Quantifying the proportion of deaths due to body mass index‐ and waist circumference‐defined obesity

Abstract

Objective

To determine the risk of mortality associated with and quantify the deaths attributable to combinations of body mass index (BMI) and waist circumference (WC).

Methods

This study included 41,439 participants. For the hazard ratio (HR) calculation, adiposity categories were defined as: BMI^N/WC^N, BMI^N/WC^O, BMI^O/WC^N, and BMI^O/WC^O (N = non‐obese, O = obese). For the population attributable fraction analysis, obesity was classified as: (i) obese by BMI and/or WC; (ii) obese by BMI; and (iii) obese by WC. Mortality data was complete to the end of 2012.

Results

The prevalence of BMI^N/WC^N, BMI^N/WC^O, BMI^O/WC^N, and BMI^O/WC^O was 73%, 6%, 6%, and 15%, respectively. There was an increased risk of all‐cause and cardiovascular disease (CVD) mortality in those with BMI^N/WC^O (HR (95% CI) 1.2 (1.2, 1.3) and 1.3 (1.1, 1.6)) and BMI^O/WC^O (1.3 (1.3, 1.4) and 1.7 (1.5, 1.9)) compared to those with BMI^N/WC^N. The estimated proportion of all‐cause and CVD mortality attributable to obesity defined using WC or using BMI and/or WC was higher compared to obesity defined using BMI.

Conclusions

Current population obesity monitoring misses those with BMI^N/WC^O who are at increased risk of mortality. By targeting reductions in population WC, the potential exists to prevent more deaths in the population than if we continue to target reductions in BMI alone.

Introduction

Although obesity in populations is most commonly defined using body mass index (BMI), the anthropometric measure(s) that actually best identify high‐risk adiposity has been much debated. Methods including dual‐energy X‐ray absorptiometry and bioelectrical impedance analysis reliably estimate total body fat. However, they are too cumbersome for use in large‐scale population surveys, and anthropometric measures remain a necessary, simpler, and cheaper alternative. Much work has been done to compare the predictive ability of adiposity measures such as BMI, waist circumference (WC), waist‐hip ratio (WHR), and waist‐height ratio (WHtR) in relation to health risks (1-4). The general consensus has been that other measures for obesity were not clearly superior to BMI, and thus the latter has remained the primary measure used to determine obesity.

Recently published findings in Australian adults reported that of all individuals with a high‐risk WC, almost 40% had a BMI below the obese range (5). A similar discordance was reported in Chinese adults where a large number of individuals with abdominal obesity would not have been identified by their BMI, and that among people with a non‐obese BMI, those with a high‐risk WC had a greater risk of developing hypertension compared to those with a non‐high‐risk WC (6). The BMI and WC cut‐points used to categorize obesity in these studies were ones recommended in clinical guidelines (7-10) and which are commonly used in population surveillance of obesity. The implication of this discordance between BMI and WC is that there is a subgroup of the population who may be at increased health risk due to excess adiposity and who would be missed by current obesity monitoring which relies heavily on BMI measures alone (11).

There is currently little empirical evidence on the health risks of those with a discordant BMI and WC: non‐obese BMI and high‐risk WC, and obese BMI and non‐high‐risk WC. While there have been suggestions that within each BMI category those with a high‐risk WC are at increased risk of hypertension, type 2 diabetes, dyslipidemia, and the metabolic syndrome compared to those with a non‐high‐risk WC (6, 12), the relationship with mortality has not been explored. Furthermore, no study has examined how the mortality risk in those with a discordant BMI and WC compares to individuals who have a non‐obese BMI and non‐obese WC and individuals who have both an obese BMI and obese WC. The aim of the present study was to determine the risk of all‐cause and cardiovascular disease (CVD) mortality in different adiposity categories and to quantify the deaths attributable to excess adiposity defined using combinations of BMI and WC.

