Associations of Skeletal Muscle Mass and Fat Mass With Incident Cardiovascular Disease and All-Cause Mortality: A Prospective Cohort Study of UK Biobank Participants.

Background There is debate whether body mass index is a good predictor of health outcomes because different tissues, namely skeletal muscle mass (SMM) and fat mass (FM), may be differentially associated with risk. We investigated the association of appendicular SMM (aSMM) and FM with fatal and nonfatal cardiovascular disease (CVD) and all-cause mortality. We compared their prognostic value to that of body mass index. Methods and Results We studied 356 590 UK Biobank participants aged 40 to 69 years with bioimpedance analysis data for whole-body FM and predicted limb muscle mass (to calculate aSMM). Associations between aSMM and FM with CVD and all-cause mortality were examined using multivariable Cox proportional hazards models. Over 3 749 501 person-years of follow-up, there were 27 784 CVD events and 15 844 all-cause deaths. In men, aSMM was positively associated with CVD incidence (hazard ratio [HR] per 1 SD 1.07; 95% CI, 1.06-1.09) and there was a curvilinear association in women. There were stronger positive associations between FM and CVD with HRs per SD of 1.20 (95% CI, 1.19-1.22) and 1.25 (95% CI, 1.23-1.27) in men and women respectively. Within FM tertiles, the associations between aSMM and CVD risk largely persisted. There were J-shaped associations between aSMM and FM with all-cause mortality in both sexes. Body mass index was modestly better at discriminating CVD risk. Conclusions FM showed a strong positive association with CVD risk. The relationship of aSMM with CVD risk differed between sexes, and potential mechanisms need further investigation. Body fat and SMM bioimpedance measurements were not superior to body mass index in predicting population-level CVD incidence or all-cause mortality.

appendicular skeletal muscle mass

bioimpedance analysis

fat mass

skeletal muscle mass

Clinical Perspective

What Is New?

Reports of the relationship between muscle mass and cardiovascular disease (CVD) are inconsistent and have rarely been considered in the context of adiposity; analysis of body composition measured by bioimpedance in this large cohort of UK adults showed that fat mass showed strong positive associations with CVD events.

Appendicular skeletal muscle mass had a curvilinear association with CVD events in women and a positive association in men; the associations of appendicular skeletal muscle mass and fat mass with all‐cause mortality followed a J‐shape in both men and women.

Measurements of body fat and skeletal muscle mass were not superior to body mass index in predicting CVD events or mortality.

What Are the Clinical Implications?

Body mass index has been criticized as an inaccurate measure of health risks, but at a population level, more specific measurements of body composition, namely appendicular skeletal muscle mass and fat mass, were generally not more predictive of CVD events or mortality.

Although body mass index may be the simplest measurement to assess health risk, which is important from a public health perspective, some of this risk may not be attributable solely to adiposity, particularly if the association observed with appendicular skeletal muscle mass in men is confirmed.

Further research is needed to better understand the biological mechanisms and impact of different body tissue compartment on health outcomes.

The increasing prevalence of obesity is a significant public health concern because it is a known risk factor for several noncommunicable diseases, , , , , estimated to account for 56 million deaths globally in 2017. Evidence from prospective cohort studies , , , and meta‐analyses of such studies , , has repeatedly shown a J‐ or U‐shaped relationship between body mass index (BMI), cardiovascular disease (CVD), and all‐cause mortality, even after efforts to account for confounding and reverse causality. , A potential explanation for this is that BMI does not distinguish between fat mass (FM) and skeletal muscle mass (SMM), , yet their contribution to the pathogenesis of disease is likely to be different.

A systematic literature review of the associations between body composition and CVD or mortality (Data S1, Tables S1 and S2) showed that the majority of studies found null or inverse associations with SMM, although a minority of studies reported positive or curvilinear associations. More studies have investigated the relationship of FM with these outcomes, with the majority of them reporting positive associations. Very few studies have investigated the combined impact of both types of tissues, yet weight change is associated with changes in both these body tissue compartments.

Dual energy X‐ray absorptiometry (DEXA) or magnetic resonance imaging techniques are considered to be reference methods for the measurement of body composition because of their precision and reliability. However, these are often not feasible for large studies given they are expensive and not easily portable. Bioimpedance analysis (BIA) is a noninvasive and practical method to assess FM and SMM in clinical practice and at scale in population‐based studies. , , The UK Biobank uses a bioimpedance analyzer previously validated against DEXA in a mixed population of children and adults, and body composition estimates were found to be more accurate than those obtained from previous BIA estimates. A recent validation study comparing BIA to DEXA in a subsample of the UK Biobank participants showed BIA to be a valid method for the assessment of appendicular skeletal muscle mass (aSMM) and FM.

In this study we aimed to use BIA‐derived aSMM and FM measurements to look at their associations with incident CVD and all‐cause mortality in the UK Biobank population. Furthermore, to investigate the prognostic value of these BIA‐derived measurements in comparison to more traditional measures such as BMI, grip strength, and waist circumference.

