Dietary Fat, Fiber, and Carbohydrate Intake in Relation to Risk of Endometrial Cancer

Endometrial cancer is the seventh most common malignancy among women worldwide and the most common invasive gynecologic cancer in U.S. women, with 40,100 new cases projected to occur in 2008 (1). The role of increased levels of estrogen is well established (2, 3), and circulating estrogen levels are associated with increased risk of endometrial cancer (4, 5). Most of the risk factors for endometrial cancer can be substantially explained within the framework of the unopposed estrogen hypothesis (6), although other mechanisms such as inflammation also have been proposed (7).

Several components of diet have been proposed to influence endometrial cancer risk by modulating the production, metabolism, and excretion of endogenous hormones, especially estrogen. One in vitro study suggested that physiological diurnal elevations in plasma free fatty acids which are increased by high fat consumption, obesity, and stress may increase free estradiol, thereby exerting a promotional effect on endometrial cancer (8). In the only animal study, in Donryu rats, a high-fat diet led to an early elevation and continued high levels of the serum estrogen/progesterone ratio and an increase in the incidence of uterine adenocarcinomas (9). Glycemic load and glycemic index are used to indicate the carbohydrate content of foods according to their postprandial glycemic effects and high glycemic load and glycemic index diet are thought to result in high circulating insulin levels (10) and thereby might influence the incidence of endometrial cancer (6).

To date, evidence for an association between dietary fat and fiber and endometrial cancer is limited and results from observational studies inconsistent. Eleven of 14 case-control studies supported the hypotheses that high fat or low dietary fiber intake could increase the risk of endometrial cancer (11–21), 3 found no association (22–24). In few prospective studies, dietary fiber and most of the fatty acids were not associated with risk (25, 26). In 1 case–control (27) and 4 cohort studies (28–31), carbohydrate intake or glycemic load were positively associated with risk of endometrial cancer whereas 2 other reported no significant associations (18, 25).

In the current study, we evaluated the associations of dietary fat, fiber, and carbohydrates (including glycemic load) with risk of endometrial cancer by using 26 years of prospectively collected data from the Nurses' Health Study (NHS) cohort.

The NHS began in 1976, and included 121,701 female registered nurses aged 30–55 years who resided in 1 of 11 states in the United States at that time. The cohort has been followed biennially by mailed questionnaire to update exposure information and any new disease diagnoses; the follow-up rate has been at least 90% for each follow-up cycle. Deaths are confirmed through report by family members and the National Death Index.

In this analysis, the follow-up began in 1980 when dietary intake was first queried. At baseline, we excluded women who had hysterectomy (n = 20,612), did not respond to, or had more than 10 missing items on the 1980 food frequency questionnaire (FFQ), or total energy intakes of less than 500 or more than 3,500 kcal/day (n = 28,487), died before 1980 (n = 747), or reported any type of cancer before 1980 (excluding nomelanoma skin cancer, n = 3,660). Because obesity is an important risk factor for endometrial cancer, we also excluded women with missing body mass index (BMI) at baseline (n = 166, these women could reenter the analysis in subsequent cycles once information on BMI was available). A total of 68,070 women remained for analysis.

During each follow-up cycle, we excluded deaths, those with a diagnosis of any cancer including endometrial cancer, as well as those who had their uterus removed in the previous period. The women with missing BMI during the last 2 consecutive periods were also excluded in the next follow-up cycle but could reenter the analysis in subsequent cycles once information on BMI was available. For subsequent FFQs with about 130 items, those with more than 70 items blank food items, or total caloric intakes of less than 600 or more than 3,500 kcal/d were excluded.

From 1978 forward, participants were asked to report in their questionnaires any new diagnosis of endometrial cancer. From June 1, 1980 through May 31, 2006, a total of 1,384 women reported endometrial cancer. For 1,104 cases, medical records including the diagnosis, histologic type, presence of invasion, and stage were obtained. From these, we identified 669 cases of invasive adenocarcinoma defined by the International Federation of Gynecology and Obstetrics as stage IB to IVB. The primary reasons for exclusion were that the tumor was noninvasive (n = 316) or nonepithelial (n = 60) or a type of epithelial cancer (n = 58) other than adenocarcinoma (e.g., clear cell).

A 61-item FFQ for collecting dietary information was administered in 1980. An expanded FFQ with approximately 130 food items was sent to women in 1984, 1986, 1990, 1994, 1998, and 2002 to assess usual food intakes in the previous year. A common unit or portion size for each food was specified, and participants were asked how often, on average, they had consumed that amount of food or beverage during the previous year. The average daily intake of each nutrient was calculated by multiplying the frequency of consumption of each item by its nutrient content per serving and totaling the nutrient intake for all food items.

