Body Size and Incident Colorectal Cancer: A Prospective Study of Older Women

Each year, nearly 150,000 new and 50,000 fatal colorectal cancer (CRC) cases are recorded in the United States (1). Environmental exposures are thought to have a major influence on colorectal carcinogenesis, but the full spectrum of potentially modifiable risk factors remains incompletely defined. In the United States, approximately two thirds of the adult population is overweight or obese (2, 3), which represents a putative risk factor for multiple target organ malignancies, including CRC (4). Yet, for reasons that remain incompletely defined, the relationship between excess body size and CRC risk appears to differ by gender, with generally stronger associations reported for men than for women (5–7).

Large-scale epidemiology investigations have typically relied on self-reported data to assess body size relationships. Compared with measured values, self-reported height, weight, body mass index (BMI), hip circumference, waist circumference, and waist-to-hip ratio (WHR) data appear to be highly accurate (8–10) and appropriate for application to population-based studies. BMI provides a useful approximation of general obesity, whereas waist circumference and WHR provide better representations of abdominal obesity. Several recent studies have suggested that waist circumference may be a better indicator of chronic disease risk than BMI (11–15), although data referent to CRC risk remain relatively limited (11, 16–20).

In the current prospective study, we examined associations between body size and incident CRC among participants enrolled in the population-based Iowa Women's Health Study (IWHS). Multiple body size measures were recorded at baseline (height, weight, hip circumference, and waist circumference), along with self-recalled weights at ages 50, 40, 30, and 18 years. These data allowed us to address the overarching hypothesis that increased body size is positively associated with incident CRC in older women, with the ability to comprehensively model CRC risks based on indicators of general and abdominal obesity, as well as age-specific and cumulative exposures. Moreover, because proximal and distal CRCs have been shown to exhibit distinct clinicopathologic features (21–23), we estimated body size–associated CRC risks overall and by anatomic subsite. These findings update and extend prior reports from the IWHS cohort, which were based on fewer exposure variables, shorter follow-up duration, and more limited outcome data (excluding rectal cancers; refs. 11, 24).

Approvals for this study were obtained from the Institutional Review Boards for Human Research at Mayo Clinic Rochester and the University of Minnesota.

Details regarding the methods used for recruitment and enrollment of IWHS participants have been previously reported (25). Briefly, a 16-page baseline questionnaire was mailed out in January 1986 to 99,826 randomly selected women, ages 55 to 69 years, who resided in Iowa and held a valid driver's license. A total of 41,836 women (42%) returned the baseline questionnaire and these subjects constitute the parent IWHS cohort. An earlier report showed that demographic characteristics and CRC incidence rates were similar between initial survey responders and nonresponders (26). Vital status and state of residence were determined by mailed follow-up questionnaires in 1987, 1989, 1992, 1997, and 2004, as well as through linkage to Iowa death certificate records. Nonrespondents were checked via the National Death Index to identify decedents. Recent estimates suggest the migration rate out of Iowa for this cohort is approximately 1% per year (27). For this study, women with follow-up less than 1 day (n = 10), history of malignancy other than nonmelanoma skin cancer (n = 3,830), or incomplete baseline body size information (n = 1,178) were excluded (not mutually exclusive), leaving 36,941 subjects in the final analytic cohort.

Self-reported height and weight values were recorded at baseline, along with recalled weights at ages 50, 40, 30, and 18 years. In addition, subjects were provided with a paper tape measure with instructions to have a friend or spouse assess waist circumference 1-inch superior to the umbilicus and hip circumference at the maximum hip dimension (each recorded to the nearest one-quarter inch). Subjects were asked to provide 2 waist and hip circumference measurements each to ensure reliability. This approach to body size measurement has been shown to be accurate in this cohort (10). Other potentially relevant exposure variables were also recorded at baseline, including age at menopause, exogenous estrogen use, history of oral contraceptive use, smoking status, alcohol consumption, physical activity level, self-reported diabetes mellitus, and dietary habits (using a 127-item food frequency questionnaire, similar to the Nurses' Health Study questionnaire; ref. 28). Since family history of CRC and nonsteroidal anti-inflammatory drug use were not systematically recorded at baseline, these variables were not included in the current data analyses.