Methods

Data Availability Statement

Researchers can apply to use the UK Biobank resource and access the data used. No additional data are available.

Study Design and Participants

The UK Biobank recruited 502 664 participants aged 40 to 69 years between 2006 and 2010 (response rate 5.5%) via mailed invitations to the general public living within 25 miles of one of the 22 assessment centers in England, Scotland, and Wales. , At the baseline assessment clinic, participants completed a touch‐screen questionnaire and computer‐assisted interview, had physical measurements taken, and biological samples collected. , UK Biobank received ethical approval from the North West Multi‐centre Research Ethics Committee (REC reference: 11/NW/03820). All participants gave written informed consent before enrolment in the study, which was conducted in accordance with the principles of the Declaration of Helsinki.

Participants were excluded from analyses if they had prior CVD (defined later) or diseases that may affect body composition including fractures in the past year, respiratory diseases, musculoskeletal conditions, and some infectious diseases (n=116 679; Figure S1). Participants were also excluded if they had missing data for the exposures (n=6751) or were not of a White race (n=22 644), because BIA estimates are derived from algorithms in White populations, which represent ≈95% of the UK Biobank sample. , ,

Measurement of Exposures

Measures of body weight and body composition (muscle mass and fat mass) were derived from BIA in bare‐footed participants wearing light clothing using a Tanita BC418MA single frequency segmental body‐composition analyzer (Tanita, Tokyo, Japan) at the baseline assessment center visit. The aSMM (kg) was calculated as the sum of the predicted muscle mass from the 4 limbs. Whole body FM (kg) was also obtained from BIA. Standing height was measured using a Seca 202 scale (Seca, Hamburg, Germany). BMI was calculated by dividing weight (kg) by the squared height in meters.

Waist circumference (cm) was measured at the umbilicus using a tape measure. The mean grip strength (kg) of the left and right hands was taken once using a Jamar J00105 hydraulic hand dynamometer.

Ascertainment of Outcomes

Participants were followed via linkage to National Health Service hospital in‐patient data from hospital episode statistics in England, the Scottish Morbidity Records, and the Patient Episode Database for Wales. Patients were identified if they died of any cause or developed incident CVD, defined using International Statistical Classification of Diseases, Tenth Revision (ICD‐10) categories: coronary heart disease (I21–I24, 125.6, I42, I43, K49, K50, K75, K40–K46), congestive heart failure or cardiomyopathy (I50, I50.1, 150.9, I11.0, I13.0, I13.2), and total stroke (I60–I64). Follow‐up was available until June 30, 2020, October 31, 2016, and February 29, 2016 for England, Scotland, and Wales respectively; and until July 31, 2020 for all‐cause mortality for all regions.

Statistical Analysis

Association of Skeletal Muscle Mass and Fat Mass With CVD and Mortality

First, age‐adjusted partial correlation coefficients between aSMM, FM, and height were calculated to examine the relationships between the body composition measurements and overall body size. As aSMM is highly correlated with FM and height, aSMM was regressed on height and FM and the residuals from this model were divided into sex‐specific quintiles for the main analysis.

Multivariable Cox regression analyses with age as the underlying timescale were used to estimate hazard ratios (HRs) and 95% CIs for the associations of sex‐specific fifths of aSMM and fifths of FM as well as per 1 SD with incident CVD and all‐cause mortality. All analyses were sequentially adjusted for height and height2 (continuous), Townsend index of deprivation (quintiles), level of education (none, vocational qualifications, any degree, higher degree, other), smoking (never, previous, current), alcohol intake (none, <1 unit/week, 1–14 units/week, 14+ units/week), physical activity derived from metabolic equivalent of task scores (low, moderate, high), and dietary factors (oily fish intake, saturated fat intake, fruit and vegetable intake [none, low, medium, high intake]), prior medical history (diabetes mellitus, cancer history >5 years ago, and menopausal status in women [binary for each]). We created a category for missing values for each of these covariates. Additionally, FM and aSMM were mutually adjusted for each other to assess the independent effects of each type of tissue. See Table S3 for the details on the derivation of these covariates.

The HRs and 95% CIs were computed using group‐specific variances ; these reflect the uncertainty in the estimate of risk in each group (including the reference group), thereby allowing comparisons between any 2 quintiles independently of the reference group. Restricted cubic splines with 5 knots were also computed to visually explore nonlinear associations for continuous exposures, and departures from linearity were tested via the likelihood ratio statistic test used to evaluate if models with linear or categorical exposures were a better fit. Five knots were specified to be consistent with the quintile analysis but also to provide enough flexibility to the model while also not being too many knots so that the model is oversensitive to the smallest fluctuations. , To correct for the measurement error that can arise from using a single baseline measurement to estimate long‐term exposure status (ie, regression dilution bias), mean values of BIA measurements at resurvey (2012–2013) from 15 694 participants were used in 2 ways. First, the HR (95% CI) in the baseline‐defined groups of aSMM and FM were plotted against the mean resurvey values in those baseline‐defined groups (termed the “usual” value). Second, where there was evidence of a log‐linear relationship, regression dilution ratios were calculated using the MacMahon‐Peto method. The log HRs (and theirSEs) per 1 SD of baseline aSMM and FM were then divided by the relevant regression dilution ratio to obtain HRs (and associated 95% CI) per 1 SD of usual aSMM and FM.

