Risk of Mortality According to Body Mass Index and Body Composition Among Postmenopausal Women

Abstract

Obesity, often defined as a body mass index (BMI; weight (kg)/height (m)²) of 30 or higher, has been associated with mortality, but age-related body composition changes can be masked by stable BMI. A subset of Women's Health Initiative participants (postmenopausal women aged 50–79 years) enrolled between 1993 and 1998 who had received dual-energy x-ray absorptiometry scans for estimation of total body fat (TBF) and lean body mass (LBM) (n = 10,525) were followed for 13.6 (standard deviation, 4.6) years to test associations between BMI, body composition, and incident mortality. Overall, BMI ≥35 was associated with increased mortality (adjusted hazard ratio (HR) = 1.45, 95% confidence interval (CI): 1.16, 1.82), while TBF and LBM were not. However, an interaction between age and body composition (P < 0.001) necessitated age stratification. Among women aged 50–59 years, higher %TBF increased risk of death (HR = 2.44, 95% CI: 1.38, 4.34) and higher %LBM decreased risk of death (HR = 0.41, 95% CI: 0.23, 0.74), despite broad-ranging BMIs (16.4–69.1). However, the relationships were reversed among women aged 70–79 years (P < 0.05). BMI did not adequately capture mortality risk in this sample of postmenopausal women. Our data suggest the clinical utility of evaluating body composition by age group to more robustly assess mortality risk among postmenopausal women.

Body mass index (BMI) is a common clinical and epidemiologic measure because it is a simple, low-cost, noninvasive proxy for obesity, chronic disease risk (1, 2), and mortality risk (3–5). The extent to which BMI predicts mortality has been debated (6, 7), partially because of its limited sensitivity (50%) in the diagnosis of excess adiposity (8) and the nonlinear relationship demonstrated between total body fat (TBF) and BMI with aging (5, 9). These findings have prompted scientists to reevaluate BMI in describing these relationships, stemming largely from evidence of adipose tissue contributions to metabolic disturbances and chronic disease (10–12).

In addition to TBF, skeletal muscle is an important contributor to metabolic homeostasis (13–15). Skeletal muscle loss increases with aging (16) and has important functional consequences (16–20) independent of age, ethnicity, health behaviors, and obesity (16). Neither adiposity nor lean body mass (LBM), nor changes in these measures over time and with advanced age, has been well characterized in relation to mortality; yet, increasingly, research suggests that loss of LBM and gain of TBF are deleterious to health, and these changes may be masked by weight stability (9, 21, 22). Thus, it may be important to employ more precise technology, such as dual-energy x-ray absorptiometry (DXA), to measure body composition and test the relationship between the body compartments and mortality. With the increased clinical availability of DXA, it is now feasible to complete more precise assessments and evaluate these relationships more robustly.

The Women's Health Initiative (WHI) Clinical Trials and Observational Study were designed to prospectively evaluate causes of morbidity and mortality in postmenopausal women. A subcohort underwent DXA scans at enrollment, which could be leveraged to assess the predictive value of body composition measures to evaluate mortality risk. Associations between LBM and mortality were of particular interest due to the paucity of research in this area. Our primary aim was to investigate associations between DXA measures of body composition (LBM and TBF) and incidence of all-cause mortality. We also sought to determine whether body composition and BMI were differentially associated with incident mortality and whether associations varied by age, based on prior work demonstrating age-specific BMI and body composition responses to risk-reducing physical activity (23, 24). We hypothesized that DXA-derived TBF would be positively associated with incident mortality and that LBM would be negatively associated with incident mortality. We further hypothesized that associations of BMI and body composition with incident mortality would vary by age.

METHODS

Study participants

Between 1993 and 1998, a total of 161,808 postmenopausal women aged 50–79 years were enrolled in the WHI Clinical Trials and Observational Study at 40 clinical centers across the United States (25, 26). The subcohort that received DXA scans comprised WHI Clinical Trials and Observational Study participants who consented to undergo DXA at one of the WHI clinical centers in Pittsburgh, Pennsylvania; Birmingham, Alabama; or Tucson-Phoenix, Arizona (27). Racial/ethnic diversity was maximized at these sites. Only women in this DXA subcohort (n = 11,020) were included in the present analyses. Women who were missing %TBF (n = 385) or follow-up (n = 28) data or died within the first 2 years of follow-up (n = 82) were excluded. The final sample size was 10,525, with an average follow-up time of 13.6 (standard deviation (SD), 4.6) years. The protocol and consent forms were approved by the institutional review board at each site, and all participants provided written informed consent. The study design and participant characteristics have been described in detail elsewhere (25, 26, 28).

Deaths

Information on vital status was collected for all WHI participants at least annually during scheduled telephone contacts. When participants could not be reached, relatives or caregivers were contacted to retrieve this information. Medical records and death certificates were used to verify deaths. The National Death Index was also used for confirmation if medical records and death certificates could not be obtained. Time to death was defined as the number of days between enrollment and death, censored at the time of the last contact for women still living (29).

Body composition assessment

Weight and height were measured in the clinic on a balance-beam scale to the nearest 0.1 kg and on a wall-mounted stadiometer to the nearest 0.1 cm, respectively. BMI was calculated as weight (kg)/height (m)². Waist circumference to the nearest 0.5 cm was measured at the level of the umbilicus over nonbinding undergarments. Body composition was determined by performing total body DXA scans (QDR2000, 2000+, or 4500W; Hologic Inc., Bedford, Massachusetts) at baseline. We calculated %TBF and %LBM as TBF (kg)/weight (kg) × 100 and LBM (kg)/weight (kg) × 100, respectively. DXA-derived fat and lean mass have been validated against the gold standard, magnetic resonance imaging, within a subset (n = 101) of this study population (WHI magnetic resonance imaging precision error <1%; for fat mass, Pearson's ρ = 0.99; for lean mass, Pearson's ρ = 0.94; P < 0.001) (30).