Incident CRC cases were identified through the State Health Registry of Iowa, which participates in the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) program (29). Annual matching between a computer-generated list of all cohort members and the records of Iowans with incident cancer in the SEER program registry was completed through December 31, 2005 using combinations of first, last, and maiden names; zip code; birth date; and social security number. The accuracy of this linkage for identifying cancer cases among IWHS subjects has been previously estimated at more than 98%. Incident CRC cases were identified by ICD-O codes, with cancers located in the cecum, ascending colon, hepatic flexure, transverse colon, and splenic flexure (ICD-O codes 18.0, 18.2–18.5) categorized as proximal colon and cancers located in the descending colon, sigmoid colon, rectosigmoid junction and rectum (ICD-O codes 18.6, 18.7, 19.9, 20.9) categorized as distal colorectum.

Data were descriptively summarized using frequencies and percentages for categorical variables and mean and SD for continuous variables. Follow-up was calculated as the age at completion of the baseline survey until the age at first CRC diagnosis, age at move from Iowa, or age at death. If none of these events occurred, a woman was assumed to be alive, cancer-free, and living in Iowa through December 31, 2005. Cox proportional hazard regression analysis was used to estimate relative risks (RR) and 95% confidence intervals (CI) for associations between body size and incident CRC.

BMI was derived as weight divided by baseline height (kg/m²) and was analyzed by quartile distribution or by categories defined by the World Health Organization (30): underweight (<18.5), normal weight (18.5–24.9), overweight (25.0–29.9), obese I (30.0–34.9), obese II (35.0–39.9), and obese III (≥40.0). Change in BMI was calculated as baseline BMI minus BMI at age 18 years. Change in weight (kg) was similarly calculated as weight at baseline minus weight at age 18 years. Peak BMI was defined as the highest BMI observed within a period of interest (i.e., during premenopausal or postmenopausal years). WHR was derived as waist circumference divided by hip circumference. “Overweight years” (OWYs) were estimated as a novel measure of cumulative exposure to excess body size (conceptually similar to cigarette pack-years). More specifically, we calculated the deviation (in BMI units) above the upper limit of normal (24.9 kg/m²) for each age-specific BMI value; measurements below this limit were assigned an OWY value of zero units. OWY values were then multiplied by the estimated duration of exposure, using a simple midpoint interpolation method to assign years between consecutive BMI measures (i.e., BMI at age 18 years used to assign values between ages 18 and 25 years; BMI at age 30 years used to assign values between ages 26 and 35 years; BMI at age 40 years used to assign values between ages 36 and 45 years; BMI at age 50 years used to assign values between age 46 and midpoint from age 50 to age at baseline; and BMI at baseline used to assign values for the remaining years through baseline). Otherwise, continuous measures were grouped into approximate quartiles.

In all Cox regression analyses, incidence was modeled as a function of age, because age is a better predictor of risk in this cohort than follow-up years (31). Two sets of models were fit: 1 accounting for age at baseline only and 1 accounting for age at baseline plus other potential confounding factors: age at menopause (<45, 45–50, 50–55, >55 years), exogenous estrogen use (never, ever), oral contraceptive use (never, ever), smoking status (never, former, current), cigarette pack-years (continuous), physical activity level (low, moderate, high), self-reported diabetes mellitus (no, yes), alcohol (0, 0–3.4, > 3.4 g/d; cut point based on median intake for IWHS participants who reported any alcohol consumption), and quartiles of intake for total energy (kcal/d), total fat (g/d), red meat (g/d), fruits and vegetables (g/d), calcium (mg/d), folate (μg/d), and vitamin E (mg/d). When applicable, tests for trend were carried out for each categorized body size variable by ordering the values from the lowest to highest and including the resulting 1 degree-of-freedom term in the Cox model. Tests for trend based on categorical BMI were run after excluding underweight subjects. To rule out the possibility that observed associations were due to underlying subclinical disease, we fit a series of secondary analyses that excluded the first 5 years of follow-up for each individual. All statistical tests were 2-sided, and all analyses were carried out using the SAS (SAS Institute, Inc.) and S-Plus (Insightful, Inc.) software systems.