Sensitivity analyses were conducted to assess potential residual confounding or reverse causality so additional exclusions were made for events that occurred during the first 2 years of follow‐up to reduce the impact of reverse causality, for outliers, or for participants with BMI over 35 kg/m2 for whom BIA measurements may be less accurate. Additional adjustments were made for BMI (instead of FM in the SMM model and instead of SMM in the FM model) as well as for hypertension (diagnosed by doctor, taking medication, or blood pressure measurement), and blood cholesterol (defined as taking medication, plus levels of non‐high‐density lipoprotein cholesterol and triglycerides) to investigate if these are potential mediators of the associations. Finally, Cox regression models were conducted with BMI as the exposure as a “positive control” to confirm that the specified models would produce the same association that has been documented previously.

Associations of aSMM Within Tertiles of FM

To better assess the independent association of aSMM with the risk of disease, irrespective of its strong correlation with FM, we examined the sex‐specific associations of aSMM within subgroups of FM tertiles (subsequently referred to as "body composition groups," because we looked at low/moderate/high groups [tertiles] of aSMM within groups of FM). Multivariable Cox models adjusting for all covariates listed previously were used to assess the associations with CVD and all‐cause mortality using "moderate" aSMM as the reference category within each tertile of FM.

Prognostic Comparison of aSMM, FM, and Body Composition Groups With BMI, Waist Circumference, and Grip Strength

The relative importance of the various measures in prediction of CVD or mortality was assessed in several ways. First, where a linear association was present, the HRs associated with 1 usual SD change were compared for each measure. In order to assess the discriminatory ability of each measure with CVD and mortality Harrell's C‐statistic from the area under the receiver operating curve was computed. Third, the Wald test χ2 statistic was used to compare a model with just confounders to a model with confounders plus the exposure of interest to explore how much of the variation in risk is explained by each exposure of interest, given confounders.

All analyses were conducted using Stata 15.0 for analyses and R 3.5.2 for graphs. Analyses used 2‐sided P values (α=0.05) without any correction for multiple testing.

Results

Study Participants

After exclusions, the final sample included 356 590 adults who were followed for a median of 10.5 years during which there were 27 784 CVD events and 15 844 deaths due to all causes. The mean age at recruitment was 56 (SD 8) years. Men had a higher aSMM (median 27.2 kg in men; and 18.3 kg in women), although the difference between the sexes was smaller for FM (median 21.8 kg in men, 26.3 kg in women; Table 1, Tables S4 and S5). There were strong partial correlations between aSMM and FM (men r=0.71, women r=0.78) and aSMM and height (men r=0.52, women r=0.44) but not between FM and height (r=men 0.14, women 0.15; Table S6).

Table 1

Baseline Characteristics of the Study Population According to Appendicular Skeletal Muscle Mass and Fat Mass Quintiles in 356 590 UK Biobank Participants

