Dual-energy X-ray absorptiometry derived adiposity and colorectal cancer incidence and mortality in postmenopausal women

Abstract

Background.

Determine if dual-energy X-ray absorptiometry (DXA) derived adiposity was associated with colorectal cancer (CRC) incidence and mortality in postmenopausal women from the Women’s Health Initiative (WHI) DXA Cohort.

Methods.

Whole-body DXA scans estimated adiposity. Women with cancer history (except non-melanoma skin cancer) or missing baseline DXA were excluded. For 27 years of follow-up, outcomes and death were adjudicated. Descriptive statistics by CRC status were calculated. Fine and Gray’s competing risks regression was used to estimate sub-hazard ratios (SHR) and 95% confidence intervals (CI). Observation time was from enrollment to first CRC event or competing risk (other cancer, other cause of death); women without cancer at last follow-up were censored. Covariates included sociodemographic, clinical, and study characteristics.

Results.

After exclusions, 9,950 women were included, with 191 first-incident CRC and 88 CRC-related deaths identified. At baseline, mean (±SD) age was 63.3 (±7.4) years, and body mass index was 28.2 (±5.7) kg/m². In adjusted models, baseline continuous abdominal visceral adipose tissue (VAT) (per 100cm²) and android fat (per kg) were significantly associated with a higher risk of first-incident CRC: SHR (95% CI) 1.23 (1.04–1.45) and 1.15 (1.01–1.31), respectively. There were no significant associations between adiposity and CRC mortality.

Conclusions.

Higher amounts of abdominal VAT and android fat were associated with a higher risk of CRC incidence in postmenopausal women.

Impact.

Associations between VAT and CRC, independent of BMI, support clinical assessment of body composition across weight categories. A head-to-head comparison of VAT and BMI for CRC prediction is recommended in future research.

INTRODUCTION

Within the United States (US), colorectal cancer (CRC) is the third most common cancer and the second most common cancer-related death among women (1). In 2024, an estimated 71,270 new CRC cases and 24,310 CRC-related deaths occurred among US women (1). CRC is a well-established obesity-related cancer based on studies that have used historical body mass index categories (BMI), as a proxy for excess adipose tissue (2–6). Many studies and meta-analyses have examined CRC incidence and mortality associated with obesity (3–5,7–10). Incidence studies have highlighted a difference in the associations with obesity by sex (3–5,7). Generally, men exhibit stronger risk estimates for obesity-related CRC than women (5), while women often show weak or null associations between elevated BMI and CRC (2,7).

Though 40% of US adults are classified as obese by historical BMI categories, BMI does not provide an accurate reflection or representation of adiposity in postmenopausal women (11–15). Postmenopausal women tend to experience a shift towards abdominal deposition of adipose, and, waist circumference has been highlighted as having a stronger association than BMI with the incidence of CRC in women (2,4,16). The relationship between BMI-categorized obesity or adiposity and CRC mortality has been studied and reviewed (8–10). Evidence suggests that CRC cancer patients with higher BMI have more favorable mortality outcomes compared to normal-weight patients, despite the general relationship indicating that higher BMI increases the risk of incident disease (8–10). Therefore, understanding the relationship between directly measured adipose tissue mass and CRC risk and mortality among postmenopausal women, particularly abdominal adiposity is paramount.

The abdomen contains two primary fat compartments: subcutaneous and visceral adipose tissue (SAT; VAT). VAT is of particular interest to researchers because it is more hormonally active and hypothesized to be more deleterious than other depots due to its links with metabolic dysfunction, inflammation, and immune function – all considered hallmarks of cancer (17,18). This hypothesized relationship between VAT and cancer risk serves as a driving rationale for examining CRC and measured adiposity rather than relying on proxy measures such as BMI (19), with a recent systematic review highlighting the need for large prospective studies examining VAT and SAT specifically (19).

Within the Women’s Health Initiative (WHI) new adiposity metrics (VAT, SAT, android and gynoid fat) have been calculated (20,21), enabling assessment of abdominal adiposity at a scale and level of detail not previously available (19). The goal of this study was to assess these new metrics of adiposity in context of other measures by determining the association of measured adiposity (baseline and time-varying) with CRC incidence among postmenopausal women. We also examined the relationship between baseline abdominal adiposity and CRC-specific mortality among individuals with first-incident CRC.

