Body size, body composition and endometrial cancer risk among postmenopausal women in UK Biobank

Previous studies on the association of adiposity with endometrial cancer risk have mostly used body mass index (BMI) as the main exposure of interest. Whether more precise measures of body fat, such as body fat percentage and fat mass estimated by bioimpedance analyses, are better indicators of risk than BMI is unknown. The role of central adiposity and fat‐free mass in endometrial cancer development remains unclear. We used Cox regression models to estimate hazard ratios (HR) and corresponding 95% confidence intervals (CI) for the associations of various measures of body size/composition with the risk of endometrial cancer among 135 110 postmenopausal women enrolled in UK Biobank. During a mean follow up of 6.8 years, 706 endometrial cancers were diagnosed, with a mean age at diagnosis of 65.5 years. The HRs (95% CIs) for endometrial cancer per 1 SD increase in BMI, body fat percentage and fat mass were broadly comparable, being 1.71 (1.61‐1.82), 1.92 (1.75‐2.11) and 1.73 (1.63‐1.85), respectively. There was an indication of positive association between central adiposity, as reflected by waist circumference (HRper 1‐SD increase = 1.08, 95% CI: 1.00‐1.17) and waist to hip ratio (HRper 1‐SD increase = 1.13, 95% CI: 1.01‐1.26), and endometrial cancer risk after accounting for BMI. Fat‐free mass was not an independent predictor of risk in this cohort. These findings suggest that body fat percentage and fat mass are not better indicators of endometrial cancer risk than BMI. Further studies are needed to establish whether central adiposity contributes to risk beyond overall adiposity.

Consequently, BMI may not accurately capture the true relationship between adiposity and endometrial cancer risk. Body fat percentage and fat mass are more precise measures of overall fatness than BMI.
Whether these measures are more strongly related to endometrial cancer risk than BMI is unknown.
The distribution of body fat, in particular central adiposity, is linked to several metabolic abnormalities including insulin resistance and inflammation that are associated with endometrial cancer development. 10,11 Although several studies have evaluated central adiposity, typically assessed by waist circumference, waist to hip ratio or waist to height ratio, in relation to risk of endometrial cancer, [12][13][14][15][16][17][18][19][20][21] it remains unclear whether central adiposity contributes to risk independently of overall adiposity.
Beyond adiposity, other aspects of body size and composition may also influence endometrial carcinogenesis. Height, a marker of early life and nutritional factors, has been associated with increased risk of endometrial cancer in some studies [22][23][24][25] but not others. 13,19,[26][27][28] Fat-free mass has also been associated with an increased risk of endometrial cancer, 29 however, it is unclear whether this association reflects confounding by BMI.
We used data from women in UK Biobank to assess the associations of several measures of body size and body composition with the risk of endometrial cancer. Specifically, we investigated whether more precise measures of overall adiposity (body fat percentage and fat mass) are better indicators of endometrial cancer risk than BMI, and whether central adiposity (waist circumference, waist to hip ratio, waist to height ratio, trunk fat percentage) contributes to risk independently of overall adiposity.

| Study population
UK Biobank is a prospective study designed to investigate the associations of genetic and environmental factors with the risk of chronic disease. Details of the recruitment, data collection and follow-up are described elsewhere. 30

| Assessment of body size and composition
At baseline, trained personnel collected data on body size and composition from all participants using a standard protocol. Height (cm) was measured to the nearest 0.1 cm using a Seca 240 cm height measure, with participants standing barefoot and with their head positioned in the Frankfort plane. Waist and hip circumference (cm) were measured using a Seca 200 cm tape. Waist circumference was measured at the narrowest part of the trunk or the level of the umbilicus at the end of a normal expiration, while hip circumference was measured at the widest part of the buttocks.  the HR per one standard deviation (1-SD) increase in each body size/ composition variable was also estimated. The likelihood ratio χ 2 statistics of these models were compared to assess the amount of variation explained by each body size/composition variable. 31 All models were stratified by year of birth (≤1940, 1941-1945, 1946-1950, 1951-1955, 1956-1960, 1961-1965, ≥1966) and year of recruitment (≤2007, 2008, 2009, 2010) and adjusted for age (continuous), quintiles of Townsend deprivation score, and UK region (10 regions). Additional adjustments were made for duration of oral contraceptive use (never, 0 to <5, 5 to <10, 10 to <15, ≥15 years), age at menopause (<49, 49 to <52, 52 to <55 and ≥55 years), use of hormone replacement therapy (HRT: never, past, current), physical activity (low, moderate, high), parity (0, 1, 2, ≥3), age at menarche (<13, 13 to <15, ≥15 years) diabetes status (no, yes) and smoking status (never, past, current). For each covariate, missing values were assigned to a separate category.