	Men Appendicular Skeletal Muscle Mass Quintiles, Range (kg)			Men Fat Mass Quintiles, Range (kg)			Total
	Q1 13.5 to ≤24.0	Q3 26.1 to ≤27.6	Q5 30.1 to ≤54.5	Q1 5.0 to ≤15.7	Q3 19.5 to ≤22.9	Q5 27.6 to ≤98.6
Age at recruitment, y, mean (SD)	61.0 (6.4)	56.0 (7.8)	51.7 (7.8)	54.9 (8.3)	56.5 (8.1)	56.9 (7.8)	56.2 (8.1)
aSMM, kg, mean (SD)	24.6 (2.9)	26.9 (2.9)	30.5 (3.6)	24.6 (2.7)	26.7 (2.7)	31.0 (3.6)	27.2 (3.7)
FM, kg, mean (SD)	22.4 (7.1)	21.4 (7.4)	22.2 (9.2)	12.4 (2.4)	21.0 (1.0)	33.6 (6.2)	21.8 (7.8)
BMI, kg/m2, mean (SD)	26.0 (3.5)	27.4 (3.6)	29.7 (4.4)	23.3 (1.8)	27.1 (1.6)	33.1 (3.5)	27.6 (4.0)
Higher education, n (%)	12 740 (39.7%)	13 140 (40.9%)	12 645 (39.3%)	13 909 (43.1%)	12 987 (40.6%)	11 735 (36.7%)	64 706 (40.2%)
Current smokers, n (%)	3714 (11.6%)	3573 (11.1%)	3724 (11.6%)	4366 (13.5%)	3368 (10.5%)	3286 (10.3%)	18 005 (11.2%)
Low fruit and vegetable intake, n (%)	14 779 (46.0%)	13 977 (43.5%)	13 367 (41.6%)	13 624 (42.2%)	14 191 (44.3%)	14 349 (44.8%)	70 241 (43.7%)
High saturated fat intake, n (%)	11 706 (36.4%)	11 647 (36.2%)	11 778 (36.6%)	10 457 (32.4%)	11 699 (36.6%)	12 933 (40.4%)	58 554 (36.4%)
Low oily fish intake, n (%)	10 919 (34.0%)	11 342 (35.3%)	11 988 (37.3%)	11 197 (34.7%)	11 518 (36.0%)	11 661 (36.4%)	57 348 (35.7%)
Heavy drinkers, n (%)	19 948 (62.1%)	19 841 (61.7%)	18 666 (58.1%)	17 562 (54.4%)	20 297 (63.4%)	19 695 (61.5%)	97 948 (60.9%)
Low physical activity, n (%)	7247 (22.6%)	6300 (19.6%)	5518 (17.2%)	4580 (14.2%)	6030 (18.8%)	8757 (27.4%)	31 636 (19.7%)
Hypertension, n (%)	20 146 (62.7%)	18 102 (56.3%)	17 370 (54.0%)	12 683 (39.3%)	18 523 (57.9%)	23 544 (73.5%)	91 858 (57.1%)
Type 2 diabetes mellitus, n (%)	1300 (4.1%)	1156 (3.6%)	1408 (4.4%)	388 (1.2%)	906 (2.8%)	2883 (9.0%)	6305 (3.9%)
Cancer history (>5 y ago), n (%)	1176 (3.7%)	816 (2.5%)	652 (2.0%)	780 (2.4%)	856 (2.7%)	889 (2.8%)	4233 (2.6%)
Cholesterol medication, n (%)	6204 (19.5%)	4646 (14.6%)	3842 (12.0%)	2175 (6.8%)	4664 (14.7%)	7440 (23.4%)	24 008 (15.0%)

χ test for trend was performed with P<0.05 for all characteristics across the aSMM and FM quintiles. All characteristics were determined at the baseline assessment clinic through touch‐screen questionnaires, interviews, and/or physical measurements. Higher education: college or university degree or professional qualifications. Low physical activity: <600 metabolic equivalent (MET)‐minutes per week. Heavy alcohol drinker: >14 units of alcohol a week). Hypertension: systolic blood pressure >140 mm Hg, diastolic blood pressure >90 mm Hg, was diagnosed by a doctor or were taking medication to lower blood pressure. Diabetes mellitus and cholesterol: taking medication for these conditions or diagnosed by a doctor. Cancer history: diagnosed with cancer >5 years ago (those with more recent cancer had been excluded). Low fruit and vegetable intake: the lowest consumption tertile (<21 portions per week). High saturated fat: the highest saturated fat tertile, based on portions per week of beef, lamb, pork, and whether they consumed animal‐ or plant‐based spreads. Low oily fish: lowest consumption tertile (<1 portion per week). aSMM indicates appendicular skeletal muscle mass; BMI, body mass index; and FM, fat mass.

10.3 to ≤16.5

17.6 to ≤18.6

20.0 to ≤39.2

5.0 to ≤18.5

22.9 to ≤27.1

33.4 to ≤109.8

Participants in the highest quintiles of aSMM and FM had similar diets (ie, high saturated fat intake, low oily fish intake, but similar fruit and vegetable intakes), and a higher percentage of participants had low physical activity and a higher prevalence of type 2 diabetes mellitus and hypertension. A higher percentage of participants in the highest FM quintile were taking medication for cholesterol but there were no differences across aSMM quintiles (Table 1, Tables S4 and S5).

Associations of Skeletal Muscle Mass and Fat Mass With Health Outcomes

There was a potential curvilinear association between aSMM and CVD in women (likelihood ratio test statistic of nonlinearity [df=4], P<0.001) with the nadir approximately at the median; this curvilinear shape was even more pronounced in the cubic spline analysis (Figure 1, Figure S2, Table S7). There was a positive linear association in men with an HR per 1 usual SD of 1.07 (95% CI, 1.06–1.09). FM showed much stronger positive log‐linear associations with the risk of CVD with HRs per 1 usual SD of 1.20 (95% CI, 1.19–1.22) in men and 1.25 (95% CI, 1.23–1.27) in women (Figure 1, Figure S2, Table S7). These associations were similar across CVD subtypes (nonfatal, fatal, coronary heart disease, congestive heart failure, and stroke) for both aSMM and FM (Figure S3). The associations of aSMM and FM with all‐cause mortality generally followed a J‐shape in both men and women, although the association with FM was more clear (Figure 2, Figure S2, Table S8).