MATERIALS & METHODS

Study Population

This study used the previously described WHI dual-energy X-ray absorptiometry (DXA) Cohort (20). Briefly, the WHI (N=161,808) enrolled postmenopausal women at 40 US clinical sites between 1993 and 1998 into clinical trials or an observational study (OS). A subset from Tucson/Phoenix, AZ, Pittsburgh, PA, and Birmingham, AL comprised the WHI DXA cohort (N=11,450), which underwent serial DXA scans over 6 years (22). Participants missing baseline DXA scans (N=579), data on prevalent cancers (N=145), or with a history of cancer at baseline (excluding non-melanoma skin cancer; N=731) were excluded from the incidence analysis (Supplemental Figure S1). The present investigation also included a mortality analysis in which participants were further excluded for unknown or missing mortality information (N=183).

Fred Hutchinson Cancer Research Center IRB approved the WHI in accordance with the Declaration of Helsinki (IR# 3467-EXT). Participants provided written informed consent to participate.

Primary Exposure: Adiposity

Adiposity measures and their ascertainment have been previously described (20). In brief, at baseline and years 3 and 6, whole-body DXA scans were performed to assess total and regional adiposity (QDR2000, 2000+, or 4500W DXA models and QDR software v12.1, Hologic Inc., Bedford, MA) (23,24). Assessed measures included total body, android, gynoid, and trunk fat mass and abdominal VAT, SAT, and total abdominal adipose tissue (TAT). Definitions of adiposity have been previously described (21,24). The android region of interest (ROI) is defined by a height equal to 20% of the distance from the pelvic horizontal cut line to the neck cut line, and arm cut lines are used as lateral boundaries. Gynoid ROI is a height twice that of the android region, where its upper boundary is located at a point 1.5X the height of the android region below the pelvic horizontal cut line and using leg cut lines as lateral boundaries. The abdominal VAT, SAT, and TAT ROIs are from a 5 cm high region spanning the full abdominal width just above the iliac crest, approximately at the level of the fourth lumbar vertebrae (20).

Outcomes

Outcome ascertainment and adjudication have previously been described (20). In the main WHI study, cancer outcomes and deaths were identified by self-report every six months for clinical trial participants and annually for OS (22). In extended studies, outcomes were reported annually (22). Participants reporting colon or rectal cancer were asked to provide consent for medical record review. Local adjudicators coded the primary cancer site using ICD-O-2 codes. When participants could not be reached during the annual follow-up, listed relatives or caregivers were contacted (22). If the proxy reported the WHI participant as deceased, or if the participant was lost to follow-up, their name was matched to the National Death Index for confirmation of death (22). Underlying causes of death were classified based on death certificates, medical records, and other records. Hospitalization and autopsy records were prioritized as the most reliable sources; when available, death certificate diagnoses were used.

Covariates

Several covariates were included in the analyses. Age, race and ethnicity, education, income, family history of CRC, and alcohol consumption were self-reported at baseline. Physical activity (MET-hr/wk) was derived from annual questionnaires, with time points aligned to DXA scans (25–27). Diet quality was assessed through validated food frequency questionnaires at baseline for all and at year 3 for OS participants and the Healthy Eating Index was calculated (HEI-2015) (28–30). Smoking status was self-reported at baseline and years 1, 3, and 6. Randomization to hormone therapy (HT), calcium and vitamin D (CaD), and dietary modification (DM) clinical trial arms were included as covariates. Baseline weight and height were measured, and BMI was calculated as weight(kg)/ height(m)². Skeletal muscle index (SMI) was calculated as DXA-derived appendicular lean soft tissue (kg)/ height(m)². A composite CRC screening variable was created from annual questionnaires; participants reporting colonoscopy or blood stool testing were classified as ever screened. Local versus regional and distant cancer status were taken from ICD-O-2 codes and records.

Statistical Analysis

Descriptive statistics were calculated and compared by case status using t-tests, chi-square tests, or non-parametric tests as appropriate. Descriptive statistics by VAT quartile have been previously published (20). Spearman’s rank correlations were calculated to compare DXA-derived adiposity measures and anthropometric measures at baseline.