A total of 36,941 women, ages 55 o 69 years at baseline and representing 619,961 person-years, were included in the final analytic cohort. The mean (±SD) for baseline BMI was 26.98 (±5.08) kg/m². Additional baseline characteristics and body size parameters are provided in Table 1, by BMI category. In general, women with baseline BMIs in the obese III category reported lower exogenous estrogen use (ever), oral contraceptive use (ever), cigarette smoking (current), physical activity level (high), and daily intake of vitamin E and alcohol, along with higher diabetes mellitus and daily intake of total energy, total fat and red meat, than women with baseline BMIs in the normal, overweight, obese I, or obese II categories. Among women with baseline BMIs below the normal range, cigarette smoking (ever), and daily intake of total energy, total fat, and alcohol were higher and vitamin E intake was lower than in women with baseline BMIs in the normal range.

Greater baseline height, weight, BMI, hip circumference, waist circumference, and WHR were all strongly associated with increased CRC risk in age-adjusted and multivariable-adjusted risk models (P_trend ≤ 0.003 for each variable; Table 2). Overall, baseline BMI yielded the highest risk estimates. In the multivariable models, overall CRC risk increased progressively across the overweight (RR = 1.12; 95% CI = 0.99–1.28), obese I (RR = 1.31; 95% CI = 1.12–1.54), obese II (RR = 1.32; 95% CI = 1.03–1.68), and obese III (RR = 1.56; 95% CI = 1.10–2.22) categories, compared with the normal weight BMI category (P_trend < 0.001). Further adjustment for baseline WHR resulted in slight attenuation of the association between BMI and incident CRC (RR = 1.49; 95% CI = 1.04–2.12; obese III versus normal weight), whereas adjustment for baseline BMI had a more pronounced effect on the association between WHR and incident CRC (RR = 1.18; 95% = CI 0.99–1.41; quartile 4 versus quartile 1). Women in the underweight BMI category were also at increased CRC risk (RR = 1.62; 95% CI = 0.98–2.66), but the point estimate did not achieve statistical significance. When incident CRCs diagnosed during the first 5 years of follow-up were excluded, the association with underweight BMI became stronger (RR = 1.81; 95% CI = 1.06–3.09), although this observation was based on relatively few events (n = 16 CRC cases); none of the other risk estimates shown in Table 2 were appreciably altered by excluding these CRC cases (data not shown). Multivariate models were also fit that did not include self-reported diabetes mellitus as a covariate, which generally resulted in stronger body size–related CRC risks (data not shown).

By anatomic subsite, baseline BMI in the obese III category was associated with an 86% increase in distal colon cancer risk, but was not significantly associated with proximal colon cancer risk (RR = 1.86; 95% CI = 1.14–3.05; P_trend < 0.001 and RR = 1.30; 95% CI = 0.78–2.18; P_trend = 0.06, respectively), compared with the normal weight BMI category. Conversely, height was more closely associated with proximal than distal tumors (P_trend < 0.001 versus 0.04, respectively). Other baseline body size parameters were associated with higher distal than proximal CRC risks, with the exception of underweight BMI (Table 2). Further separation of distal colon and rectal cancers did not appreciably alter the subsite-specific risk estimates (data not shown).