Figure 1

Adjusted hazard ratios (HRs) of incident cardiovascular disease associated with appendicular skeletal muscle mass (aSMM) and fat mass (FM).

A, HRs of incident CVD associated with aSMM in men, 1 SD=5.66 kg. B, HRs of incident CVD associated with aSMM in women, 1 SD=1.45 kg. C, HRs of incident CVD associated with FM in men, 1 SD=6.75 kg. D, HRs of incident CVD associated with FM in women, 1 SD=8.28 kg. For all panels, likelihood ratio tests were used to estimate nonlinearity (aSMM in men, P=0.04; aSMM in women, P<0.001; FM in men, P=0.09; FM in women, P=0.09). Adjusted HRs and CIs obtained using the floated absolute risk method of Cox proportional hazards regression, number of cases shown above each estimate and HRs shown below. Adjusted for age (underlying timescale variable), height (as a continuous variable in FM and included by regression out of variation due to height for aSMM), Townsend index of deprivation, education, smoking status, alcohol intake, physical activity, oily fish intake, fruit and vegetable intake, saturated fat intake, diabetes mellitus, cancer history, menopause (women), and mutually adjusted for FM (in the aSMM models) and aSMM (in the FM models). HRs are plotted at the mean of the resurvey values for the baseline‐defined quintiles (“usual” values) to correct for measurement error. HRs per 1 SD given where there was no evidence of departure from linearity. CVD indicates cardiovascular disease.

Figure 2

Adjusted hazard ratios (HRs) of all‐cause mortality associated with appendicular skeletal muscle mass (ASMM) and fat mass (FM).

A, HRs of all‐cause mortality associated with aSMM in men, 1 SD=5.66 kg. B, HRs of all‐cause mortality associated with aSMM in women, 1 SD=1.45 kg. C, HRs of all‐cause mortality associated with FM in men, 1 SD=6.75 kg. D, HRs of all‐cause mortality associated with FM in women, 1 SD=8.28 kg. For all panels, likelihood ratio tests were used to estimate nonlinearity P values (aSMM in men, P=0.002; aSMM in women, P=0.008; FM in men, P<0.001; FM in women, P<0.001). Adjusted HRs and CIs obtained using the floated absolute risk method of Cox proportional hazards regression, number of cases shown above each estimate and HRs shown below. Adjusted for age (underlying timescale variable), height (as a continuous variable in FM and included by regression out of variation due to height for aSMM), Townsend index of deprivation, education, smoking status, alcohol intake, physical activity, oily fish intake, fruit and vegetable intake, saturated fat intake, diabetes mellitus, cancer history, menopause (women), and mutually adjusted for FM (in the aSMM models) and aSMM (in the FM models). HRs are plotted at the mean of the resurvey values for the baseline‐defined quintiles (“usual” values) to correct for measurement error.

These findings remained robust after sensitivity analyses (Tables S9 and S10). Exclusion of the first 2 years of follow‐up to reduce the risk of reverse causality, outliers, or participants with BMI >35 kg/m2 did not affect the associations.

Adjustment for hypertension and high blood cholesterol as mediators did not affect associations of aSMM with CVD or all‐cause mortality. These mediators explained ≈30% to 40% of the χ2 statistic in models of FM and CVD, although the association between FM and all‐cause mortality was affected less (Tables S11 and S12). However, adjustment for BMI (instead of SMM or FM in their respective models) removed the positive association between aSMM and CVD in men but did not change the association observed in women. Associations between FM and CVD and all‐cause mortality were largely attenuated after adjustment for BMI, with large % of the χ2 statistic explained in both men and women.

Analyses of body composition groups (aSMM tertiles within each FM tertile) showed positive linear associations between aSMM and incident CVD risk in men in all FM tertiles, whereas women still had curvilinear associations with incident CVD except those in the highest FM tertile (Figure 3). The associations of aSMM with all‐cause mortality within FM tertiles were broadly similar between men and women and to those observed in the main analysis, except for women in the middle tertile of FM, which showed a curvilinear association between aSMM and all‐cause mortality.

Figure 3

Adjusted hazard ratios (HRs) of cardiovascular disease and all‐cause mortality associated with appendicular skeletal muscle mass (aSMM) when participants are stratified into fat mass (FM) tertiles.

A, HRs of cardiovascular disease (CVD) associated with aSMM in low fat men. B, HRs of CVD associated with aSMM in moderate fat men. C, HRs of CVD associated with aSMM in high fat men. D, HRs of CVD associated with aSMM in low fat women. E, HRs of CVD associated with aSMM in moderate fat women. F, HRs of CVD associated with aSMM in high fat women. G, HRs of all‐cause mortality associated with aSMM in low fat men. H, HRs of all‐cause mortality associated with aSMM in moderate fat men. I, HRs of all‐cause mortality associated with aSMM in high fat men. J, HRs of all‐cause mortality associated with aSMM in low fat women. K, HRs of all‐cause mortality associated with aSMM in moderate fat women. L, HRs of all‐cause mortality associated with aSMM in high fat women. For all panels, adjusted hazard ratios (HR) and CIs obtained using Cox proportional hazards regression, number of cases shown above each estimate and HRs shown below. Adjusted for age (underlying timescale variable), height (included by regression out of variation due to height), Townsend index of deprivation, education, smoking status, alcohol intake, physical activity, oily fish intake, fruit and vegetable intake, saturated fat intake, diabetes mellitus, cancer history, menopause (women), and mutually adjusted for FM (in the aSMM models) and aSMM (in the FM models). HRs are plotted at the mean of the resurvey values for the baseline‐defined quintiles (“usual” values) to correct for measurement error.