Methods

Study population

The Melbourne Collaborative Cohort Study (MCCS) population and methods have been described in detail previously (13). Briefly, at baseline in 1990 to 1994, 41,514 participants were recruited from the Melbourne metropolitan area. The participants were aged between 27 and 76 years, with the majority (97%) aged between 40 and 69 years. Southern European migrants to Australia were oversampled to increase the genetic variation of the study population and to extend the range of lifestyle exposures. The current study included 41,439 participants. Participants were excluded if they were missing any of the following (numbers (n) are not additive): BMI (n = 31), WC (n = 38), physical activity (n = 9), alcohol consumption (n = 39), smoking (n = 10), education (n = 9), and cancer (n = 9). This study was approved by the Cancer Council Victoria's Human Research Ethics Committee and by the Alfred Hospital Human Research Ethics Committee.

Data collection

Weight was measured to the nearest 0.1 kg using digital electronic scales. Height was measured to the nearest millimetre using a stadiometer. BMI was calculated as weight (kg) divided by height (m) squared, and categorized as: (i) non‐obese: BMI <30 kg/m²; and (ii) obese: BMI ≥30 kg/m². WC was measured to the nearest millimetre using a metal anthropometric tape at the narrowest part of the torso, and categorized as: (i) non‐obese: men <102 cm, women <88 cm; and (ii) obese: men ≥102 cm, women ≥ 88 cm. Adiposity categories were created using a combination of BMI and WC as: (i) BMI^N/WC^N; (ii) BMI^N/WC^O; (iii) BMI^O/WC^N; and (iv) BMI^O/WC^O, where N = non‐obese and O = obese.

Deaths were identified using the Victorian Registry of Births, Deaths and Marriages, and the National Death Index. Mortality data was complete to end of 2012. Deaths were attributed to CVD if the underlying cause of death was coded I10‐I25, I46.1, I48, I50‐I99, or R96 according to the 2006 International Classification of Diseases10th revision (ICD‐10).

Information on country of birth, smoking, education, and physical activity were collected via questionnaire. Country of birth was categorized as: (i) Australia/New Zealand/Northern Europe; and (ii) Southern Europe. Smoking status was categorised as: (i) never smoker; (ii) ex‐smoker; and (iii) current smoker. Education was categorized as: (i) did not complete high school; (ii) completed high/technical school or other qualification (e.g., trade certificate); and (iii) completed a tertiary degree or diploma. A physical activity score was calculated based on self‐reported number of times walked, number of times exercised vigorously, and number of times exercised non‐vigorously per week over the last 6 months (14). A greater physical activity score corresponds to higher levels of physical activity. Alcohol consumption was determined using the Cancer Council Victoria's Food Frequency Questionnaire (15).

Statistical analysis

Hazard ratios and 95% confidence intervals (CI) for the relationship between adiposity categories and mortality was explored using Cox proportional hazards regression model. The multivariate analysis was adjusted for age, sex, smoking, country of birth, physical activity, education, and alcohol consumption. To test for effect modification, an interaction term was included between sex, age (dichotomized into <65 years and ≥65 years), and country of birth with adiposity categories. As significant interactions were detected for age and country of birth, all analyses were stratified by these factors. Participants were censored at death, leaving Victoria or leaving Australia, or last date known to be alive (end of 2012), whichever came first.

Population attributable fraction (PAF) was calculated to determine the proportion of deaths theoretically attributable to obesity defined in three ways: (i) obese according to a combination of BMI and/or WC (BMI^N/WC^O, BMI^O/WC^N, and BMI^O/WC^O); (ii) obese according to BMI (BMI^O/WC^N and BMI^O/WC^O); and (iii) obese according to WC (BMI^N/WC^O and BMI^O/WC^O). PAF was calculated using the “punafcc” command in Stata^®, which implements the method recommended by Greenland and Drescher (16):

Goodness‐of‐fit for each model was compared using the Akaike Information Criterion (AIC) computed as −2(log‐likelihood)+2(number of estimated parameters), with a lower AIC indicating a better fit (17).

Sensitivity analysis

As a sensitivity analysis, we excluded smokers (n = 4,673), those underweight (BMI <18.5 kg/m²; n = 269), those who died within one year of baseline (n = 95), and those known to have cancer at baseline (n = 3,174), to avoid bias in the relationship between body weight and mortality that is related to ill‐health (total excluded n = 7,760). We also tested the effect of excluding those who died within five years of baseline. The sensitivity analysis illustrates the relationship between obesity and mortality in otherwise healthy persons, but this was not included as part of the main analysis because it does not represent the actual deaths associated with obesity in the entire population, which includes the less healthy.