The WHI DXA quality assurance program included standard WHI protocols for positioning and analysis of DXA scans by technicians certified by Hologic, Inc. and the WHI Bone Density Coordinating Center; local daily and weekly phantom scans; circulating Hologic spine, hip, and block calibration phantoms scanned by each site and instrument; and machine and technician performance monitoring by means of random sampling, review of phantom scans, and review of scans with specific problems (31).

Covariates

Potential covariates were initially identified from published literature evaluating body composition and mortality. Study-specific questionnaires administered at baseline were used to ascertain baseline age, race/ethnicity, dietary energy consumption (kcal/day) and diet quality (Healthy Eating Index 2005 (32, 33) score) (34), physical activity (metabolic equivalent-hours/week) (35), physical function (36), smoking (pack-years), alcohol intake (converted to drinks/day), and self-reported prior use of hormone therapy.

Statistical analysis

Baseline characteristics were compared across quintiles of %TBF using analysis of variance (for continuous variables) and χ² tests (for categorical variables). Cox proportional hazards regression models were used to test differences in risk of incident all-cause mortality according to standard BMI categories (underweight (<18.5), normal weight (18.5–24.9), overweight (25.0–29.9), obese class I (30.0–34.9), obese class II (35.0–39.9), or obese class III (≥40.0)); quintiles of BMI, %TBF, and %LBM; and quintiles of absolute TBF (kg) and LBM (kg). The denominators for %TBF were not equal across quintiles because %TBF was rounded to the nearest tenth, thereby producing many tied values. Variables that were associated with both %TBF and incident mortality at P < 0.1 were included in the multivariate models as potential confounders: age (years; continuous), age at menopause (years; continuous), diet quality (Healthy Eating Index 2005 score (possible range, 0–100); continuous), physical activity (metabolic equivalent-hours/week; continuous), race/ethnicity (non-Hispanic white, black, Hispanic, or other/unknown), pack-years of smoking (0 (never smoker), <5, 5–<20, or ≥20), current alcohol intake (none, <1 drink/day, or ≥1 drink/day), hormone therapy status (never, former, or current use), and clinical trial arm. In the absolute TBF (kg) and LBM (kg) models, results were additionally adjusted for height and weight and mutually adjusted for TBF (quintiles) or LBM (quintiles). Mutual adjustment for lean mass and fat mass could not be performed in the %TBF and %LBM models because of strong correlation (Pearson's ρ = −0.998); absolute measures were less correlated (Pearson's ρ = 0.508).

Age interactions were evaluated because prior age interactions were detected in WHI body composition analyses (23, 24), and analyses were stratified by decade of age at baseline. The proportional hazards assumption in each age-stratified Cox model was assessed using Schoenfeld's test. Models for absolute TBF (kg) and absolute LBM (kg) were stratified by waist circumference (<88 cm or ≥88 cm) to explore whether the associations between TBF or LBM and incident mortality differed according to waist circumference. Quintile cutpoints for absolute TBF (kg) and LBM (kg) were recalculated to be waist circumference–specific to ensure a reasonable number of participants in each category. Likelihood ratio tests were conducted to measure 2-way interactions between each body size/composition measure and age group and 3-way interactions between absolute TBF or LBM, age, and waist circumference. We attempted to investigate differential associations by race/ethnicity but found insufficient power for these analyses. All statistical analyses were performed using Stata (version 13.1; StataCorp LP, College Station, Texas).

RESULTS

In the overall cohort, the mean age was 63.1 (SD, 7.4) years, and the mean BMI was in the overweight range, 28.2 (SD, 5.9). The majority of women (n = 8,428; 80.1%) demonstrated a high level of body fat according to age-appropriate normative data (>38%TBF) (37). The majority of the cohort was non-Hispanic white (77.1%), did not smoke (92.0%), and consumed less than 1 alcoholic beverage per day (91.6%). Approximately 35% of the cohort had never used postmenopausal hormone therapy (Table 1). When participants were stratified by quintile of %TBF (Table 2), mean BMI typically increased 2–3 units for each quintile increase in %TBF. Although %LBM decreased across increasing quintiles of %TBF, absolute LBM (kg) increased, reflecting the increasing total body mass. The lean mass:fat mass ratio also decreased across %TBF quintiles. Energy intake was higher with higher %TBF, and diet quality (Healthy Eating Index 2005) and physical activity were lower with higher %TBF. Current use of hormone therapy was lower among the higher %TBF groups. For the total cohort, BMI was positively correlated with TBF (kg), LBM (kg), and %TBF at baseline (Pearson's ρ = 0.90, 0.62, and 0.75, respectively); BMI was inversely associated with %LBM (Pearson's ρ = −0.74).

Table 1.