Comparing the Prognostic Value of Body Composition Measures

For CVD risk, waist circumference and FM showed the strongest associations whereas aSMM and grip strength showed the weakest associations in both men and women. For all‐cause mortality, waist circumference and grip strength showed the strongest associations, whereas aSMM remained the weakest (Figure 4, Table 2). However, the discriminatory performance (Harrell's C‐statistics) for total CVD was slightly higher for BMI compared with all the other metrics in men (C=0.63; 95% CI, 0.63–0.64) and for BMI and waist circumference in women (C=0.45; 95% CI, 0.45–0.45). Similarly, for all‐cause mortality, the discriminatory performance was highest for BMI (C=0.62; 95% CI, 0.62–0.63) and waist circumference (C=0.61; 95% CI, 0.61–0.62) in men, whereas in women it was FM (C=0.63; 95% CI, 0.62–0.63) and aSMM (C=0.63; 95% CI, 0.62–0.63). The χ2 statistic was marginally higher in the BMI and waist circumference model in men and in the BMI and the combined aSMM/FM groups in women.

Figure 4

Independent effects of body mass index (BMI), fat mass (FM), waist circumference, appendicular skeletal muscle mass (aSMM), and grip strength on cardiovascular disease (CVD) subtypes and all‐cause mortality. Adjusted hazard ratios (HRs) per SD change.

A, The independent effects of BMI, FM, waist circumference, aSMM and grip strength on CVD subtypes and all‐cause mortality in men. B, The independent effects of BMI, FM, waist circumference, aSMM, and grip strength on CVD subtypes and all‐cause mortality in women. Range excludes outliers. Adjusted hazard ratios (HR) and CIs obtained using Cox proportional hazard regression. Adjusted for age (underlying timescale variable), height (as a continuous variable in all models except aSMM where it was included by regression out of variation due to height for aSMM), Townsend index of deprivation, education, smoking status, alcohol intake, physical activity, oily fish intake, fruit and vegetable intake, saturated fat intake, diabetes mellitus, cancer history, menopause (women), and mutually adjusted for FM (in the aSMM models) and aSMM (in the FM models). HRs are corrected for regression dilution bias by the MacMahon‐Peto method.

Table 2

The Discrimination Ability of Each Body Composition Measure for the Prediction of Cardiovascular Events and All‐Cause Mortality, as Calculated by Harrell's C‐Statistic From the Area Under the Receiver Operating Curve

	Men		χ12	Women		χ12
	HR Per SD (95% CI)	Harrell's C‐Statistic (95% CI)		HR Per SD (95% CI)	Harrell's C‐Statistic (95% CI)
Cardiovascular disease
BMI	1.15 (1.14–1.16)	0.63 (0.63–0.64)	597	1.18 (1.16–1.19)	0.45 (0.45–0.45)	462
FM	1.20 (1.19–1.22)	0.56 (0.55–0.56)	445	1.25 (1.23–1.27)	0.59 (0.58–0.59)	374
Waist circumference	1.23 (1.21–1.25)	0.61 (0.61–0.62)	492	1.31 (1.28–1.33)	0.45 (0.44–0.45)	464
aSMM	1.07 (1.06–1.09)	0.54 (0.54–0.54)	73	1.00 (0.98–1.02)	0.57 (0.56–0.57)	0
Decreasing grip strength	1.09 (1.07–1.12)	0.55 (0.54–0.55)	42	1.11 (1.08–1.14)	0.44 (0.43–0.44)	51
Body composition groups	…	0.56 (0.55–0.56)	504	…	0.45 (0.44–0.45)	400
All‐cause mortality
BMI	1.08 (1.06–1.09)	0.62 (0.62–0.63)	79	1.06 (1.05–1.08)	0.61 (0.61–0.62)	40
FM	1.08 (1.06–1.1)	0.60 (0.59–0.60)	39	1.05 (1.02–1.08)	0.63 (0.62–0.63)	11
Waist circumference	1.13 (1.1–1.15)	0.61 (0.61–0.62)	84	1.09 (1.06–1.12)	0.61 (0.6–0.62)	29
aSMM	1.04 (1.01–1.06)	0.60 (0.59–0.60)	9	0.98 (0.92–1.04)	0.63 (0.62–0.63)	0
Decreasing grip strength	1.17 (1.13–1.2)	0.60 (0.60–0.61)	70	1.08 (1.04–1.11)	0.61 (0.60–0.61)	18
Body composition groups	…	0.61 (0.61–0.62)	83	…	0.61 (0.61–0.62)	41