All analyses were performed using Stata^® statistical software package version 11.2 (StataCorp, College Station, TX).

Results

The mean age of the study population was 55 years (SD 9 years) and 59% were women (Table 1). About 73% were non‐obese (BMI^N/WC^N), 6% had obesity based on their WC but not their BMI (BMI^N/WC^O), 6% had obesity based on their BMI but not their WC (BMI^O/WC^N), and 15% had obesity according to both their BMI and WC (BMI^O/WC^O). Figure 1 presents the prevalence by sex. Over the follow‐up period, 7,953 participants died and 1,445 were CVD‐related deaths.

Figure 1

Prevalence of each adiposity category (BMI^N/WC^N: non‐obese BMI and WC; BMI^N/WC^O: non‐obese BMI, obese WC; BMI^O/WC^N: obese BMI, non‐obese WC; BMI^O/WC^O: obese BMI and WC) in: (A) men and (B) women. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Table 1. Characteristics of the study population

	Adiposity categories
	BMI^N/WC^N	BMI^N/WC^O	BMI^O/WC^N	BMI^O/WC^O
n (%)	30,223 (72.9)	2,637 (6.4)	2,310 (5.6)	6,269 (15.1)
Age (years)* *P for difference between adiposity categories <0.05.	54.7 ± 8.8	59.2 ± 7.8	55.0 ± 8.3	56.8 ± 8.0
Men, n (%)* *P for difference between adiposity categories <0.05.	12,785 (42.3)	995 (37.7)	1,028 (44.5)	2.188 (34.9)
BMI (kg/m^2)* *P for difference between adiposity categories <0.05.	25.0 ± 2.7	28.1 ± 1.5	31.5 ± 1.5	34.2 ± 3.6
Waist circumference (cm)* *P for difference between adiposity categories <0.05.	80.4 (73.0‐88.5)	95.2 (90.0‐103.3)	87.0 (83.0‐97.6)	102.4 (94.3‐108.5)
Country of birth, n (%)* *P for difference between adiposity categories <0.05.
Australia/New Zealand/Northern Europe	24,201 (80.1)	1,861 (70.6)	1,308 (56.6)	3,741 (59.7)
Southern Europe	6,022 (19.9)	776 (29.4)	1,002 (43.4)	2,528 (40.3)
Physical activity score* *P for difference between adiposity categories <0.05.	4 (1.5‐6)	4 (1.5‐4)	3 (0‐4)	3 (0‐4)
Alcohol (g/day)* *P for difference between adiposity categories <0.05.	4.8 (0‐17.4)	2.6 (0‐17.3)	2.1 (0‐15.5)	0.8 (0‐14.7)
Smoking status, n (%)* *P for difference between adiposity categories <0.05.
Never smoker	17,345 (57.4)	1,420 (53.9)	1,323 (57.3)	3,705 (59.1)
Ex‐smoker	9,429 (31.2)	896 (34.0)	720 (31.2)	1,928 (30.8)
Current smoker	3,449 (11.4)	321 (12.2)	267 (11.6)	636 (10.2)
Education, n (%)* *P for difference between adiposity categories <0.05.
Did not complete high school	15,874 (52.5)	1,788 (67.8)	1,624 (70.3)	4,572 (72.9)
Completed high/technical school or some tertiary education	6,672 (22.1)	469 (17.8)	392 (17.0)	1,032 (16.5)
Completed tertiary degree or diploma	7,677 (25.4)	380 (14.4)	294 (12.7)	665 (10.6)
Cancer, n (%)* *P for difference between adiposity categories <0.05.	2,334 (7.7)	267 (10.1)	149 (6.5)	424 (6.8)
Deaths, n (%)* *P for difference between adiposity categories <0.05.	5,193 (17.2)	777 (29.5)	412 (17.8)	1,571 (25.1)
CVD deaths, n (%)* *P for difference between adiposity categories <0.05.	887 (2.9)	153 (5.8)	86 (3.7)	319 (5.1)

• Data is presented as mean ± SD or median (interquartile range), unless otherwise specified.
• *P for difference between adiposity categories <0.05.