Baseline Characteristics of Postmenopausal Women Participating in the Women's Health Initiative (n = 10,525), 1993–1998^a

Characteristic	Mean (SD)	No.	%
Age at baseline, years	63.1 (7.4)
Age at menopause, years	47.5 (6.8)
Energy intake, kcal/day	1,669 (689)
Diet quality (HEI-2005 scoreb)	66.2 (11.0)
Physical activity, MET-hours/week	11.5 (13.9)
Height, m	1.62 (0.06)
Weight, kg	74.0 (16.4)
Body mass indexc	28.2 (5.9)
Total lean body mass, kg	37.8 (5.4)
Total fat body mass, kg	32.6 (11.6)
Total lean body mass, %	53.3 (7.0)
Total fat body mass, %	43.8 (7.3)
Lean mass:fat mass ratio	1.29 (0.48)
Race/ethnicity
Non-Hispanic white		8,119	77.1
Black or African-American		1,502	14.3
Hispanic		681	6.47
Other/unknown		223	2.12
Smoking, pack-years
0 (never smoker)		5,682	55.9
<5		1,367	13.5
5–<20		1,339	13.2
≥20		1,779	17.5
Alcohol intake, drinks/day
None		5,217	51.5
<1		4,061	40.1
≥1		849	8.38
Hormone use
Never use		3,371	35.1
Former use		2,360	24.5
Current use		3,884	40.4

Abbreviations: HEI-2005, Healthy Eating Index 2005; MET, metabolic equivalent; SD, standard deviation.

^a Missing data: for age at menopause, 1,103 women (10.5%); for energy intake, 398 women (3.8%); for diet quality, 398 women (3.8%); for physical activity, 1,190 women (11.3%); for height, 49 women (0.5%); for weight, 11 women (0.1%); for body mass index, 56 women (0.5%); for smoking, 358 women (3.4%); for alcohol intake, 398 women (3.8%); and for hormone use, 910 women (8.6%).

^b Range of possible scores, 0–100. In these participants, scores ranged from 22.7 to 91.1.

^c Weight (kg)/height (m)².

Table 2.

Baseline Characteristics of Postmenopausal Women Participating in the Women's Health Initiative (n = 10,525), by Quintile of Total Body Fat Percentage, 1993–1998^a

Characteristic	Quintile of Total Body Fat
	Quintile 1 (7.3%–38.1%) (n = 2,138)			Quintile 2 (38.2%–42.5%) (n = 2,111)			Quintile 3 (42.6%–46.0%) (n = 2,079)			Quintile 4 (46.1%–50.0%) (n = 2,095)			Quintile 5 (50.1%–64.9%) (n = 2,102)
	Mean (SD)	No.	%	Mean (SD)	No.	%	Mean (SD)	No.	%	Mean (SD)	No.	%	Mean (SD)	No.	%
Age at baseline, years	63.0 (7.7)			63.4 (7.4)			63.3 (7.4)			63.3 (7.3)			62.5 (7.2)
Age at menopause, years	48.2 (6.4)			47.8 (6.7)			47.5 (6.7)			47.1 (7.0)			46.7 (7.3)
Energy intake, kcal/day	1,577 (589)			1,617 (633)			1,673 (701)			1,689 (710)			1,791 (780)
Diet quality (HEI-2005 scoreb)	68.4 (10.7)			67.6 (10.9)			66.2 (10.8)			65.1 (10.7)			63.6 (11.3)
Physical activity, MET-hours/week	17.3 (17.3)			13.0 (14.1)			10.5 (11.9)			9.47 (12.1)			7.10 (10.3)
Height, m	1.62 (0.06)			1.62 (0.06)			1.62 (0.06)			1.61 (6.6)			1.61 (6.6)
Weight, kg	59.6 (10.1)			67.3 (10.9)			72.6 (10.9)			78.8 (12.4)			91.9 (16.0)
Body mass indexc	22.5 (2.9)			25.6 (3.3)			27.7 (3.6)			30.3 (4.1)			35.2 (5.5)
Total lean body mass, kg	36.6 (4.3)			37.0 (4.8)			37.6 (5.4)			38.1 (5.6)			39.7 (6.2)
Total fat body mass, kg	19.4 (4.2)			26.6 (3.8)			31.7 (4.7)			37.1 (5.7)			48.5 (9.6)
Total lean body mass, %	63.5 (4.3)			56.5 (1.3)			52.8 (1.0)			49.3 (1.2)			44.1 (2.6)
Total fat body mass, %	33.2 (4.4)			40.5 (1.3)			44.3 (1.0)			48.0 (1.2)			53.4 (2.7)
Lean mass:fat mass ratio	1.98 (0.57)			1.40 (0.07)			1.19 (0.05)			1.03 (0.05)			0.83 (0.09)
Race/ethnicity
Non-Hispanic white		1,793	83.9		1,670	79.1		1,614	77.6		1,567	74.8		1,475	70.2
Black or African-American		197	9.2		263	12.5		300	14.4		322	15.4		420	20.0
Hispanic		97	4.5		137	6.5		130	6.3		173	8.3		144	6.9
Other/unknown		51	2.4		41	1.9		35	1.7		33	1.6		63	3.0
Smoking, pack-years
0 (never smoker)		1,056	51.0		1,115	54.8		1,170	58.2		1,175	57.9		1,166	57.6
<5		298	14.4		267	13.1		257	12.8		262	12.9		283	14.0
5–<20		315	15.2		266	13.1		250	12.4		259	12.8		249	12.3
≥20		401	19.4		386	19.0		335	16.7		332	16.4		325	16.1
Alcohol intake, drinks/day
None		826	40.0		973	47.9		1,063	52.8		1,095	54.6		1,260	62.8
<1		981	47.5		856	42.1		805	40.0		765	38.1		654	32.6
≥1		260	12.6		203	10.0		146	7.3		146	7.3		94	4.7
Hormone use
Never use		581	29.9		620	32.2		681	35.7		723	37.5		766	40.0
Former use		433	22.3		465	24.1		427	22.4		523	27.2		512	26.8
Current use		927	47.8		843	43.7		799	41.9		680	35.3		635	33.2

Abbreviations: HEI-2005, Healthy Eating Index 2005; MET, metabolic equivalent; SD, standard deviation.