Age-specific BMIs were associated with lower CRC risks than baseline BMI (Table 3) and only BMI at age 50 years was significantly associated with incident CRC (P_trend = 0.005). Simultaneous adjustment for each age-specific BMI did not appreciably alter these associations, with BMI at age 50 years still associated with moderately increased risk for CRC overall (RR = 1.51; 95% CI = 0.83–2.77 for obese III versus normal weight). Peak BMI during the premenopausal and postmenopausal periods was also evaluated (Table 3). For peak BMI during the postmenopausal period, comparison of obese III versus normal weight BMI yielded risk estimates of 1.52 (95% CI = 1.10–2.11; P_trend = 0.002) for any CRC, 1.64 (95% CI = 1.02–2.63; P_trend = 0.008) for distal CRC and 1.41 (95% CI = 0.89–2.24; P_trend = 0.11) for proximal colon cancer. For peak BMI during the premenopausal period, comparison of obese III versus normal weight BMI yielded risk estimates of 1.20 (95% CI = 0.75–1.93; P_trend = 0.05) for any CRC, 1.70 (95% CI = 0.95–3.05; P_trend = 0.22) for distal CRC, and 0.79 (95% CI = 0.35–1.79; P_trend = 0.16) for proximal colon cancer. After adjusting for peak premenopausal BMI, the association between peak postmenopausal BMI and overall CRC risk remained significantly increased (RR = CRC 1.57; 95% CI = 1.05–2.36; obese III versus normal), whereas the association with peak premenopausal BMI was attenuated by adjustment for peak postmenopausal BMI (RR = 0.89; 95% CI = 0.52–1.51; obese III versus normal). Simultaneous adjustment for the age- and period-specific BMI measures also resulted in attenuated proximal and distal CRC risks, similar to the patterns described for overall CRC risk above (data not shown).

Change in BMI and change in weight (from age 18 years to baseline for each measure) were both positively associated with incident CRC in age-adjusted risk models (Table 4). However, risk estimates adjusted for baseline BMI and other potential confounding variables were not statistically significant. Exclusion of incident CRCs diagnosed within the first 5 years of follow-up did not appreciably alter these associations (data not shown). Greater cumulative exposure to increased body size, as estimated by the OWY variable, was significantly associated with overall CRC risk (P_trend = 0.007) and distal CRC risk (P_trend = 0.01), but not proximal colon cancer risk (P_trend = 0.15). OWYs accrued during the premenopausal and postmenopausal periods demonstrated subsite-specific risk patterns that were similar to those based on the peak BMI data.

In this large, prospective study of older women, multiple body size measures were positively associated with incident CRC. Our findings do not support the previously proposed theory that weight status has limited, if any, influence on CRC risk among women after menopause (32). Rather, these data strongly suggest that increased body size represents a potentially modifiable CRC risk factor in this demographic subgroup. Further investigation of when increased body size has the greatest effect on CRC risk (i.e., early adulthood versus later adulthood) might also be informative, particularly with respect to defining subsite-specific pathways of colorectal carcinogenesis.

Previous studies of general obesity and CRC risk among women have typically observed weak or null associations, as recently reviewed (33–35). In a recent meta-analysis of data from 21 prospective studies, Larsson et al. (6) estimated a 12% higher colon cancer risk (RR = 1.12; 95% CI = 1.07–1.18) for every 5 kg/m² increment in women. Summary risk estimates for proximal colon (RR = 1.13; 95% CI = 0.93–1.36), distal colon (RR = 1.14; 95% CI = 1.01–1.28), and rectal (RR = 1.03; 95% CI = 0.99–1.08) cancer subsites were not statistically different. Renehan et al. (7) analyzed data from 19 and 14 prospective studies to estimate risk ratios per 5 kg/m² BMI increase of 1.09 (95% CI = 1.05–1.14) and 1.02 (95% CI = 0.99–1.04) for colon and rectal cancers, respectively, among women. Using a slightly different approach, Moghaddam et al. (5) summarized data from 9 cohort studies and found that neither overweight (BMI = 25.0–29.9 kg/m²) nor obese (BMI > 30 kg/m²) women were at significantly increased CRC risk compared with “normal” weight (BMI < 25 kg/m²) women, with risk estimates of 1.03 (95% CI = 0.96–1.10) and 1.07 (95% CI = 0.97–1.18), respectively, after correcting for publication bias. Of note, underweight (<18.5 kg/m²) women were included in the reference group, which may have diminished the investigators' ability to accurately define the true risk association. Among IWHS participants, women who were underweight at baseline experienced a similar overall CRC risk as women who were extremely obese (although the subsite-specific associations were different). Moreover, after excluding incident CRCs diagnosed during the first 5 years of IWHS follow-up, the association between underweight status and incident CRC was strengthened rather than attenuated, arguing against reverse causality or residual confounding by cigarette smoking or other risk factors. Thus, it seems plausible that a parabolic relationship (i.e., U-shaped curve) exists between body size and CRC risk, particularly proximal CRC, which may have contributed to prior reports of weak or null associations among women in other studies.