In the multivariate analysis, individuals with BMI^N/WC^O and BMI^O/WC^O, but not BMI^O/WC^N, had an increased risk of all‐cause and CVD mortality compared to those with BMI^N/WC^N (Table 2). The magnitude of increase in risk was similar for those with BMI^N/WC^O and BMI^O/WC^O, and the relationship was not modified by age or country of birth. In those aged ≥65 years, the risk of CVD mortality was similar across all adiposity categories compared to BMI^N/WC^N. Exclusion of those with indicators of poorer health did not substantially alter our results (Supporting Information Table 1).

Table 2. Multivariate hazard ratio (HR) for mortality according to body weight categories

		All‐cause mortality, HR (95% CI)		CVD mortality, HR (95% CI)
	N	Crude	Adjusteda^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).	Crude	Adjusteda^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).
Total population
BMI^N/WC^N	30238	1.0 (ref)	1.0 (ref)	1.0 (ref)	1.0 (ref)
BMI^N/WC^O	2641	1.6 (1.5, 1.7)	1.2 (1.2, 1.3)	2.0 (1.7, 2.4)	1.3 (1.1, 1.6)
BMI^O/WC^N	2313	0.9 (0.8, 1.0)	1.0 (0.9, 1.1)	1.2 (0.9, 1.5)	1.3 (1.0, 1.6)
BMI^O/WC^O	6279	1.3 (1.2, 1.4)	1.3 (1.3, 1.4)	1.7 (1.5, 2.0)	1.7 (1.5, 1.9)
Age <65 years
BMI^N/WC^N	25232	1.0 (ref)	1.0 (ref)	1.0 (ref)	1.0 (ref)
BMI^N/WC^O	1855	1.5 (1.3, 1.7)	1.5 (1.3, 1.6)	1.6 (1.2, 2.2)	1.6 (1.2, 2.1)
BMI^O/WC^N	1988	1.0 (0.9, 1.1)	1.0 (0.8, 1.1)	1.3 (0.9, 1.8)	1.1 (0.8, 1.6)
BMI^O/WC^O	5134	1.4 (1.3, 1.5)	1.5 (1.3, 1.6)	2.1 (1.7, 2.5)	2.1 (1.7, 2.5)
Age ≥65 years
BMI^N/WC^N	5006	1.0 (ref)	1.0 (ref)	1.0 (ref)	1.0 (ref)
BMI^N/WC^O	786	1.2 (1.1, 1.3)	1.2 (1.1, 1.4)	1.4 (1.2, 1.8)	1.5 (1.2, 1.8)
BMI^O/WC^N	325	0.9 (0.8, 1.1)	1.0 (0.8, 1.1)	1.4 (0.9, 1.9)	1.4 (0.9, 1.8)
BMI^O/WC^O	1145	1.2 (1.1, 1.3)	1.3 (1.2, 1.4)	1.4 (1.1, 1.7)	1.5 (1.2, 1.8)
Country of birth: Australia/New Zealand/Northern Europe
BMI^N/WC^N	24212	1.0 (ref)	1.0 (ref)	1.0 (ref)	1.0 (ref)
BMI^N/WC^O	1865	1.6 (1.5, 1.8)	1.2 (1.1, 1.3)	1.9 (1.6, 2.4)	1.3 (1.0, 1.6)
BMI^O/WC^N	1308	1.0 (0.9, 1.1)	1.0 (0.9, 1.1)	1.1 (0.8, 1.5)	1.2 (0.9, 1.6)
BMI^O/WC^O	3748	1.5 (1.4, 1.6)	1.4 (1.3, 1.5)	1.8 (1.5, 2.1)	1.7 (1.4, 1.9)
Country of birth: Southern Europe
BMI^N/WC^N	6026	1.0 (ref)	1.0 (ref)	1.0 (ref)	1.0 (ref)
BMI^N/WC^O	776	1.6 (1.4, 1.8)	1.3 (1.1, 1.5)	2.2 (1.6, 3.0)	1.5 (1.1, 2.1)
BMI^O/WC^N	1005	1.0 (0.8, 1.1)	1.0 (0.8, 1.1)	1.4 (0.9, 1.9)	1.4 (0.9, 1.9)
BMI^O/WC^O	2531	1.3 (1.2, 1.4)	1.3 (1.2, 1.4)	1.6 (1.3, 2.1)	1.6 (1.3, 2.1)

• ^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).
• Bold text indicates statistical significance (P < 0.05).

