial cancer during follow-up	0.6	1.36 (92)	1.36 (77)	1.33 (64)	1.15 (15)	1.13 (4.0)	1.03 (0.6)
Colon cancer during follow-up	0.8	1.15 (23)	1.21 (41)	1.20 (34)	1.18 (33)	1.28 (19)	1.14 (19)
Death	5.4	1.14 (137)	1.23 (332)	1.23 (309)	1.21 (305)	1.38 (229)	1.18 (208)

The strength of association was measured by the odds or hazard ratios associated with an increase in the obesity measure of 1 sd. The associated χ² value with one degree of freedom is given in parentheses. Covariables in all analyses included age, race, region, education level, income, physical activity, smoking, alcohol, source of participants, and RCT treatments.

^aThe strength of association and χ² values have been adjusted by including BMI in the equation.

^bThe analyses of systolic blood pressure (SBP) was unique in two ways. 1) The strength of association measure was the average increase in blood pressure for an increase of 1 sd of the obesity measure. 2) Subjects treated for hypertension were eliminated.

Based on χ² values, waist or WHtR were the most statistically significant single risk factors for all conditions except for history of joint replacement and a history of endometrial or breast cancer at baseline. Results for waist and WHtR were about the same as expected given their high correlation.

The χ² value for WHR added to BMI was greater than for waist added to BMI for diabetes, hypertension, systolic blood pressure, and to a lesser extent MI and stroke. For example, the χ² value for WHR added to BMI was 2153 for new diabetes, and the χ² value for waist added to BMI was 1365 for new diabetes. The higher χ² value signifies that BMI plus WHR is more strongly associated with new diabetes than BMI plus waist.

Figure 1 graphs the relative risk of developing new diabetes for participants in a specific BMI and WHR (or waist) group compared with the group with BMI less than or equal to 25 kg/m² and WHR (or waist) in the lowest 20th percentile. For each level of BMI, the risk of diabetes increases with either WHR or waist especially for the most obese participants. Also shown in the figure is that for a given value of WHR or waist, the risk of diabetes increases with increasing BMI. There were two differences between the WHR and waist graphs. One is that the risk of diabetes increases more for increasing WHR than for increasing waist (at least for obese and normal-weight participants). The second difference is that there are 15 times as many obese participants with low WHR as with low waist and 23 times as many normal-weight participants with high WHR as with high waist. (This result is not surprising because the correlation between BMI and WHR was 0.39 compared with a correlation of 0.85 between BMI and waist.) Therefore, if BMI is used to assess diabetes risk, this assessment will be altered for many more patients based on WHR than on waist.

Fig. 1.

Relative hazard ratios for diabetes according to BMI and WHR or waist. The three WHR (waist) groups were lowest 20th percentile, 20–80th percentile, and 80th percentile. The hazard ratios were obtained from the Cox proportional hazard model that included covariables. The reference group was normal-weight subjects in the lowest 20th percentile of WHR (waist).

Figure 2 shows the results for hypertension for the combination of adiposity measures. For a given level of BMI, the increases in hypertension are similar for increases with waist and WHR, but as with diabetes, the assessment of risk based on BMI will be altered for many more patients based on WHR than on waist.

Fig. 2.

Relative odds ratios for hypertension according to BMI and WHR or waist. The three WHR (waist) groups were lowest 20th percentile, 20–80th percentile, and 80th percentile. The hazard ratios were obtained from the Cox proportional hazard model that included covariables. The reference group was normal-weight subjects in the lowest 20th percentile of WHR (waist).

We tested each of the characteristics in Table 1 to examine their influence on the association between the adiposity measures and diabetes. Most characteristics did not influence this association. The participant characteristics that increased the association of BMI with diabetes at the P < 0.0001 level were white race (χ² = 112), younger age (χ² = 23), and greater income (χ² = 46). The same factors significantly increased the association of waist and WHtR with diabetes, white race (χ² = 96 and 73), younger age (χ² = 54 and 71), and greater income (χ² = 21 and 30). The association of WHR with diabetes had less interaction with other participant characteristics. The factors that significantly increased the association of WHR with diabetes at the P < 0.0001 level were white race (χ² =28) and younger age (χ² =36). Therefore, the associations between WHR and diabetes were less influenced by race and age than were the associations of diabetes with BMI or waist. Also, the association between WHR and diabetes was not influenced by any other factors, such as income.