To date, relatively few prospective studies have evaluated abdominal obesity as a CRC risk factor among women (16–20). In the 2 largest studies reported to date, colon cancer risks based on comparison of extreme levels for BMI, waist circumference, and WHR were 1.06 (95% CI = 0.79–1.42), 1.48 (95% CI = 1.08–2.03), and 1.52 (95% CI = 1.12–2.05) for participants in the European Prospective Investigation into Cancer and Nutrition (EPIC) study (18) and 1.45 (95% CI = 1.02–2.07), 1.48 (95% CI = 0.89–2.46), and 1.48 (95% CI = 0.88–2.49), respectively, for participants in the among Nurses' Health Study (16). Of note, adjusting for BMI slightly attenuated the observed waist circumference- and WHR-associated colon cancer risk estimates in both studies. Among IWHS participants, baseline BMI and WHR were both positively associated with incident CRC overall, although simultaneous adjustment did attenuate the observed risk estimates somewhat. However, we did not detect an appreciable difference in CRC risk based on measures of general obesity versus abdominal obesity in our cohort.

Change in body size over time has also been evaluated as a potential CRC risk factor among women in a limited number of previous investigations. Among participants in the prospective Nurses' Health Study (16), weight gain of 20 kg or more versus stable weight, over an interval of 16 to 41 years, had no appreciable effect on colon cancer risk (RR = 1.08; 95% CI = 0.79–1.48). Another cohort study from Austria (36) found no significant difference in CRC risk for women with annual BMI increases of 0.5 kg/m² or more versus less than 0.1 kg/m² (RR = 1.31; 95% CI = 0.66–2.61). Similarly, null associations were observed for women who gained 21 kg or more versus 1 to 5 kg after age 20 years, both in a population-based study from Canada (OR = 1.13; 0.88–1.46; ref. 37) and in an international study involving Colon Cancer Family Registry participants (OR = 1.08; 95% CI = 0.80–1.47; ref. 38). In contrast, case–control studies from Wisconsin (39) and Kentucky (40) have reported statistically significant increases in colon cancer risk among women with higher adulthood weight gain (OR = 1.3; 95% CI = 1.0–1.6 for extreme quartiles in percent weight change and OR = 2.74; 95% CI = 1.27–5.92 for weight gain ≥ 10 kg/m² versus < 5 kg/m², respectively). Among IWHS participants, baseline BMI was associated with the highest CRC risks (particularly distal tumors), whereas BMI at age 50, peak BMI during the postmenopausal period, and total OWY demonstrated slightly weaker risk associations. BMI at younger ages (40, 30, 18 years) and changes in BMI or weight from age 18 years to baseline were also less predictive of CRC risk than baseline body size in our study cohort. Of note, height was strongly associated with incident CRC (particularly proximal tumors), which has been reported in other observational studies as well (18, 37, 41, 42). Because adult height is thought to reflect childhood energy intake (43), it is possible that early life nutritional factors, perhaps not adequately captured by recalled weight (or derived BMI), are relevant to CRC risk and may influence different carcinogenic pathways than later life exposure to increased body size.