As shown in Tables 3 and 4, the proportion of the population that had an obese BMI (21%) was similar to the proportion that had an obese WC (22%). The prevalence of obesity defined using a combination of BMI and/or WC and defined using WC alone was higher in the older compared to the younger age group, while the prevalence of obesity defined using BMI alone did not differ by age. The prevalence of obesity according to each definition was approximately double in those born in Southern Europe compared to those born in Australia/New Zealand/Northern Europe.

Table 3. Estimated proportion of total deaths attributable to excess adiposity according to definition of excess adiposity (BMI or WC)

	Obese by BMI and/or WC			Obese by BMI			Obese by WC
	PAF (95% CI)	Obesity prevalence (%)	AIC	PAF (95% CI)	Obesity prevalence (%)	AIC	PAF (95% CI)	Obesity prevalence (%)	AIC
Total population (n = 41,439)
Total population	8.0 (6.6, 9.5)	27.1	154,799	3.3 (2.1, 4.6)	20.7	154,888	8.5 (7.2, 9.8)	21.5	154,739
Total populationa^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).	6.8 (5.2, 8.3)	27.1	149,591	4.4 (3.1, 5.6)	20.7	149,619	6.9 (5.5, 8.3)	21.5	149,561
Age <65 years (n = 34,183)
Total population	8.6 (6.6, 10.6)	26.2	85,180	5.0 (3.3, 6.6)	20.8	85,221	8.6 (6.9, 10.3)	20.4	85,153
Total populationa^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).	8.9 (6.9, 10.9)	26.2	84,546	5.4 (3.6, 7.1)	20.8	84,584	9.2 (7.5, 10.9)	20.4	84,505
Age ≥65 years (n = 7,256)
Total population	4.4 (2.2, 6.6)	31.0	55,815	2.0 (0.3, 3.8)	20.2	55,824	4.8 (2.8, 6.8)	26.6	55,807
Total populationa^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).	6.1 (3.8, 8.3)	31.0	55,500	3.3 (1.6, 5.1)	20.2	55,514	6.4 (4.4, 8.3)	26.6	55,489
Country of birth: Australia/New Zealand/Northern Europe (n = 31,111)
Total population	8.4 (6.9, 9.9)	22.2	111,143	4.2 (3.0, 5.4)	16.2	111,224	8.5 (7.2, 9.8)	18.0	111,103
Total populationa^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).	5.8 (4.2, 7.4)	22.2	107,074	4.0 (2.7, 5.2)	16.2	107,086	5.8 (4.4, 7.2)	18.0	107,060
Country of birth: Southern Europe (n = 10,328)
Total population	10.5 (6.8, 14.2)	41.7	34,498	4.0 (0.7, 7.2)	34.2	34,522	10.9 (7.8, 13.9)	32.0	34,481
Total populationa^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).	8.6 (4.4, 12.6)	41.7	33,430	4.5 (9.7, 7.9)	34.2	33,440	9.3 (5.9, 12.6)	32.0	33,417

• Obese by BMI or WC: BMI ≥30 kg/m² or WC ≥102 cm for men and ≥88 cm for women; obese by BMI: BMI ≥30 kg/m²; obese by WC: WC ≥102 cm for men and ≥88 cm for women.
• ^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).

Table 4. Estimated proportion of CVD deaths attributable to excess adiposity according to definition of excess adiposity (BMI or WC)