Harrell's C‐statistic and hazard ratios (HR) per SD change calculated from the fully‐adjusted model, which adjusted for: age (underlying timescale variable), height, Townsend index of deprivation, education, smoking status, alcohol intake, physical activity, oily fish intake, fruit and vegetable intake, saturated fat intake, diabetes mellitus, cancer history, menopause (women), and mutually adjusted for FM (in the aSMM models) and aSMM (in the FM models). HRs are corrected for regression dilution bias using the MacMahon‐Peto method. One SD of aSMM is 3.34 kg (men), 1.95 kg (women) and FM is 6.79 kg (men), 8.29 kg (women). The model for aSMM in men is for increasing aSMM; in women is for decreasing aSMM. Wald test statistic was used to compare a model with just confounders to a model with confounders plus the exposure of interest. aSMM indicates appendicular skeletal muscle mass; BMI, body mass index; and FM, fat mass.

Discussion

In this prospective study of 356 590 UK adults, FM had a strong positive log‐linear association with the risk of CVD in both sexes. There was also a positive log‐linear association with aSMM for men and a curvilinear association for women. The associations of aSMM and FM with all‐cause mortality followed a J‐shape in both men and women. Analysis of the association of aSMM within tertiles of FM supported these associations with CVD and all‐cause mortality. The discriminatory ability of BMI was similar to, or better than, more specific measures of body composition (aSMM and FM), waist circumference, or grip strength in relation to CVD events or all‐cause mortality.

Few previous studies have specifically examined the association between distinct body compartments with either incident CVD or mortality. In line with previous studies we consistently observed a positive association between FM and CVD. This is consistent with previous analyses from UK Biobank that found significant associations between body fat percentage, waist circumference, and waist‐to‐hip ratio on CVD outcomes as well as meta‐analyses of prospective cohort studies assessing various adiposity measures. ,

The role of aSMM has been investigated in fewer studies, most of which used older populations with small sample sizes or a proxy for aSMM such as fat‐free mass. , , , , Our rationale for using aSMM as opposed to whole body muscle or fat‐free mass is because this tissue is more likely to be modifiable by lifestyle factors such as physical activity than other components of fat‐free mass and it is less likely to be confounded by FM given that higher abdominal FM is often accompanied by greater muscle in the trunk region. However, aSMM is a large contributor to whole body muscle and it is likely that participants would be classified in the same quintile regardless of the measure used. Although some studies have shown an inverse association between aSMM and CVD risk, our finding of a positive log‐linear association among men has been observed previously. The Aerobics Center Longitudinal Study found a similar pattern with fat‐free mass index measured by skinfold thicknesses as well as hydrostatic weighing (for which aSMM is the largest contributor) and had a comparably aged, predominantly male study population. A plausible physiological mechanism linking higher aSMM with higher CVD risk may be a higher circulating blood volume, which increases cardiac output and increases systolic blood pressure and the risk of heart failure, a phenomenon previously described mainly among people with obesity. , , A recent literature review has provided a more counterintuitive view of the role of lean mass on metabolic health, proposing the possibility of publication bias, especially if unexpected results were found. Our analysis within tertiles of FM confirmed the increased risk of CVD with aSMM even among men with lower FM levels, reducing the possibility of residual confounding by FM although this cannot be completely ruled out. Our exploratory mediation analyses showed that the association between aSMM and CVD was no longer significant in men after adjusting for BMI. This implies that if aSMM increases, FM plus all the other body compartments have to decrease in order to hold BMI constant, such that changes in body composition that increase skeletal muscle while lowering total body fat, as expected with physical training, may not be associated with increased CVD risk. Furthermore, the large changes observed in the χ2 statistic after adjustment for BMI in the aSMM model suggest a large part of the association may be explained by confounding by BMI, especially among men.

Nevertheless, it is unusual to find a CVD risk factor with such different associations between sexes. A potential explanation for this disparity could be because of differences in lifestyle factors between men and women classified as high aSMM within each FM tertile. For example, compared with women, a higher percentage of men in the same aSMM and FM tertile, reported poorer diets (low fruit and vegetable intake, high saturated fat intake), heavy drinking (over 14 units/week, National Health Service guidelines), or presented a higher prevalence of hypertension and cholesterol medication. Our findings are largely consistent with those reported from a recent study of 38 000 middle‐aged men that demonstrated a U‐shaped association between predicted lean mass and CVD death and mortality; however, these participants may have been healthier because they recruited health professionals rather than the general population. Although we adjusted for several potential lifestyle confounders our study may still have residual confounding in relation to lifestyle factors.