Discussion

The present study used a large comprehensive dataset to compare the value of the BMI, waist, WHR, and WHtR as risk factors for health outcomes in postmenopausal women. The large sample size and careful assessment of many outcomes made it possible to have precise estimates of the association of the adiposity measure with outcome and to compare the adiposity measures for their associations with many outcomes. All adiposity measures were highly associated with diabetes, hypertension, joint replacement, and gallbladder disease. They were less strongly associated with MI, endometrial cancer, and death and least strongly associated with colon cancer, stroke, and breast cancer.

Waist had a stronger association than BMI with all outcomes except for joint replacement, endometrial cancer, and breast cancer. WHtR was highly correlated with waist (0.97) and had similar associations with most outcomes.

Although waist was a stronger single risk factor than BMI for most conditions, the combination of adiposity measures most significantly associated with diabetes and hypertension, (and to a lesser extent MI and stroke) was WHR and BMI. The relative risk of each BMI/WHR group was compared with the group with BMI less than or equal to 25 kg/m² and WHR (or waist) in the lowest 20th percentile. For obese patients in the upper 20th percentile of WHR, relative risk of developing diabetes was 11.5 and of having hypertension was 5.5. WHR identified many more obese participants than waist or WHtR as having only moderate risk of diabetes or hypertension.

BMI is a measure of overall adiposity and WHR of body fat distribution. These measures are relatively independent as shown by their correlation of 0.39 and studies that found they have an independent genetic basis (17). Body fat distribution may be an important risk factor for some conditions because adipose tissue on the waist is more metabolically active than adipose tissue on the hips (18). Waist and WHtR are the best single measures of risk because they are highly correlated with both overall adiposity and body fat distribution. However, they provided less information about risk than the combination of more specific measures: one for overall adiposity (BMI) and one for body fat distribution (WHR).

Previous studies

There has been no consensus among previous studies as to the best adiposity measures. In most cases, the adiposity measures have been compared as individual risk factors for diabetes, hypertension, cardiovascular disease, and death. Some studies have found that WHR was as good or better risk factor than other adiposity measures (10, 19–22). Others found that the obesity measures provided similar measures of risk (23–27), that waist was the best measure of risk (28, 29), that WHR was the least valuable of these risk factors (30, 31), or that WHtR was the best measure (32–34).

Some authors considered BMI and WHR as complementary rather than competing measures (10, 15, 35). They found that WHR provided additional risk information beyond BMI, more so for women than for men. Other studies found that WHR and waist were also independent predictors of diabetes (36, 37).

Conclusions from different studies may differ for a number of reasons. One is that the populations may differ. Our study found that the associations between an obesity measure and disease may depend on race, age, or income. Other studies have found that the associations may be influenced by age (21, 31), race (22, 38), and gender (10, 26). Results may also be influenced by the statistical methodology used, which varied widely.

The results in the present study used girth measurements that were taken by professionals. The association of girth measurements with outcomes may be weaker if self-measurements are taken by the subjects. However, this may not be a major problem; the studies that first identified WHR as a risk factor and found it added more to BMI than did waist were based on self-measurement (6, 7).

Conclusion

According to the American Heart Association, WHR is less accurate than BMI or waist circumference and is no longer recommended (Body Composition Tests: http://www.heart.org/HEARTORG/GettingHealthy/NutritionCenter/Body-Composition-Tests_UCM_305883_Article.jsp). Our findings do not support this assessment for white postmenopausal women. Instead, the findings suggest that risk assessment for diabetes or hypertension in these women can be improved by incorporating information on WHR in addition to BMI.