	Obese by BMI and/or WC			Obese by BMI			Obese by WC
	PAF (95% CI)	Obesity prevalence (%)	AIC	PAF (95% CI)	Obesity prevalence (%)	AIC	PAF (95% CI)	Obesity prevalence (%)	AIC
Total population (n = 41,439)
Total population	15.7 (12.2, 19.1)	27.1	29,840	9.0 (6.0, 11.9)	20.7	29,888	14.3 (11.2, 17.3)	21.5	29,832
Total populationa^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).	12.7 (9.0, 16.3)	27.1	27,818	9.3 (6.3, 12.2)	20.7	27,826	11.0 (7.7, 14.2)	21.5	27,819
Age <65 years (n = 34,183)
Total population	17.8 (12.6, 22.7)	26.2	13,639	14.2 (9.6, 18.6)	20.8	13,648	15.9 (11.3, 20.3)	20.4	13,635
Total populationa^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).	16.8 (11.5, 21.8)	26.2	13,341	13.3 (8.6, 17.8)	20.8	13,348	15.9 (11.2, 20.2)	20.4	13,329
Age ≥65 years (n = 7256)
Total population	10.7 (5.7, 15.4)	31.0	13,173	5.6 (1.8, 9.2)	20.2	13,184	8.7 (4.3, 13.0)	26.6	13,176
Total populationa^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).	11.6 (6.6, 16.4)	31.0	13,088	6.2 (2.3, 9.9)	20.2	13,099	9.7 (5.1, 14.0)	26.6	13,091
Country of birth: Australia/New Zealand/Northern Europe (n = 31,111)
Total population	13.5 (9.8, 17.1)	22.2	21,077	7.8 (4.7, 10.8)	16.2	21,109	12.9 (9.5, 16.1)	18.0	21,068
Total populationa^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).	10.1 (6.2, 13.8)	22.2	19,516	7.5 (4.5, 10.5)	16.2	19,518	9.2 (5.7, 12.6)	18.0	19,515
Country of birth: Southern Europe (n = 10,328)
Total population	22.0 (13.1, 30.0)	41.7	7,077	11.7 (3.9, 18.8)	34.2	7,093	17.9 (10.3, 24.8)	32.0	7,078
Total populationa^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).	19.1 (9.6, 27.6)	41.7	6,612	12.8 (5.0, 19.9)	34.2	6,618	14.9 (6.9, 22.2)	32.0	6,614

• Obese by BMI or WC: BMI ≥30 kg/m² or WC ≥102 cm for men and ≥88 cm for women; obese by BMI: BMI ≥30 kg/m²; obese by WC: WC ≥102 cm for men and ≥88 cm for women.
• ^a Adjusted for sex, smoking status, physical activity, education, and alcohol intake (total population adjusted for age and country of birth; age strata adjusted for country of birth; country of birth strata adjusted for age).

As the hazard ratios indicated an increased risk of mortality in those with BMI^N/WC^O, we calculated PAFs for mortality according to different definitions of obesity to assess the effect of including those with BMI^N/WC^O in the obesity definition (as “obese by WC” and “obese by BMI and/or WC”). The PAF indicates that the estimated proportion of all‐cause and CVD mortality attributable to obesity was up to 50% and 30% higher, respectively, when obesity was defined using WC or using BMI and/or WC compared to obesity defined using BMI, after adjustment for potential confounders (Tables 3 and 4). In all subgroups of country of birth and age, the estimated proportion of all‐cause and CVD mortality attributable to obesity was lower when obesity was defined using BMI alone than when defined using WC alone or using BMI and/or WC combined, though only the difference in all‐cause mortality PAFs for obesity by BMI and obesity by WC in those aged <65 years reached statistical significance. The magnitude of the PAFs was generally higher in the younger age group and in those born in Southern Europe compared to the older age group and those born in Australia/New Zealand/Northern Europe, respectively. After excluding those with indicators of poorer health (Supporting Information Table 2), the estimated proportion of all‐cause and CVD mortality attributable to obesity was up to 1.6 and 1.3 times higher, respectively, when obesity was defined using WC or using BMI and/or WC compared to obesity defined using BMI. Similar results were found when we additionally excluded those who died within 5 years of baseline (data not shown).

In the total population, in relation to all‐cause mortality, AIC was lowest (optimal) for the multivariate model with obesity defined using WC alone compared to the model with obesity defined using a combination of BMI and/or WC, or using BMI only (Table 3). This did not differ by age or by country of birth. In relation to CVD mortality, AIC for the total population was lowest for the multivariate model with obesity defined using a combination of BMI and/or WC and highest for the model with obesity defined using BMI alone (Table 4). There was little difference in AIC between models with obesity defined using a combination of BMI and/or WC and models with obesity defined using WC alone. Similar results for AIC were generally observed by age and by country of birth.