| Statistical analysis
To assess whether central adiposity and fat-free mass contribute to endometrial cancer risk independently of overall adiposity, the associations of central adiposity measures and fat-free mass with risk were further adjusted for BMI using the residual method. 32 For this analysis, residuals for each of the central adiposity and fat-free mass measures were obtained from a linear regression of that variable on BMI. These residuals (categorised into quartiles or standardised to SD units) were investigated in relation to endometrial cancer risk in multivariable Cox models, which were additionally adjusted for BMI. The effect of adjustment for fat mass (instead of BMI) on the association of fat-free mass with risk of endometrial cancer was also evaluated.
To allow for comparison between any two groups, even if neither is the referent category, the corresponding group-specific confidence intervals (gsCIs) for the HRs were calculated and presented in tables and figures 33 ; however, conventional 95% CIs are given in the texts.
The proportional hazard assumption was tested using log-log plots and tests based on Schoenfeld residuals. To correct for regression dilution bias, the HRs based on quartiles of body size/composition variables were plotted against the repeat assessment mean values within the same baseline categories. 34 For HRs per 1-SD increase, the log HRs and standard errors for the different measures were divided by their respective regression dilution ratios (RDR), which were estimated by regressing the repeat assessment values on the baseline measures in the subset of the cohort with repeat assessment data.
Several sensitivity analyses were performed. First, we repeated all analyses among never users of HRT, since HRT use is known to attenuate the associations of adiposity with endometrial cancer risk. 35 Second, we assessed the associations of all measures of adiposity with the risk of type I tumours, as it has been suggested that the aetiology of type I and type II tumours may differ. 36 There were too few cases of type II tumours (n = 79) to allow for any meaningful analysis of associations with this subtype. Third, to assess the potential impact of missing data, we excluded participants with missing data on any of the covariates. Fourth, we adjusted the associations of central adiposity measures with endometrial cancer risk for BMI using the standard method (ie, with BMI included as a covariate) instead of the residual method. Lastly, we used standard categories for BMI (<25, 25 to <30, 30 to <35, ≥35 kg/m 2 ), waist circumference (<80, 80 to <88, ≥88 cm) and waist to hip ratio (<0.85, ≥0.85) to enable comparison with other study populations.
All analyses were performed using STATA version 15.0.