^a Missing data: for age at menopause, 1,103 women (10.5%); for energy intake, 398 women (3.8%); for diet quality, 398 women (3.8%); for physical activity, 1,190 women (11.3%); for height, 49 women (0.5%); for weight, 11 women (0.1%); for body mass index, 56 women (0.5%); for smoking, 358 women (3.4%); for alcohol intake, 398 women (3.8%); and for hormone use, 910 women (8.6%).

^b Range of possible scores, 0–100.

^c Weight (kg)/height (m)².

There were 1,762 deaths (16.7%) during follow-up. BMI-defined class II obesity (hazard ratio (HR) = 1.45, 95% confidence interval (CI): 1.16, 1.82) and class III obesity (HR = 1.84, 95% CI: 1.40, 2.43) were significantly associated with increased risk of death in the fully adjusted model, in comparison with normal-weight women (Table 3). Participants who were underweight, overweight, or obese class I demonstrated nonsignificantly increased mortality risk. The highest BMI quintile was significantly associated with mortality (HR = 1.26, 95% CI: 1.05, 1.51). DXA-derived %TBF and %LBM were not associated with mortality in the overall cohort. However, a significant interaction was detected between age and body composition (for both %LBM and %TBF, P for interaction < 0.001) in relation to mortality.

Table 3.

Association Between Body Size or Body Composition and All-Cause Mortality Among Postmenopausal Women in the Women's Health Initiative, 1993–2013

Body Size or Composition	BMIa Range			Mortality Incidenceb			Age-Adjusted HR	95% CI	Multivariate HRc	95% CI
	Minimum	Median	Maximum	No. of Deaths	Total No. of Women	%
BMI category
Underweight (<18.5)	14.4	17.9	18.5	18	86	20.9	1.26	0.79, 2.02	1.09	0.62, 1.89
Normal weight (18.5–<25.0)	18.5	22.9	25.0	520	3,334	15.6	1.00	Referent	1.00	Referent
Overweight (25.0–<30.0)	25.0	27.3	30.0	615	3,696	16.6	1.06	0.94, 1.19	1.03	0.89, 1.18
Obese I (30.0–<35.0)	30.0	32.0	35.0	353	2,061	17.1	1.21	1.05, 1.38	1.12	0.95, 1.32
Obese II (35.0–<40.0)	35.0	37.0	40.0	154	841	18.3	1.61	1.34, 1.92	1.45	1.16, 1.82
Obese III (≥40.0)	40.0	42.8	69.1	94	451	20.8	2.13	1.71, 2.65	1.84	1.40, 2.43
BMI quintile
Q1 (≤23.3)	14.4	21.8	23.3	355	2,094	17.0	1.00	Referent	1.00	Referent
Q2 (23.4–25.8)	23.4	24.7	25.9	315	2,094	15.0	0.90	0.78, 1.05	0.92	0.77, 1.10
Q3 (25.9–28.7)	25.9	27.2	28.7	345	2,095	16.5	0.98	0.84, 1.14	0.87	0.73, 1.04
Q4 (28.8–32.6)	28.7	30.4	32.6	357	2,093	17.1	1.08	0.93, 1.25	1.02	0.86, 1.22
Q5 (≥32.7)	32.6	36.1	69.1	382	2,093	18.3	1.44	1.24, 1.66	1.26	1.05, 1.51
Total body fat, %
Q1 (≤38.1)	14.4	22.3	66.0	362	2,138	16.9	1.00	Referent	1.00	Referent
Q2 (38.2–42.5)	16.4	25.1	67.9	357	2,111	16.9	1.00	0.86, 1.16	0.96	0.81, 1.14
Q3 (42.6–46.0)	18.6	27.3	65.5	337	2,079	16.2	0.95	0.82, 1.10	0.89	0.74, 1.06
Q4 (46.1–50.0)	20.1	29.9	65.9	345	2,095	16.5	1.01	0.87, 1.17	0.95	0.79, 1.13
Q5 (≥50.1)	18.9	34.6	69.1	361	2,102	17.2	1.16	1.00, 1.35	1.00	0.83, 1.20
Lean body mass, %
Q1 (≤47.2)	18.9	34.5	69.1	351	2,105	16.7	1.00	Referent	1.00	Referent
Q2 (47.3–51.1)	20.1	29.8	65.9	352	2,105	16.7	0.91	0.78, 1.05	1.02	0.85, 1.21
Q3 (51.2–54.5)	18.6	27.3	65.5	340	2,105	16.2	0.83	0.72, 0.96	0.89	0.75, 1.07
Q4 (54.6–58.7)	16.2	25.1	67.9	360	2,105	17.1	0.88	0.76, 1.02	0.98	0.82, 1.17
Q5 (≥58.8)	14.4	22.3	66.0	359	2,105	17.1	0.87	0.75, 1.01	1.01	0.84, 1.21

Abbreviations: BMI, body mass index; CI, confidence interval; HR, hazard ratio; Q, quintile.

^a Weight (kg)/height (m)².

^b Includes all women, irrespective of missing data for covariates.