Discussion

In this population of Australian adults, over one‐quarter of the population had obesity based on either their BMI or WC. The prevalence of BMI‐defined and WC‐defined obesity was similar (21% and 22%, respectively). However, while some of those identified as having obesity by their BMI and by their WC were the same individuals, there was also some discordance where 6% of those with an obese BMI had a non‐obese WC and 6% of those with an obese WC had a non‐obese BMI. The estimated proportion of deaths attributable to obesity was lower when obesity was defined using BMI alone compared to when obesity was defined using a combination of BMI and/or WC or using WC alone. AIC was highest for models with obesity defined using BMI alone, indicating that out of the three different methods used to define obesity, the definition using BMI alone had the poorest fit. When compared to the non‐obese (BMI^N/WC^N), those with BMI^N/WC^O and BMI^O/WC^O, but not those with BMI^O/WC^N, had an increase in risk of all‐cause and CVD mortality.

While the discordance between BMI and WC in classifying individuals as having obesity has been described in a number of populations (6, 18, 19), less has been done to characterize the health risks associated with the different combinations of BMI and WC. There is some evidence that across all BMI categories, those with an obese WC have an increased risk of hypertension, type 2 diabetes, dyslipidemia, and metabolic syndrome compared to those with a non‐obese WC (6, 12). Our study further showed that those with an obese WC, regardless of whether their BMI was in the obese range, also had an increased risk of mortality compared to their non‐obese counterparts. To the best of our knowledge, only one previous study has compared the PAFs for mortality associated with obesity defined using BMI or WC, demonstrating a similar PAF for BMI‐defined and WC‐defined obesity among non‐smoking men (10.1% and 10.2%) (20). No relationship was found in women or current and ex‐smokers. This is in contrast to our study which found no effect modification by sex.

PAF calculations take into account both the risks associated with an exposure and its prevalence in the population. In our study, the prevalence of BMI‐defined and WC‐defined obesity was similar; thus it is likely that a difference in risk estimates rather than a difference in exposure prevalence explains the difference in PAF for obesity defined using BMI and using WC. Importantly, we demonstrated first the presence of a subgroup of people who are not captured as having obesity by their BMI, but who have a large WC, who have an increased risk for mortality, and subsequently we showed that the estimated proportion of deaths attributable to obesity was higher when WC was taken into account in defining obesity compared to when focusing on BMI alone. Thus, our findings indicate that the estimated proportion of deaths (or more precisely, the excess risk of mortality) that could theoretically be avoided by targeting those with an obese WC, and reducing population WC from obese to non‐obese, would be greater than the proportion of deaths that would be avoided by targeting population BMI alone (the status quo of current obesity prevention and intervention strategies).

We note that a limitation of the PAF lies in its interpretation, insofar that it actually refers to the proportion of risk that could be eliminated if the risk in the exposed were to decrease to the same level as the risk in the unexposed, while all other risk factors and the absolute risk in the unexposed remain unchanged (21). The real magnitude of effect would therefore depend on the method used for reducing BMI and WC (e.g., lifestyle modification vs. surgery). Weight‐loss strategies, when effective in reducing obesity, are likely to also improve other risk factors. Moreover, calculations of PAF do not account for etiologic cases and may represent an underestimation of the true proportion of risk attributable to a given factor (22). Lastly, we acknowledge that the same strategies would be used to reduce BMI and WC, given that no intervention clearly targets one over the other. However, additional people (those with BMI^N/WC^O) would be included as targets for intervention when WC is taken into account in addition to BMI. Thus, our primary message is not in the absolute magnitudes of the PAFs presented, but rather in their relative differences when comparing the different obesity measures. Importantly, the pattern of differences between different obesity measures was consistent across univariate, multivariate, and sensitivity analyses, and across different subgroups of age and country of birth.

Interestingly, in our study, the combination of BMI and/or WC to define obesity was not beneficial in terms of PAFs and AIC compared to simply defining obesity using WC alone. This is likely because those with BMI^O/WC^N did not have an increased risk of mortality; thus combining these individuals with those who have an obese WC would not be expected to improve the association with mortality. Alternatively, it may simply be that WC alone is sufficient to capture the risks associated with obesity and that combining WC with BMI adds little value in this population. This strongly supports the need to include WC measures in population obesity monitoring and risk assessment.