In both incidence and mortality analyses, we used competing-risks regression based on Fine and Gray’s proportional sub-hazards model (31). The following definitions were used to define incidence and case-specific mortality:

• Incidence: Diagnosis of CRC as the first cancer diagnosis for a participant in a population of women without cancer at WHI baseline (n=9,950)

• Case-specific mortality: Death from CRC in women with a first cancer diagnosis of CRC (n=191)

To this end, we created three-level outcome variables unique to the incidence and mortality analyses. Additional descriptive statistics by the competing risks outcome variable were conducted to illustrate the differences in obesity, adiposity with significant findings, age, and follow-up time. For incidence analyses, the outcome variable was 1) survived cancer-free to last contact (censored), 2) developed CRC as the first incident cancer type observed (event), 3) developed another type of cancer first or died before developing cancer (competing risks). In a sensitivity analysis for the incidence models the event was limited to either colon or rectal cancer status, with the other cancer specified as a competing risk (Table 1). For example, in the colon cancer model, the event was colon cancer and competing risks included death before developing cancer or developed another type of cancer (including rectal cancer). In CRC case-specific mortality analyses, all first incident CRC cases were included, with the event defined as death from CRC and competing risks as death from other causes. Time since the WHI baseline was the underlying time metric, with a maximum follow-up of 27 years.

Table 1.

Total number of WHI DXA Cohort participants and mean years of follow-up, and baseline age, BMI, VAT, and android fat by analysis type

Analysis type	Participants N (%)	Follow-up Years Mean (±SD)	Baseline
			Age (Years) Mean (±SD)	BMI Mean (±SD)	VAT (100cm2) Mean (±SD)	Android fat (kg) Mean (±SD)
First incident cancer of CRC (n=9950) a
Censored: Alive at last follow-up	4,548 (45.7%)	22.8 (±1.4)	60.0 (±6.3)	27.9 (±5.5)	1.6 (±0.8)	2.5 (±1.2)
Competing risk: Other first cancer or died	5,211 (52.4%)	13.7 (±6.6)	66.0 (±7.1)	28.5 (±5.9)	1.7 (±0.8)	2.6 (±1.2)
Event: First cancer of CRC	191 (1.9%)	9.6 (±6.4)	64.0 (±7.2)	29.0 (±6.4)	1.8 (±0.9)	2.7 (±1.3)
First incident cancer of colon cancer (n=9950) b
Censored: Alive at last follow-up	4,548 (45.7%)	22.8 (±1.4)	60.0 (±6.3)	27.9 (±5.5)	1.6 (±0.8)	2.5 (±1.2)
Competing risk: Other first cancer or died	5,241 (52.7%)	13.7 (±6.6)	66.0 (±7.1)	28.5 (±6.1)	1.7 (±0.8)	2.6 (±1.3)
Event: First cancer of colon cancer	161 (1.6%)	9.4 (±6.4)	64.3 (±7.3)	28.7 (±6.1)	1.8 (±0.9)	2.7 (±1.3)
First incident cancer of rectal cancer (n=9950) c
Censored: Alive at last follow-up	4,548 (45.7%)	22.8 (±1.4)	60.0 (±6.3)	27.9 (±5.5)	1.6 (±0.8)	2.5 (±1.2)
Competing risk: Other first cancer or died	5,372 (54.0%)	13.6 (±6.6)	66.0 (±7.1)	28.5 (±5.9)	1.7 (±0.8)	2.6 (±1.3)
Event: First cancer of rectal cancer	30 (0.3%)	10.8 (±6.4)	62.7 (±6.7)	30.5 (±7.6)	1.9 (±0.9)	3.1 (±1.7)
First incident cancer of CRC case-specific mortality (n=191) d
Censored: Alive at last follow-up	66 (34.6%)	23.2 (±1.3)	60.3 (±6.0)	28.0 (±5.7)	1.7 (±0.7)	2.6 (±1.1)
Competing risk: Died of another cause	62 (32.4%)	16.8 (±5.2)	66.0 (±6.2)	30.3 (±7.0)	2.1 (±1.0)	3.0 (±1.5)
Event: Cause of death is CRC	63 (33.0%)	11.8 (±6.5)	66.0 (±7.8)	28.8 (±6.3)	1.7 (±0.8)	2.6 (±1.4)

^aAny CRC diagnoses after another primary cancer are included as competing risks not events;

^bRectal cancer diagnosis included as a competing risk;

^cColon cancer diagnosis included as a competing risk;

^dCRC case-specific mortality only includes first-incident CRC cancer patients. There are too few rectal cancer deaths to stratify for analysis; WHI – Women’s Health Initiative; DXA – Dual-energy X-ray absorptiometry; BMI – Body mass index; VAT – Visceral adipose tissue; CRC – Colorectal cancer