| RESULTS
The baseline characteristics of the 135 110 women according to quartiles of BMI as measured at recruitment are shown in Table 1. Compared to women in the lowest quartile, women in the highest quartile of BMI were more likely to be younger at menarche and to have a history of diabetes. They were less likely to be nulliparous, ever users of contraceptive and current smokers. With the exception of height, the mean values of all body size and body composition variables increased in parallel with increasing BMI (Table 1). During a mean follow up of 6.8 years, 706 endometrial cancers were identified, with a mean (SD) age at diagnosis of 65.5 (5.3) years.
The age-adjusted partial correlation coefficients (r) between the various measures of body size and composition are shown in Table 2. BMI was positively correlated with weight, body fat percentage, fat mass and measures of central adiposity, with correlation coefficients (r) ranging from 0.45 for waist to hip ratio to 0.94 for fat mass. Waist to hip ratio showed weak to moderate correlations with other adiposity measures (r < .45), with the exception of waist circumference (r = .74) and waist to height ratio (r = .74). All bioimpedance-derived measures of adiposity were highly correlated with each other (r > .80).
Fat-free mass was correlated with BMI (r = .69) and to varying extents with other measures of adiposity (r > .25). Height was not correlated with body fat percentage (r = .01), waist circumference (r = .03) or waist to hip ratio (−0.08) but showed weak correlations with other measures of adiposity ( Table 2). As expected, measures of central adiposity and fat-free mass that were adjusted for BMI (using the residual method) were not correlated with BMI (r ranged from −.0043 for waist to height ratio to .0045 for fat-free mass; data not shown). Table 3 and Figure 1 show the HRs for the associations of height and overall adiposity measures with the risk of endometrial cancer. All Baseline characteristics of participants according to quartiles of BMI  Table 3).
All central adiposity measures (waist circumference waist to hip ratio, waist to height ratio and trunk fat percentage) were also associated with an increased risk of endometrial cancer in an approximately linear fashion (P trend < .0001, Table 4 and Figure 1).  (Table S1).
However, additional adjustment for BMI completely attenuated these associations (P trend > .20; Table 4). Adjustment of the association between fat-free mass and endometrial cancer for fat mass (instead of BMI) produced similar results (P trend = .20, data not shown). The majority of incident endometrial cancers in this cohort were type I tumours (~80%). In analyses restricted to this subtype, the associations of adiposity measures with risk were slightly strengthened but the overall pattern remained unchanged (Table S2). Similarly, restricting the analysis to postmenopausal never HRT users did not substantially alter the overall pattern of the associations (Table S3).
Results for analyses that excluded participants with missing data on any covariates were not materially different from that of the main analysis (data not shown). Note: Minimally adjusted models stratified by year of birth and year of recruitment and adjusted for age, deprivation score and region. Multivariable-adjusted models include all the variables defined above, plus use of HRT, smoking status, physical activity, parity, age at menopause, oral contraceptive use, age at menarche and diabetes status. Standard deviations for the body size/composition measures were as follows: 12.3 cm for waist circumference, 10.1 cm for hip circumference, 0.07 for waist to hip ratio, 0.08 for waist to height ratio and 7.6% for trunk fat percentage and 4.8 kg for fat-free mass. Each HR (95% CI) per 1-SD increase was corrected for regression dilution bias by dividing the log HR and SE by the regression dilution ratio (RDR). The RDRs for waist circumference, hip circumference, waist to hip ratio, waist to height ratio, trunk fat percentage and fat-free mass were 0.87, 0.89, 0.63, 0.88, 0.83 and 0.91, respectively. Likelihood ratio where risk increased by 58% for each 1-SD increase in body fat percentage (that was also measured by bioimpedance). 37 Greater fat mass was also associated with an increased risk of endometrial cancer.
Although both body fat percentage and fat mass are more precise measures of overall adiposity than BMI, the results of our study suggest that they are not better indicators of endometrial cancer risk.
Previous prospective studies examining the independent effect of central adiposity on endometrial cancer risk have reported conflicting results. [12][13][14][15][16][17][18][19][20] In the California Teachers' Study, waist circumference and waist to hip ratio were positively associated with endometrial cancer risk after adjustment for BMI. 17 In contrast, the Nurses' Health Study did not report an independent association with these measures, 14 while the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort reported an independent association with waist circumference but not waist to hip ratio. 19 A 2015 meta-analysis also found an independent association with waist circumference but not waist to hip ratio. 5 One study that used Mendelian randomisation found no evidence to support an association between waist to hip ratio and endometrial cancer risk. 38 In the present analysis, a small increase in risk remained for waist circumference and waist to hip ratio after adjusting for BMI, suggesting that central adiposity as reflected by these measures may confer risk above and beyond overall adiposity.
Hip circumference was not an independent predictor of endometrial cancer risk in our study, which is consistent with results from the EPIC cohort, 19  or indirectly by increasing bioavailability of oestrogen and insulinlike growth factor-1 (IGF-1). [40][41][42][43] Adiposity may also influence endometrial cancer risk by decreasing the circulating levels of adiponectin, which protects against the risk of endometrial cancer. 44,45 Obesity-induced low-grade chronic inflammation may also mediate this association. 11,44 The strengths of our study include its large size, prospective design and breadth of data on body size and composition, all of which were measured rather than self-reported, thereby minimising potential for differential misclassification, which has been particularly shown for self-reported body weight and height. 46 Repeat measures available in a subset of the cohort allowed for assessment of the extent of measurement error (quantified by RDRs) associated with each measure and the correction of risk estimates for regression dilution bias. With the exception of waist to hip ratio, the RDRs for the body size/composition indices were small, suggesting that these measures did not change substantially over a 5-year period.
Several limitations of our study need to be considered. The associations between body size/composition variables and endometrial cancer risk reported in our study largely reflect associations with type I tumours, which account for majority of the incident cases (~80%).
The small number of type II tumours precluded analysis of associations with this subtype. Participants in UK Biobank were predominantly of European origin; therefore, the results of our study may not be generalisable to other ethnic group. Missing values for each covariate were assigned to a separate category to prevent loss of data. This method of handling missing data has limitations, including the possibility of producing biased results due, at least in part, to incomplete adjustment for confounders. 47 However, results from a sensitivity analysis that excluded participants with missing data on any covariate were not appreciably different from that of the main analysis.
Body composition parameters in our study were estimated by bioimpedance analysis, which is a valid technique for estimating body composition. 48 Unlike imaging techniques such as dual-energy X-ray absorptiometry (DEXA) and magnetic resonance imaging (MRI) that measure body composition with high precision, bioimpedance measures the resistance/impedance of body tissues to a small alternating current and uses the impedance readings in a predictive equation to estimate various body fractions. 49 The accuracy of bioimpedance-derived measurements of body composition depends on several factors including hydration status, presence of medical conditions that may impact fluid and electrolyte balance and the prediction equation used. 50 Prediction equations are often population-specific as they describe the empirical relationship between impedance and body composition estimated by reference methods in a given population. 49 In the present analysis, estimates of body composition were based on prediction equations incorporated into the software of the Tanita BC418MA body composition analyser, which may not best represent the UK Biobank cohort. Plans are, however, under way to validate bioimpedance-derived measures of body composition against MRI and DEXA scans currently being taken in a large subset of this cohorts.
In summary, BMI, body fat percentage and fat mass showed similar associations with endometrial cancer risk. Central adiposity, as reflected by waist circumference and waist to hip ratio, may be associated with endometrial cancer risk independently of BMI, but this requires further investigation.