One advantage of BMI over WC is the relative ease and smaller measurement error in measuring height and weight compared to measuring WC (23). The latter is dependent on numerous factors including the type of tape measure used and the landmarks used to determine positioning of the tape measure. The current consensus appears to be that there would be modest to no benefit in moving from BMI to WC to assess obesity (24, 25). This is largely a product of studies that have compared predictive abilities and strengths of association of different anthropometric measures, usually assessed using area under the receiver operating characteristic curve and differences in odds ratio/relative risk (1-4). As there is greater scope for measurement error with WC compared to BMI, the increase in predictive ability using WC needs to be sufficiently large to produce a worthwhile trade‐off. However, the predictive ability of a risk factor is not the only important element. Our study showed that the proportion of all‐cause and CVD‐related deaths in the population that could theoretically be reduced if obesity was reduced was much greater when WC was taken into account. It is also important to acknowledge that BMI and WC do not always identify the same high‐risk individuals. With greater increases in WC than BMI over time the likely consequence is that the WC criteria for obesity may be increasingly capturing a different group of the population to the BMI criteria (26-31). This may be particularly important in our aging population given the discrepancy between BMI and WC appears particularly in older age, where the prevalence of WC‐defined obesity is higher than in younger ages while the prevalence of BMI‐defined obesity was slightly lower. This is in line with previous findings that people tend to lose weight after the age of 65 but continue to gain WC (5). Thus, the debate around which anthropometric measure best captures obesity and its related burden of disease needs to look beyond its ability to predict outcome, but also consider which measure (or combination of measures) best identifies individuals who are at increased risk of outcomes such as mortality, who should be targeted for treatment and prevention efforts including increased physical activity and reduced energy intake while still consuming a healthy diet.

For many years, researchers have discussed at great lengths the limitations of BMI as a measure for adiposity. The suggestion to move away from BMI as the sole population measure of obesity is not novel, and national weight management guidelines in several countries now suggest that health professionals consider both BMI and WC (7-10). Yet, current population monitoring for obesity and identification of individuals at risk of health outcomes in clinical practice still rely heavily on BMI. It is uncertain why routine assessment of WC is not currently done. We hypothesize that it may be related to the aforementioned greater measurement error associated with WC measures, potentially greater time required to measure WC well, and difficulty in accurately identifying landmarks such as the hip bone and lower rib to ascertain where the tape measure should be placed, especially in people with higher levels of body fat. Further exploration of potential barriers to routine WC measurement is important for progress, as only then can we develop strategies to overcome them.

The strengths of this study include a large sample size, a long follow‐up period, and objectively measured anthropometry. One potential limitation is that the MCCS is an older cohort recruited between 1990 and 1994, and the study population was predominantly older (40‐69 years at baseline); thus the findings may not necessarily reflect (current) relationships between BMI and WC with mortality in younger populations. In this study, we defined obesity using cut‐points derived for Europid populations. While our study population appears to comprise only of Europid individuals, ethnicity in this study was determined using country of birth. Lastly, in our univariate analyses, a significant difference in PAFs was observed between obesity defined by BMI and obesity defined by WC or by BMI and/or WC. Although the pattern of the estimates remained similar, statistical significance was lost in the multivariate analyses. There were also a modest number of CVD deaths in this cohort; thus our analyses may have been underpowered to detect significant differences between obesity definitions.

In conclusion, current population monitoring for obesity misses those in the population who have an obese WC but a BMI that is not within the obese range. We have shown that by incorporating WC to identify those with high‐risk adiposity and who should be targeted for intervention, the potential exists to prevent more all‐cause and CVD mortality in the population. Better identification of those with high‐risk adiposity is important as it would allow for better prioritization of healthcare resources.

Acknowledgments

This study was made possible by the contribution of many people, including the original investigators and the diligent team who recruited the participants and completed follow‐up. Authors would like to express their gratitude to the many thousands of Melbourne residents who continued to participate in the study. The MCCS was funded by VicHealth, The Cancer Council Victoria, and the National Health and Medical Research Council. These funding sources had no input into study design or collection, analysis, and interpretation of the data.