We fit a series of three models for the incidence analyses for each primary exposure, with the primary model being multivariable adjusted (all covariates listed in the covariates section above; Model 1). The remaining two models conducted were the multivariable-adjusted model plus either BMI (Model 2) or SMI (Model 3). In both models, BMI and SMI were added as additional covariates due to anticipated interest in their effects. However, variance inflation factors for BMI (>5) and SMI (>4) were notably high across all models and are reported in Supplemental Table S1 due to concerns for collinearity. In CRC case-specific mortality analysis, due to sample size only one model was fitted: multivariable-adjusted including age-at-diagnosis, local vs. regional or distant disease, and ever-screened for CRC. The sub-hazard ratio (SHR) and 95% confidence interval (95% CI) were presented for each model.

Continuous model results for abdominal adiposity variables (VAT, SAT, and TAT) are expressed per 100 cm² (10cm by 10cm square, roughly the size of a drink coaster) to reflect meaningful increases in adiposity. Total body, android, gynoid, and trunk fat mass continuous models results are per kg. Measurements from DXA scans at baseline, year 3 and 6 were used to assess time-varying adiposity. The models used the same covariates as baseline models, but when available, the covariates were also time-varying from the time points baseline, years 3 and 6. Time-varying exposures and covariates were entered as updated exposure values.

In sensitivity analyses, based on a priori information, models were stratified by age at baseline, race and ethnicity, and BMI category. In these models, interaction with VAT or SAT was also tested, and no statistically significant associations were found. Thus, interaction terms were not included in the final incidence and mortality analyses models. Their respective stratified models removed age, race and ethnicity, and BMI as covariates.

Multiple imputation using chained equations (MICE) with predictive mean matching was used to impute missing variables. In baseline analyses, the covariates education (n=64), income (n=734), race and ethnicity (n=374), height at baseline (n=22), alcohol intake (n=82), smoking status (n=132), physical activity (MET-hrs/wk; n=964), physical function (SF 36 score; n=218), total energy intake (kcal/day; n=22), HEI-2015 score (n=22), and first degree relative with CRC (n=444), were imputed when missing. In time-varying analyses, adipose variables were additionally imputed using predictive mean matching for years 3 (n=2456) and 6 (n=3057) when missing.

Data were analyzed using Stata 18 (RRID: SCR 012763; StataCorp, College Station, Texas) and SAS 9.4 (RRID: SCR 008567; SAS Institute Inc., Cary, North Carolina). All analyses used a type I error rate of 0.05.

Data Availability

This research utilizes pre-existing data from the WHI study (RRID: SCR 011789). Persons interested in these data can follow the formal procedures for manuscript proposals codified on the WHI website; data use agreements and research instructions are also available (www.whi.org).

RESULTS

Incidence

After exclusions, 9,950 participants were included in the incidence analysis, for a total of 177,295 person-years (Supplemental Figure S1). The baseline mean age was 63.3 (±7.4) years, and the mean BMI was 28.2 (±5.7) kg/m². Of the participants in the incidence analysis, 191 developed CRC as their first incident cancer, 1,827 developed another type of first primary cancer, 3,384 died without a cancer diagnosis, and 4,548 survived until the last follow-up (Table 1). The mean age at diagnosis of CRC was 74.3 (±8.5) years and ranged from 55 to 94 years. Among women who developed CRC, 161 were colon cancer cases, and 30 were rectal cancer cases. CRC cases had higher baseline BMI (29.0 ±6.4 kg/m²), VAT per 100cm² (1.8 ±0.9), and android fat per kg (2.7 ±1.3) than women who survived cancer-free or had a competing risk. Further, CRC cases consumed more alcohol and were more likely to have a female relative with cancer than non-cases (Table 2). Cases also had higher waist circumference (cm), TAT (100cm²), trunk fat (kg), total body lean mass (kg), appendicular lean mass (kg), and SMI than non-cases (Table 2).

Table 2.

Baseline demographic characteristics of colorectal cancer cases and non-cases in the WHI DXA Cohort (n=9,950).