A functional link between obesity and CRC risk seems biologically plausible (44), based on increased colonic exposure to circulating growth hormones, cytokines and/or energy substrates (35, 45–50). Some reports have suggested that higher estrogen levels, derived from adipose tissue, might counterbalance the procarcinogenic effects of obesity (32, 51, 52). However, data from several recent cohort studies remain inconclusive. In the United Kingdom's Million Women Study (53), incremental BMI change (per 10 kg/m²) had no apparent influence on CRC risk within the cohort at large. After stratifying by menopausal status, a positive association was observed for premenopausal women (RR = 1.61; 95% CI = 1.05–2.48), whereas a null association was observed for postmenopausal women who had never taken hormone replacement therapy (HRT; RR = 0.99; 95% CI = 0.88–1.12; ref. 53). Among postmenopausal women in the EPIC study (18), BMI was not significantly associated with incident colon cancer for either HRT users or nonusers (RR = 0.72; 95% CI = 0.31–1.70 and RR = 1.12; 95% CI = 0.75–1.67, respectively; comparison of extreme quintiles). However, waist circumference (RR = 1.68; 95% CI = 1.06–2.64; P_trend = 0.02) and WHR (RR = 1.76; 95% CI = 1.14–2.72; P_trend = 0.002) were positively associated with colon cancer risk in HRT nonusers only. In the Cancer Prevention Study-II Nutrition Cohort (54), Wang and colleagues observed similar BMI-associated CRC risks for postmenopausal women who were never-users, former-users, and current-users of HRT (RR = 1.13; 95% CI = 1.02–1.25; RR = 1.01; 95% CI = 0.87–1.18; and RR = 1.08; 95% CI = 0.92–1.27; per 5-unit increase; P_interaction = 0.50). Because IWHS subjects were all ages 55 years or more at baseline, we were unable to evaluate potential effect modification by menopausal status in our cohort, although we did find that, on the basis of self-recalled data, peak BMI and OWY accrued during the premenopausal and postmenopausal periods were associated with somewhat different subsite-specific CRC risk patterns. If confirmed by other studies, these observations could generate new insights regarding the seemingly complex relationship between body size, estrogen status, and colorectal carcinogenesis.

Major strengths of our prospective study include the extensive baseline and prebaseline body size data, comprehensive CRC case ascertainment methods, and ability to adjust for multiple potential confounding variables, which allowed us to model overall and subsite-specific CRC risks with sufficient statistical power. As was observed for IWHS participants, increased body size has been more strongly associated with distal than proximal CRC risks in other observational studies (16, 51, 55–57), supporting the consistency of our findings in this regard. Separating underweight from “normal” weight women in our Cox regression models resulted in more precise risk estimates for overweight and obese women. One limitation of our study was the reliance on self-reported height and weight measures. Recent data from the National Health and Nutrition Examination Epidemiological Follow-Up study suggest that adult self-reported weight profiles may be slightly underestimated by older women (58). If this reporting bias was present in the IWHS cohort, the observed CRC risks associated with increased body size may have been somewhat attenuated. It is also important to note that our findings were based on a relatively homogeneous subject cohort (i.e., older, predominately Caucasian women) and may not apply to other more demographically diverse groups.

In summary, results from our population-based cohort study strongly support a positive association between body size and incident CRC among older women. The CRC risk patterns observed in our study were generally similar across multiple body size parameters, with no appreciable differences in predictive capacity based on surrogate markers of general obesity versus abdominal obesity. On the basis of recent data regarding the high prevalence of overweight and obese women in the United States (2, 3), broader recognition of the relationship between increased body size and incident CRC should provide additional motivation for avoidance of excess weight gain and signal the need for clinicians to be vigilant in promoting early detection, lifestyle modification, and/or other prevention strategies in this high-risk group. Given the CRC subsite-specific risk associations observed in our study, coupled with molecularly defined subtype associations described in other recent reports (38), further evaluation of the carcinogenic pathways influenced by increased body size (including with respect to positive energy balance in earlier versus later life) may be informative.

Dr. Limburg served as a consultant for Genomic Health, Inc. from 8/12/08-4/19/10. Mayo Clinic has licensed Dr. Limburg's intellectual property to Exact Sciences and he and Mayo Clinic have contractual rights to receive royalties through this agreement. No other potential conflicts of interest were disclosed.

Grant Support

This study was funded in part by National Cancer Institute grants R01 CA39742 and R01 CA107333.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.