^c Adjusted for age at baseline (years; continuous), age at menopause (years; continuous), physical activity (metabolic equivalent-hours/week; continuous), diet quality (Healthy Eating Index 2005 score; continuous), race/ethnicity (non-Hispanic white, black, Hispanic, or other/unknown), pack-years of smoking (0 (never smoker), <5, 5–<20, or ≥20), alcohol intake (none, <1 drink/day, or ≥1 drink/day), hormone use (never, former, or current use), and clinical trial arm.

Upon stratification by decade of life (age 50–59, 60–69, or 70–79 years), the association between BMI and mortality was limited to obese class II and obese class III for women aged <70 years only (Table 4). In contrast, striking differential associations by age category were observed for %TBF, %LBM, and mortality. Specifically, for postmenopausal women aged 50–59 years, there was a significantly increased risk of mortality among those in %TBF quintiles 2–5 compared with quintile 1, with the highest risk being seen in quintile 5 (HR = 2.44, 95% CI: 1.38, 4.34). Additionally, the highest quintile of %LBM demonstrated a significant risk reduction in women aged 50–59 years as compared with quintile 1 (HR = 0.41, 95% CI: 0.23, 0.74). In contrast, the relationship between body composition and mortality was reversed among older women (ages 70–79 years): High %TBF was protective against death (quintile 5 vs. quintile 1: HR = 0.74, 95% CI: 0.56, 0.98), whereas higher %LBM was associated with increased mortality risk (quintile 5 vs. quintile 1: HR = 1.37, 95% CI: 1.03, 1.82). Protection from death was observed for the middle quintiles of %TBF (HR = 0.71, 95% CI: 0.53, 0.95) and %LBM (HR = 0.67, 95% CI: 0.51, 0.90) in women aged 60–69 years, but there was no association for quintile 5 in this age group. Adjustment for height did not significantly influence the %TBF and %LBM models (data not shown).

Table 4.

Association Between Body Size or Body Composition and All-Cause Mortality Among Postmenopausal Women in the Women's Health Initiative, by Age Group, 1993–2013

Body Size or Composition	Age Group, years
	<60					60–69					≥70
	No. of Deaths	Total No. of Women	%	HRa	95% CI	No. of Deaths	Total No. of Women	%	HRa	95% CI	No. of Deaths	Total No. of Women	%	HRa	95% CI
BMIb category
Underweight (<18.5)	0	15	0.00	N/A	N/A	6	27	22.2	1.80	0.79, 4.12	7	23	30.4	0.79	0.37, 1.69
Normal weight (18.5–24.9)	36	840	4.29	1.00	Referent	143	1,061	13.5	1.00	Referent	199	630	31.6	1.00	Referent
Overweight (25.0–29.9)	47	844	5.57	1.18	0.76, 1.84	165	1,177	14.0	0.98	0.78, 1.23	225	644	34.9	1.01	0.83, 1.23
Obese I (30.0–34.9)	35	505	6.93	1.46	0.90, 2.36	109	655	16.6	1.17	0.90, 1.52	112	334	33.5	1.02	0.80, 1.30
Obese II (35.0–39.9)	26	248	10.5	2.26	1.33, 3.83	51	267	19.1	1.40	1.00, 1.97	32	82	39.0	1.21	0.82, 1.78
Obese III (≥40.0)	20	151	13.5	2.46	1.36, 4.45	28	125	22.4	1.78	1.16, 2.71	16	39	41.0	1.41	0.83, 2.38
BMI quintile
Q1 (≤23.3)	24	520	4.62	1.00	Referent	102	673	15.2	1.00	Referent	136	410	33.2	1.00	Referent
Q2 (23.4–25.8)	26	511	5.09	1.03	0.59, 1.81	82	652	12.6	0.80	0.60, 1.07	126	384	32.8	1.01	0.79, 1.29
Q3 (25.9–28.7)	23	459	5.01	0.96	0.54, 1.73	87	673	12.9	0.79	0.59, 1.06	119	362	32.9	0.90	0.70, 1.16
Q4 (28.8–32.6)	34	512	6.64	1.24	0.72, 2.11	108	658	16.4	1.03	0.78, 1.35	119	352	33.8	0.99	0.76, 1.27
Q5 (≥32.7)	57	601	9.48	1.68	1.01, 2.80	123	656	18.8	1.21	0.91, 1.60	91	244	37.3	1.14	0.86, 1.50
Total body fat, %
Q1 (≤38.1)	18	570	3.16	1.00	Referent	108	654	16.5	1.00	Referent	133	376	35.4	1.00	Referent
Q2 (38.2–42.5)	38	524	7.25	2.24	1.27, 3.94	100	676	14.8	0.80	0.60, 1.05	125	379	33.0	0.92	0.72, 1.18
Q3 (42.6–46.0)	31	515	6.02	1.80	1.00, 3.25	87	655	13.3	0.71	0.53, 0.95	126	366	34.4	0.89	0.69, 1.14
Q4 (46.1–50.0)	34	484	7.02	2.01	1.12, 3.60	87	675	12.9	0.72	0.54, 0.97	126	356	35.4	0.98	0.77, 1.26
Q5 (≥50.1)	44	522	8.43	2.44	1.38, 4.34	122	671	18.2	0.99	0.75, 1.30	85	287	29.6	0.74	0.56, 0.98
Lean body mass, %
Q1 (≤47.2)	44	536	8.21	1.00	Referent	117	666	17.6	1.00	Referent	81	282	28.7	1.00	Referent
Q2 (47.3–51.1)	37	491	7.54	0.92	0.59, 1.42	98	685	14.3	0.83	0.63, 1.08	125	350	35.7	1.38	1.04, 1.83
Q3 (51.2–54.5)	31	522	5.94	0.75	0.47, 1.20	82	668	12.3	0.67	0.51, 0.90	131	373	35.1	1.27	0.96, 1.69
Q4 (54.6–58.7)	36	517	6.96	0.90	0.57, 1.42	100	669	15.0	0.83	0.64, 1.10	127	383	33.2	1.28	0.96, 1.71
Q5 (≥58.8)	17	549	3.10	0.41	0.23, 0.74	107	643	16.6	1.03	0.78, 1.37	131	376	34.8	1.37	1.03, 1.82
Total body fat, kgc
Q1 (1.5–22.9)	18	515	3.50	1.00	Referent	94	625	15.0	1.00	Referent	138	429	32.2	1.00	Referent
Q2 (23.0–28.3)	31	508	6.10	1.50	0.82, 2.73	100	687	14.6	0.87	0.65, 1.17	145	397	36.5	1.06	0.82, 1.37
Q3 (28.4–33.7)	27	460	5.87	1.32	0.69, 2.52	86	673	12.8	0.67	0.48, 0.93	118	362	32.6	0.89	0.66, 1.19
Q4 (33.8–41.4)	33	513	6.43	1.28	0.64, 2.54	107	687	15.6	0.78	0.54, 1.11	107	327	32.7	0.89	0.62, 1.27
Q5 (41.5–83.6)	55	608	9.05	1.34	0.56, 3.23	116	644	18.0	0.77	0.47, 1.25	83	237	35.0	0.99	0.61, 1.60
Lean body mass, kgc
Q1 (8.7–33.2)	30	371	8.09	1.00	Referent	103	661	15.6	1.00	Referent	163	509	32.0	1.00	Referent
Q2 (33.3–35.8)	25	462	5.41	0.53	0.31, 0.93	81	695	11.7	0.72	0.54, 0.98	130	400	32.5	1.15	0.90, 1.47
Q3 (35.9–38.5)	20	501	3.99	0.37	0.20, 0.67	84	662	12.7	0.81	0.59, 1.11	119	360	33.1	1.13	0.86, 1.47
Q4 (38.6–41.9)	34	574	5.92	0.39	0.22, 0.70	108	658	16.4	1.11	0.79, 1.55	100	286	35.0	1.34	0.99, 1.83
Q5 (42.0–66.5)	55	696	7.90	0.37	0.19, 0.72	127	640	19.8	1.49	0.99, 2.24	79	197	40.1	1.97	1.33, 2.94