Variable	Cases (n=191) N (column %)	Non-cases (n=9759) N (column %)	P
Ethnicity: Hispanic or Latina	11 (5.76%)	663 (6.79%)	0.58
Race
American Indian or Alaska Native*	<10	152 (1.56%)	1.00
Asian or Pacific Islander*	<10	36 (0.37%)	0.51
Black	32 (16.75%)	1366 (14.00%)	0.28
White	152 (79.58%)	7901 (80.96%)	0.54
Education			0.33
Less than high school (includes no educ.)	15 (7.85%)	871 (8.93%)
High school or GED completed	47 (24.61%)	2227 (22.82%)
Vocational or technical school, or some college	71 (37.17%)	3647 (37.37%)
College degree	16 (8.38%)	807 (8.27%)
Some postgraduate or professional school	21 (10.99%)	813 (8.33%)
Graduate degree	20 (10.47%)	1331 (13.64%)
Income			0.43
Less than $20,000	53 (27.75%)	2416 (24.76%)
$20,000 to $34,999	58 (30.37%)	2542 (26.05%)
$35,000 to $49,999	35 (18.32%)	1706 (17.48%)
$50,000 to $74,999	25 (13.09%)	1380 (14.14%)
$75,000 and greater	13 (6.81%)	988 (10.12%)
Smoking Status			0.08
Never	89 (46.60%)	5309 (54.40%)
Former	84 (43.98%)	3552 (36.40%)
Current	16 (8.38%)	768 (7.87%)
Alcohol intake			0.02
Nondrinker	29 (15.18%)	1650 (16.91%)
Past drinker	39 (20.42%)	2122 (21.74%)
Minimal (<1 drink per week)	51 (26.70%)	3157 (32.35%)
Moderate (1 to <7 drinks per week)	46 (24.08%)	1992 (20.41%)
Heavy (7+ drinks per week)	24 (12.57%)	758 (7.77%)
Relative with cancer
Female relative with cancer	100 (52.36%)	4372 (44.80%)	0.04
Male relative with cancer	77 (40.31%)	3239 (33.19%)	0.06
Female relative with colorectal cancer	13 (6.81%)	744 (7.62%)	0.28
Male relative with colorectal cancer	18 (9.42%)	757 (7.76%)	0.98
Any relative with colorectal cancer	29 (15.18%)	1425 (14.60%)	0.90
BMI category			0.38
Underweight (<18.5) a	<10	75 (0.77%)
Normal (18.5 – 24.9)	60 (31.41%)	3107 (31.84%)
Overweight (25.0 – 29.9)	64 (33.51%)	3434 (35.19%)
Obesity I (30.0 – 34.9)	35 (18.32%)	1941 (19.89%)
Obesity II (35.0 – 39.9)	21 (10.99%)	787 (8.06%)
Extreme Obesity III (≥ 40)	11 (5.76%)	389 (3.99%)
	mean ± SD	mean ± SD
Age at baseline (y)	64.03 ± 7.21	63.23 ± 7.38	0.14
Height at baseline (cm)	162.16 ± 6.25	161.61 ± 6.13	0.22
Physical activity (MET-hrs/wk)	10.16 ± 11.29	11.44 ± 13.79	0.46
Physical function (SF 36 score)	77.62 ± 20.92	78.60 ± 21.27	0.53
Total energy intake (kcal/day)	1,713.19 ± 708.15	1,654.21 ± 801.32	0.26
HEI-2015 score	64.05 ± 10.45	63.47 ± 10.68	0.45
BMI	28.99 ± 6.37	28.19 ± 5.70	0.08
Waist Circumference (cm)	87.82 ± 14.04	85.82 ± 13.13	0.04
DXA Body Composition
VAT (cm2)	180.62 ± 86.89	166.23 ± 81.76	0.02
SAT (cm2)	399.22 ± 145.48	380.44 ± 138.04	0.06
TAT (cm2)	579.84 ± 217.97	546.67 ± 206.75	0.03
VAT to SAT ratio	0.44 ± 0.14	0.43 ± 0.14	0.07
Total body fat (%)	44.50 ± 7.37	43.87 ± 7.29	0.24
Total body fat (kg)	34.33 ± 12.89	32.58 ± 11.48	0.07
Android fat (kg)	2.74 ± 1.35	2.53 ± 1.23	0.02
Gynoid fat (kg)	5.81 ± 1.84	5.61 ± 1.80	0.14
Trunk fat (kg)	16.19 ± 6.86	15.12 ± 6.20	0.03
Skeletal muscle index (kg/m2)	5.80 ± 1.07	5.65 ± 0.98	0.04

^aCell sizes are not reported due to small sizes. HEI: Healthy eating index; BMI: Body mass index; and Abdominal VAT: Visceral adipose tissue; SAT: Subcutaneous adipose tissue; TAT: Total adipose tissue at L4 ROI

Spearman’s rank correlations are illustrated in Supplemental Figure S2. Six correlations had scores above 0.95: 1) android fat and TAT (0.99), 2) android fat and trunk fat (0.97), 3) TAT and SAT (0.97), 4) trunk fat and TAT (0.96), 5) android fat and SAT (0.95), and 6) trunk fat and total body fat mass (0.95). All correlations above 0.95 come from adiposity measures with overlapping ROIs. BMI was most highly correlated with total body fat mass (0.93), trunk fat (0.91), and TAT (0.90).