Clinical and Public Health Implications

BMI has been criticized as an inaccurate measure of health risks, , but at a population level, more specific measurements of body composition, namely aSMM and FM, were generally not more predictive of CVD events or mortality; an observation that also been reported elsewhere. , , The moderately improved prognostic value of BMI may reflect the combined effects of height, FM, and SMM that are each individually associated with CVD risk. In addition, BMI has less measurement error than other measures that could contribute to its marginally stronger prognostic ability. Waist circumference and other measures of central adiposity have been reported to better discriminate CVD risk in some studies, although not superior to BMI in others as happened in our study. However, waist circumference is particularly liable to observer error, whereas measures of central adiposity do not indicate whole‐body adiposity, nor is there an equivalent measure for fat‐free mass. Grip strength is often used as a functional indicator of SMM; however, it includes a volitional component and the European Working Group of Sarcopenia in Older People recommends measuring the amount of SMM to assess risk. However, the commonly used measure of the mid‐upper arm circumference is vulnerable to overestimation because it cannot distinguish between muscle fibers and intramuscular fat deposits.

However, although BMI may be the simplest measurement to assess health risk, which is important from a public health perspective, some of this risk may not be attributable solely to adiposity, particularly if the association observed with aSMM in men is confirmed, although further research is needed to better understand the biological mechanisms and impact of different body tissue compartment on health outcomes. It could therefore be beneficial to reframe BMI as a composite measure of risk. , In addition, at the individual level, additional measurements of CVD risk factors (eg, blood lipids, blood pressure) in addition to BMI or body composition are needed to classify individuals at risk and propose adequate treatments.

Strengths and Weaknesses of the Study

The strengths of this study include its large sample size, which reduces the risk of chance findings owing to random error, and the detailed measurements of the exposures, potential confounders, and outcomes from the hospital episode statistics follow‐up. Some participants had repeated measurements taken at resurvey, which allowed for the correction for random measurement error and consequent regression dilution bias.

Although BIA has many practical strengths in research and clinical settings, it is not as accurate as other methods that use physical properties of the body to measure composition, such as densitometry or DEXA or imaging methods such magnetic resonance imaging scans, and is vulnerable to estimation errors, especially at the extreme ranges of BMI or in people with conditions that affect water retention. , However, validation studies against DEXA show that it performs well in healthy individuals with a stable electrolyte and water balance. Because algorithms to estimate body composition by BIA vary, it may be that our results can be replicated only using a Tanita BC‐418 MA segmental body composition analyzer. However, studies comparing different analyzers from this manufacturer or others have reported only small differences in % body fat (eg, equivalent to 0.7 kg of difference in FM), suggesting that a participant would likely fall into the same quintile regardless of the method used.

Despite the large sample of participants studied here, one of the main limitations is that the UK Biobank presents a low response rate for the United Kingdom (5.5%); however, the associations in this study should still be valid and not affected by selection bias. The participants were predominantly people of White race, which limits the generalizability of our findings. Despite the performance of BMI in this high‐income population of UK Biobank, there is emerging evidence that suggests that BMI is not an informative measure of risk for mortality in lean populations from low‐ and middle‐income countries. For example, research from a large cohort study of 0.5 million adults in India found little association between BMI and cardiac mortality. It may be that BMI is a better indicator of risk in populations where fat mass is the dominant type of body tissue, supported in the current study by the largely equivalent associations of BMI and fat mass with the risk of CVD. Research has also suggested that the distribution of fat mass within equivalent levels of BMI may have distinct associations with the risk of cardiometabolic diseases in different ethnic groups. Thus, although BMI may be the most effective tool for risk assessment within high‐income populations further work is needed to compare its prognostic ability with more detailed measures of body composition in diverse populations. Finally, as this is an observational study we cannot eliminate the possibility that residual confounding affected our results.

Conclusions

FM showed a strong positive association with CVD risk whereas SMM showed a positive log‐linear association with CVD risk in men but curvilinear in women. Although BMI has been criticized as an inaccurate measure of risk, more specific measurements of body composition did not demonstrate improved prognostic ability to detect the risk of CVD or all‐cause mortality.

Sources of Funding

This research was funded by the National Institute for Health Research (NIHR) School of Primary Care Research (SPCR) (Piernas and Jebb); NIHR Applied Research Collaboration Oxford (Piernas and Jebb), and the NIHR Biomedical Research Centre Oxford (Jebb, Carter, Lewington, and Bennett). Additional support was obtained from the Nuffield Department of Population Health (scholarship placement for Knowles), core grants to Clinical Trial Service Unit from the Medical Research Council (Carter and Lewington, Clinical Trial Service Unit A310) and the British Heart Foundation (Carter and Lewington, CH/1996001/9454). The views expressed are those of the author(s) and not necessarily those of the National Health Service, the NIHR, or the Department of Health and Social Care.

Disclosures

None.

Data S1

Tables S1–S12

Figures S1–S3

References , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,

Click here for additional data file.