Abbreviations: BMI, body mass index; CI, confidence interval; HR, hazard ratio; N/A, not applicable; Q, quintile.

^a Adjusted for age at baseline (years; continuous), age at menopause (years; continuous), physical activity (metabolic equivalent-hours/week; continuous), diet quality (Healthy Eating Index 2005 score; continuous), race/ethnicity (non-Hispanic white, black, Hispanic, or other/unknown), pack-years of smoking (0 (never smoker), <5, 5–<20, or ≥20), alcohol intake (none, <1 drink/day, or ≥1 drink/day), hormone use (never, former, or current use), and clinical trial arm.

^b Weight (kg)/height (m)².

^c Models for total body fat (kg) and lean body mass (kg) included additional adjustment for height (m; continuous) and weight (kg; continuous). Mutual adjustment for total body fat (kg; quintiles) and lean body mass (kg; quintiles) in these models did not significantly affect results (data not shown).

The relationship between absolute TBF (kg) and mortality for women aged 50–59 years and 60–69 years was similar to the relationship between %TBF and mortality for those age groups, although the associations were generally attenuated and nonsignificant. Among women aged ≥70 years, there was no relationship between absolute TBF and mortality, unlike the case for %TBF. The protective association of LBM among women aged 50–59 years was demonstrated across all quintiles of LBM (kg), instead of confined to quintile 5, as in %LBM models. The risk of mortality was reduced by 47%–63% across all LBM quintiles (quintile 2: HR = 0.53, 95% CI: 0.31, 0.93; quintile 5: HR = 0.37, 95% CI: 0.19, 0.72) as compared with quintile 1 in women aged 50–59 years. The associations between absolute LBM and mortality were similar to those for %LBM and mortality among women aged ≥60 years. Significantly increased risk of mortality was demonstrated for the highest level of LBM (kg) among women aged ≥70 years (HR = 1.97, 95% CI: 1.33, 2.94). No significant change in the results was observed when the absolute TBF and LBM models included mutual adjustment for the other factor, except for a slight strengthening of the LBM-mortality relationship in women aged 70–79 years (data not shown).

In separate analyses, when the results were stratified by waist circumference, the protective association of higher LBM (kg) in younger women (age <60 years) was seen only among those with low waist circumference (<88 cm) and not those with larger waists (≥88 cm; Table 5). Furthermore, the increased mortality risk among older women (ages 60–69 and ≥70 years) with higher LBM (kg) was seen only among those with a high waist circumference (likelihood ratio test for 3-way interaction between LBM, waist circumference, and age: P = 0.002). Stratification by waist circumference did not substantially change the associations between death and absolute TBF overall (likelihood ratio test for 3-way interaction between TBF, waist circumference, and age: P = 0.226). However, among younger women (age <60 years), TBF was positively associated with mortality in the low waist circumference group but not in the high waist circumference group.