In baseline, multivariable-adjusted models, VAT (per 100cm²) and android fat (per kg) had statistically significant associations with a higher risk of CRC: SHR (95% CI) 1.22 (1.01, 1.48) and 1.15 (1.01, 1.31), respectively (Figure 1). In baseline multivariable quartile models, there were no significant associations between abdominal measures of adiposity and CRC risk (Supplemental Table S2).

Figure 1: Multivariable-adjusted associations between baseline adiposity variables (continuous) and first incidence of colorectal cancer and colorectal cancer case-specific mortality in the WHI DXA Cohort.

First-incident CRC model 9,950 women observed over 177,295 person-years total (cases=191). Model adjusted for: height at baseline, region, education, income, race and ethnicity, hormone replacement therapy trial arm, diet modification trial arm, calcium and vitamin D trial arm, alcohol intake, smoking status, physical activity (MET-hrs/wk), physical function (SF 36 score), total energy intake (kcal/day), HEI-2015 score, and relatives with CRC. Competing risks: death without developing any type of cancer and developing a first primary cancer other than colorectal cancer. Case-specific mortality model 191 women observed over 3,313 person-years total (deaths=63). CRC case-specific mortality only includes first-incident CRC cancer patients. Models adjusted for: age at diagnosis, tumor stage, and ever screened for colorectal cancer. Competing risks are: death other than colorectal cancer. Multivariable adjusted models used MICE to address missing covariates. VAT: Visceral adipose tissue; SAT: Abdominal subcutaneous adipose tissue; TAT: Total abdominal adipose tissue; BMI: Body mass index; SMI: Skeletal muscle index; SHR: Sub-hazard ratio; 95% CI: 95% confidence interval; CRC: Colorectal cancer

In sensitivity analyses, colon and rectal cancer were examined separately (Supplemental Table S3). In these models, there were no statistically significant associations between adiposity and the incidence of colon cancer. In contrast, VAT (per 100cm²), SAT (per 100cm²), TAT (per 100cm²), total body fat (per kg), android fat (per kg), gynoid fat (per kg), and trunk fat (per kg) were all statistically significantly associated with higher risk of incident rectal cancer, with increased risk ranging from 5–62%.

CRC incidence models were stratified by age group, BMI category, and race and ethnicity. VAT (per 100cm²) and VAT/SAT ratio (per 0.1) had statistically significant associations with higher risk of CRC in women aged 50–64 at baseline, but not in other age groups (Figure 2). In baseline BMI-stratified multivariable models, TAT (per 100cm²), android fat (per kg), and trunk fat (per kg) had statistically significant associations with incident CRC in women who were obese (Supplemental Table S4). The remaining adiposity measures had no significant associations with CRC incidence in women who were obese. None of the adiposity measures were associated with altered risk in women who were normal weight or overweight. In race and ethnicity stratified multivariable models, no statistically significant findings for any of the associations were found between adiposity and CRC incidence (Supplemental Table S5).

Figure 2: Multivariable adjusted associations between baseline adiposity variables (continuous) and incident colorectal cancer stratified by baseline age category in the WHI DXA Cohort (multivariable-adjusted models).

Multivariable model adjusted for: height at baseline, region, education, income, race and ethnicity, hormone replacement therapy trial arm, diet modification trial arm, calcium and vitamin D trial arm, alcohol intake, smoking status, physical activity (MET-hrs/wk), physical function (SF 36 score), total energy intake (kcal/day), HEI-2015 score, and relatives with colorectal cancer. Age 50–64 category: n=5,460 and cases=96; Age 65–79 category: n=4,460 and cases=95; Competing risks are: death without developing any type of cancer and developing a first primary cancer other than colorectal cancer. Multivariable adjusted models used MICE to address missing covariates; VAT: Visceral adipose tissue; SAT: Abdominal subcutaneous adipose tissue; TAT: Total abdominal adipose tissue; SHR: Sub-hazard ratio; 95% CI: 95% confidence interval.

In baseline multivariable models further adjusted for BMI (model 2), the association seen for VAT was attenuated (Supplemental Table S1). In model 2, android fat (per kg) had a statistically significant positive association with risk: SHR (95% CI) 1.27 (1.01, 1.59). There were no significant findings in models further adjusted for SMI (Model 3). In time-varying models, there were no significant associations between adiposity and risk of CRC (Supplemental Table S1).

Mortality and Case-specific Survival

In case-specific analyses, 191 CRC cases were included, for a total of 3,313 person-years. Of these cases, 63 died of CRC (58 colon cases; 5 rectal cases), 62 died of other causes (50 colon cases; 12 rectal cases), and 66 survived with CRC (54 colon cases and 12 rectal cases) (Table 1). CRC cases that died of CRC had a mean baseline age of 66.0 (±7.8) years and a baseline BMI of 28.8 (±6.3) kg/m², VAT of 1.7 (±0.8) 100cm², and android fat of 2.6 (±1.4) kg. There were no statistically significant associations between baseline and time-varying CRC case-specific survival analyses (Figure 1).

DISCUSSION

Among postmenopausal women in the WHI DXA Cohort, we found a significantly higher risk of CRC incidence with higher baseline VAT (22%) and android fat (15%). VAT is included in the android ROI for most of the participants (21), and the two were highly correlated (Spearman’s ρ 0.92, p=0.0001). Though the other measures of adiposity did not have significant positive findings, their lower CI was notably close to one. Time-varying analyses were null, but the direction of the relationship with abdominal adiposity aligned with our hypothesis. These findings suggest that there should be a focus on abdominal adiposity when assessing CRC incident risk, and future research with a larger sample could better assess the potential significance of other regions.

As mentioned earlier, VAT is a metabolically active tissue linked to several cancer hallmarks, though a direct molecular connection to CRC remains unidentified (18,19,32). VAT secretes various proinflammatory factors that affect signaling pathways involved in cancer development and progression (18,19,32). SAT also participates in tumorigenesis, as it is white adipose tissue, and inflamed SAT may contribute to the microenvironment, but likely to a lesser extent (32). This biological context supports the relevance of our findings regarding abdominal adiposity.

Kabat et al. (2013) previously examined obesity and CRC incidence in the WHI DXA Cohort using the historic DXA scan values for fat mass, fat-free mass, and percent body fat (33). They found no significant associations or trends between quartiles of BMI, waist circumference, or waist-hip-ratio and CRC incidence (p-trends: 0.55, 0.23, 0.40, respectively) (33). Other WHI researchers later calculated and validated the additional body composition measures of android fat, gynoid fat, VAT, SAT, and TAT from these DXA scans (21). In both our current study and the study by Kabat et al. (2013), there were no significant associations between total body fat, body fat percent, and trunk fat with CRC in women of the WHI DXA Cohort. This again illustrates the potential importance of regional adiposity in favor of whole-body fat.

In research, body composition and carcinoma are not commonly assessed (19). Body composition data are primarily derived from bioelectrical impedance (BIA) scales, computerized tomography (CT) scans, or magnetic resonance imaging (MRI) (18,19,34). Major national studies, such as the UK Biobank, tend to use BIA scales to assess body fat composition because it is less expensive than CT and MRI, requires a relatively short time to perform, and generates no radiation concerns, but this limits measures to fat mass and fat-free mass for the majority of participants (19,35). Due to the differences in the 2-component model of BIA and the 3-component model of DXA, it is difficult to compare body composition measures taken from these separate instruments. Furthermore, few studies have examined the association between measured VAT and the risk of CRC in women, and have largely been conducted in smaller hospital-based cohorts (19). An example is the study of Lee et al. (2014), who observed a significant increase in risk for the third tertile of CT-derived VAT in postmenopausal women [odds ratio 2.96 (95% CI 1.38–6.33)] (36). While the study design and instrument were different, they similarly illustrated the higher risk for CRC among women with higher levels of VAT. More commonly, researchers have used anthropometric data to predict VAT in large prospective cohorts (37,38). Given that anthropometric predictions of VAT are not measured, it is difficult to compare findings, but the association between VAT and a higher risk of CRC is consistent. Overall, anthropometric measures are easier in the field, but DXA scans are more accessible than CT and more precise than anthropometrics. Beyond using different tools, many studies assess adenoma rather than CRC due to its higher prevalence. Studies with adenoma have shown that increased VAT is associated with increased adenoma risk (19,39). However, comparisons between adenoma and carcinoma findings are limited, as only about 5% of adenomas progress to cancer (40).