Table 5.

Association Between Body Size or Body Composition and All-Cause Mortality Among Postmenopausal Women in the Women's Health Initiative, by Age and Waist Circumference, 1993–2013

Body Size or Composition	Age Group, years
	<60		60–69		≥70
	HRa	95% CI	HRa	95% CI	HRa	95% CI
Waist circumference <88 cm	n = 2,175		n = 2,578		n = 1,466
Total body fat, kg
Q1 (2.5–20.9)	1.00	Referent	1.00	Referent	1.00	Referent
Q2 (21.0–24.1)	2.73	1.19, 6.27	0.95	0.65, 1.37	0.95	0.69, 1.30
Q3 (24.2–27.6)	1.77	0.74, 4.21	0.84	0.58, 1.23	0.87	0.64, 1.19
Q4 (27.7–31.6)	1.74	0.72, 4.23	0.63	0.42, 0.94	0.96	0.70, 1.30
Q5 (31.7–59.7)	2.41	1.04, 5.57	0.72	0.48, 1.08	0.75	0.52, 1.07
Lean body mass, kg
Q1 (23.1–32.0)	1.00	Referent	1.00	Referent	1.00	Referent
Q2 (32.1–34.2)	0.36	0.17, 0.77	0.72	0.50, 1.05	1.10	0.82, 1.48
Q3 (34.3–36.2)	0.52	0.27, 1.01	0.59	0.40, 0.87	1.07	0.79, 1.46
Q4 (36.3–38.7)	0.26	0.13, 0.56	0.71	0.49, 1.04	0.97	0.69, 1.35
Q5 (38.8–55.2)	0.31	0.16, 0.61	0.81	0.55, 1.19	1.33	0.94, 1.89
Waist circumference ≥88 cm	n = 1,486		n = 1,901		n = 886
Total body fat, kg
Q1 (1.5–33.6)	1.00	Referent	1.00	Referent	1.00	Referent
Q2 (33.7–38.1)	0.51	0.23, 1.15	0.79	0.53, 1.18	1.01	0.70, 1.46
Q3 (38.2–42.7)	0.70	0.33, 1.50	0.92	0.62, 1.37	1.12	0.76, 1.65
Q4 (42.8–49.9)	0.84	0.41, 1.72	0.86	0.58, 1.28	0.82	0.54, 1.23
Q5 (50.0–83.6)	1.08	0.55, 2.11	1.00	0.68, 1.49	0.96	0.58, 1.58
Lean body mass, kg
Q1 (8.7–36.5)	1.00	Referent	1.00	Referent	1.00	Referent
Q2 (36.6–39.4)	1.49	0.60, 3.68	0.97	0.64, 1.47	0.98	0.67, 1.45
Q3 (39.5–41.9)	1.16	0.46, 2.89	1.03	0.67, 1.57	1.12	0.76, 1.64
Q4 (42.0–45.3)	1.08	0.43, 2.68	1.10	0.73, 1.67	1.31	0.85, 2.01
Q5 (45.4–66.5)	1.58	0.68, 3.70	1.67	1.11, 2.52	1.89	1.21, 2.97

Abbreviations: CI, confidence interval; HR, hazard ratio; Q, quintile.

^a Adjusted for age at baseline (years; continuous), age at menopause (years; continuous), physical activity (metabolic equivalent-hours/week; continuous), diet quality (Healthy Eating Index 2005 score; continuous), race/ethnicity (non-Hispanic white, black, Hispanic, other/unknown), pack-years of smoking (0 (never smoker), <5, 5–<20, or ≥20), alcohol intake (none, <1 drink/day, or ≥1 drink/day), hormone use (never, former, or current use), clinical trial arm, height (m; continuous), and weight (kg; continuous).

Stratification by race/ethnicity showed no significant interactions. In addition, a sensitivity analysis in which the models were restricted to never smokers demonstrated very similar results, although with less precision due to smaller sample sizes (data not shown).

DISCUSSION

Principal findings

Only BMI was significantly associated with all-cause mortality incidence in the full cohort; however, the associations of age-by-body composition interactions with incident mortality substantially altered the results and subsequent interpretation of these relationships. Upon age stratification, both BMI and body composition were associated with mortality in postmenopausal women. Increased risk of mortality was limited to BMI-defined obese class II and class III individuals, regardless of age, although the association was nonsignificant in the oldest group. However, what is typically considered adverse body composition (high %TBF and/or low %LBM) was associated with higher mortality only among younger postmenopausal women. In fact, there was an opposing association in older women. This relationship held for absolute LBM (kg) but not for absolute TBF (kg), where the pattern of association was similar to that in the %TBF models but attenuated. The protective association of LBM (kg) among women aged 50–59 years was striking, with up to a 63% reduction in risk of incident mortality among those in the highest quintile of LBM (kg).

To our knowledge, this is the largest study to have evaluated the relationship between DXA-derived body composition and incident mortality (n = 10,525) and to have compared TBF and LBM with BMI in regards to their associations with mortality. According to previously reported normative standards for body fat, >38% TBF for middle-aged women and >35% TBF for elderly women would be considered a high level of adiposity (37). According to these standards, the majority of the community-dwelling postmenopausal women in the study (80.1%; approximately quintiles 2–5) would be considered to have excess adiposity, although %TBF quintiles 2–5 encompassed nearly the full spectrum of BMIs represented in the cohort (16.4–69.1). Such excess body fat conferred more than double the risk of mortality in younger postmenopausal women. Thus, many postmenopausal women with normal BMI but excess TBF upon DXA assessment of body composition may be at increased risk of mortality, but such women may not receive appropriate lifestyle and therapeutic recommendations to address their adiposity-associated mortality risk. Additionally, since weight loss alone induces loss of both LBM and TBF, lifestyle recommendations should include the use of LBM-preserving strategies, such as physical activity, while losing weight (38, 39). The approximately 60% reduction in mortality risk among younger postmenopausal women for those in the highest LBM quintile (>58.8% or 42.0–66.5 kg), independent of TBF, lends strong support for LBM preservation early in menopause.