Interestingly, we observed no significant associations between adiposity and colon cancer alone, whereas most of the adiposity measures were associated with rectal cancer alone. Women who developed rectal cancer had higher BMI, VAT, and android fat than the women who developed colon and colorectal cancer (Table 1). In various meta-analyses examining anthropometric obesity and CRC, rectal cancer specific associations were weaker or null compared to colon and colorectal associations in women (2,3,41–43). Further, studies using BIA scales or DXA scans tended to show no significant associations between fat mass, fat-free mass, and percent total body fat and risk of colon or rectal cancer in women (35,44,45). While statistically significant associations were observed, it is unclear if additional rectal cancer cases would attenuate these associated, given patterns historically seen in the literature (2,3,41–43). Therefore, further research is necessary to examine these associations with VAT and android fat.

We decided a priori to stratify by age, and there was no significant interaction between VAT and age with respect to CRC risk (p=0.08). Overall, there were significant associations found among younger postmenopausal women. There was a significantly higher risk of CRC in women aged 50–64 with higher VAT (35%) and VAT/SAT ratio (12%). The VAT/SAT ratio was of particular interest because it allowed us to account for adipose distribution in the abdominal region. This finding suggests that we need to consider the balance between the two abdominal depots, specifically in younger postmenopausal women.

In our BMI-stratified models for those women categorized as obese, it may be that any body composition measure is enough to indicate a higher risk for CRC. In regression models that adjusted for BMI, we found that higher android fat (27%) and trunk fat (6%) were significantly associated with CRC risk. However, caution is required in interpreting these findings given the collinearity between the adiposity measures and BMI, which is best illustrated through the increase in the width of the CIs in these models (Supplemental Table S1). Further, BMI is highly correlated with the adiposity measures with correlation coefficients from 0.80–0.93 for 8 of the 9 values assessed (Supplemental Figure S2). Therefore, caution is warranted when interpreting the BMI-adjusted models.

In this WHI DXA Cohort, we observed no significant associations between body composition and CRC case-specific mortality, with SHRs notably close to one. Adiposity measures taken proximal to diagnosis may better assess effects on incidence than mortality, potentially leading to these null findings. Cheng et al. (2022) conducted a meta-analysis of adiposity and cancer survival and similarly found no associations between VAT, SAT, and TAT and CRC mortality (8). In contrast, studies using waist-related anthropometric measures, a proxy for abdominal obesity, have shown that higher abdominal obesity is significantly associated with higher risk of CRC mortality (8). However, studies that utilized waist measurements are rarer highlighting potential issues of publication bias (8). Larger scale studies with sufficient power are needed for mortality analysis comparisons between anthropometric measures and image-derived adiposity.

Strengths and Limitations

The WHI Study conducted a robust characterization and follow-up of its participants. This immense collection of information allowed adjustment for a multitude of potential confounders and the study of cancer incidence and death via adjudication over 27 years of follow-up. Further, the DXA Cohort allowed access to validated adiposity variables that are not commonly examined in population-based cohorts. Regardless, power was limited by the relatively small number of cases and deaths across analyses. This limitation was especially evident in race and ethnicity strata, which had null findings for groups previously shown to have elevated risks using other measures (7,46). A more racially and ethnically diverse study is needed to better understand potential differences in associations. Due to the small number of events within 5 years (55 incident cases, 21 CRC-specific deaths), we were unable to conduct mortality analyses with adiposity closer to time of death. Another limitation is the use of DXA, which provides a two-dimensional measure of adiposity, in contrast to the three-dimensional imaging of CT or MRI (18). This variation in equipment complicated comparisons across DXA, CT, and MRI-derived adiposity measures. Additionally, the WHI did not report proximal and distal colon cancer separately, preventing assessment of these outcomes. Finally, the findings are not generalizable to men.

Conclusion

Baseline DXA-derived abdominal VAT and android fat were significantly associated with increased CRC risk in postmenopausal women in the WHI DXA Cohort. In future research, VAT and android fat from DXA may help clarify the relationship between regional adiposity and CRC incidence and mortality. A direct comparison of abdominal adiposity and BMI for CRC prediction is also recommended. Replication in men and more diverse populations is needed to better understand adiposity and CRC associations.