Among women aged ≥60 years, a reduced risk of mortality with lower LBM and higher TBF was suggested, limited to middle quintiles of body composition in which the median BMI was in the overweight range. These results imply a survival benefit of moderately elevated TBF for women aged 60–69 years, potentially due, in part, to the survival bias imparted by higher risk in the 50- to 59-year age group. In women aged 70–79 years, the TBF and LBM associations observed in the youngest group were reversed, with the highest level of TBF being associated with reduced risk of mortality and quintiles 2–5 of LBM being associated with increased risk (although the estimates for some quintiles were not significant). These results strongly support the notion of clinically evaluating body composition and BMI measurements in conjunction with age when assessing mortality risks among postmenopausal women (9). Additional energy stores may be important for women over age 60 years but not for younger women (3, 4).

One of our primary interests in this analysis was to examine the association between LBM and incident mortality, particularly because of the inconsistent outcomes and small number of studies in this area. Bioelectrical impedance and anthropometric estimates have favored survival with higher LBM (40–42), but imaging studies (DXA, computed tomography) are rarer, and results have been mixed. In the youngest age group (50–59 years), a survival benefit was demonstrated only for the highest category of %LBM but across quintiles 2–5 of absolute LBM (kg), independent of body size and TBF. The latter finding aligns with a study which found that DXA-derived sarcopenia, the lowest level of LBM, was necessary for increased mortality (43). Furthermore, in the Netherlands Longitudinal Aging Study, DXA-derived LBM was shown to be nonsignificantly positively associated with mortality among women (n = 235) aged 74.0 (SD, 6.3) years (44), similar to the older women in the WHI cohort. However, our results for the oldest women were not consistent with those for the similar-age women in the Health, Aging and Body Composition Study cohort (n = 1,168 women aged 70–79 years at baseline), for which no significant associations between mortality and LBM by DXA or computed tomography scan were found (45).

We did not have a direct measure of visceral fat, but we were able to explore the role of central adiposity with waist circumference. Importantly, persons with lower age and waist circumference were at increased risk of death with higher TBF. Although prior literature suggests that central adiposity, particularly visceral fat, is associated with greater adverse metabolic and inflammatory outcomes, which may contribute to chronic diseases and death (7, 46), our data suggest that overall body composition should be considered even in women with low waist circumference.

Strengths and limitations

Although being overweight or obese, defined by BMI, has been previously associated with mortality, our results were well aligned with a recent meta-analysis that suggested that being obese class I may not be associated with increased mortality and that being overweight may, in fact, be protective across a broad age range, especially for women aged ≥65 years (3, 4, 47–50). Notably, few researchers have stratified their results by age or have examined body composition in comparison with BMI to assess mortality risk (4), as we have done here; thus, differential associations by age may have been missed. Body composition contributes to mortality risk in women yet is not routinely assessed clinically. Further, shifts in body composition with aging can contribute to mortality risk and would likely be masked by weight stability (9, 21, 22). These results imply that anthropometric measures, including BMI, may not be the best proxy for risk assessment in older populations. Other strengths of our study include adjustment for smoking history and the exclusion of deaths occurring within 2 years of study entry to minimize potential reverse causation (4, 51).

The increased risk of death among elderly women with higher LBM is counterintuitive and deserves further exploration. A study of the quality of the LBM or skeletal muscle, in addition to the quantity, across a broad range of ages would be an important contribution to the literature. Neither strength measures nor muscle biopsies for all of the women who received DXA scans were available in WHI.

Additionally, although the relative homogeneity of the WHI population in terms of age, sex, and menopausal status provided us with greater statistical power to observe associations between body composition and mortality, it limits the generalizability of these findings, excluding men and younger women. We also were limited by sample size in evaluating differences by race/ethnicity. Evaluation of heterogeneous fat distribution across racial/ethnic groups with respect to risk of death (52, 53) and chronic conditions has begun (54–56), but data on many cohorts are limited to anthropometric measurements, and investigators do not have adequate sample sizes to explore multiple racial/ethnic backgrounds simultaneously. Lastly, DXA scans at the time of WHI baseline measurements did not include estimates of subcutaneous, visceral, android, gynoid, or intramuscular fat, and the additional stratification by waist circumference resulted in wide confidence intervals due to reduced sample size per stratum. Further enriching cohorts for racial/ethnic diversity and fully exploring the ramifications of total and regional body composition in specific causes of death across groups would be an important contribution to the field.

Conclusion

Our results suggest that body composition is an important factor for evaluating mortality risk in postmenopausal women, especially women aged 50–59 years, and higher mortality risk may not be captured by BMI measurements alone. Further, appropriate recommendations for lifestyle changes cannot be provided to women with low-risk BMI but high-risk body composition if body composition is not measured. Additionally, the associations between body composition and death appear to be significantly modified by age, such that age should be routinely considered when evaluating mortality risk and making lifestyle recommendations in postmenopausal women.