Values for total, animal, and vegetable fat and specific fatty acids were computed primarily from the USDA database (32). Association of Official Analytical Chemists (AOAC) fiber (the enzymatic gravimetric method), the standard total dietary fiber method in the United States, was used to calculate dietary fiber. Total fiber, fiber from cereals and separately from vegetables, also was calculated. Glycemic load values were calculated by multiplying each food's glycemic index by its carbohydrate value and by the frequency of consumption, then summing over all foods (33).

Dietary variables were energy adjusted as previously described (34). This approach addresses whether the composition of the diet, independent of total energy intake, is most relevant to the risk of endometrial cancer. Given dietary intake was assessed up to 7 times over a 26-year period, we assessed intake in several ways: cumulative average intake (the average intake from the available FFQs up to each follow-up cycle), recent intake (current intake, at the beginning of a 4-year interval), and baseline intake were used to assess the effect of cumulative exposure, short latency, and long latency, respectively. If intake on 1 or more questionnaires was missing, the cumulative average intake was calculated by averaging the available data. For recent intake, we carried forward the last available dietary data, and if data from more than 1 questionnaire was missing, recent intake was considered missing. All the dietary data were used prospectively.

In addition, in secondary analyses, we assessed nonenergy adjusted nutrient intake, the percentage of energy from fat and carbohydrates, and grams of fiber per 1,000 kcal, all in relation to endometrial cancer risk.

Information on most potential confounding factors, including weight, smoking, oral contraceptive (OC) use, postmenopausal hormone use (PMH), age at menopause, age at last birth, hypertension, and diabetes was collected in 1976 and subsequent questionnaires. If data were not available, those women were assigned to a missing category for that period. BMI was calculated from height at baseline and from the biennially updated report of current weight. We carried forward the weight reported in the prior questionnaire cycle if it was missing in the current cycle. A nurse was classified as postmenopausal from the time she returned a questionnaire reporting natural menopause. PMH was first assessed in 1976; women were queried about current and past postmenopausal hormone therapy use and duration. From 1978, information on the type of hormone used was collected. OC use was queried biennially until 1984 when women were 38 to 63 years of age and few women were still current users. Neither alcohol intake nor physical activity were confounders in these analyses, and hence were not included in the final multivariable models.

Each participant contributed person-time from the date of the return of the 1980 questionnaire through June 1, 2006, hysterectomy, death, loss to follow-up, or diagnosis of endometrial cancer or other cancer, whichever came first. Incidence rates of endometrial cancer in each category of the exposure variable were calculated as the number of incident cases divided by the total person-time at risk. Incidence rate ratios were computed as the ratio of the incidence rate in the exposure category of interest to the incidence rate in the referent category. To adjust the RRs for multiple covariates, we used Cox proportional hazard models conditioned on age (months) and follow-up cycle. In all multivariate models, we included the following covariates (see Table 3 for detail on how covariates were controlled in multivariate models): total energy, smoking, OC use, PMH use, age at menopause, parity, age at last birth, age at menarche, hypertension, diabetes, and BMI. In secondary analyses, to show the impact of not accounting for total energy intake in multivariable models, we present the association between macronutrients that were not energy adjusted with disease risk. Because there was no prior rationale for specific cut points for intakes of fat or dietary fiber, we categorized exposure in quintile categories to maintain an adequate number of participants in each category. Tests for linear trend for the nutrients were calculated by including the median for each quintile in the final multivariate model.

To assess whether the relationships between fat and dietary fiber intake and endometrial cancer risk varied across categories of other risk factors, we conducted stratified analyses. Further, interaction terms were calculated as the products of a binary stratification factor (e.g., menopausal status) and the median for each quintile of our dietary exposures (modeled continuously) and evaluated by using the likelihood ratio test.

A total of 669 cases of invasive adenocarcinoma were identified among about 1.3 million person-years of follow-up. Characteristics of the study population at the midpoint (1990) are summarized in Table 1. In 1990, women with high total fat intake were more likely to be obese and diabetic, and were less likely to be postmenopausal or to use PMH. Women with high fiber intake were less likely to smoke, to have used OCs or to be hypertensive, and were more likely to use PMH. Finally, women with high carbohydrate intake were less likely to be obese and ever smokers, and had a lower prevalence of OC use and diabetes.

Table 2 shows the Spearman correlations of the cumulative averaged intake of dietary factors. The upper part of the table displays correlations for energy-adjusted dietary factors and the lower part displays correlations between nonenergy-adjusted values. All the correlation coefficients are statistically significant (P < 0.05). For the energy-adjusted values, the correlation coefficients for total fat with the fat subtypes ranged from 0.17 to 0.95, and the correlation coefficient with carbohydrates is −0.76. Saturated fat and monounsaturated fat also were highly correlated with animal fat (r = 0.90, 0.75, respectively). Similarly, total fiber was also highly correlated with vegetable fiber and cereal fiber (r = 0.70, 0.51 respectively).

The RRs according to quintiles of cumulative intake of total energy, energy-adjusted dietary fat, fiber, and carbohydrates are shown in Tables 3–5. Although we observed significantly elevated risk for the fourth quintile of total energy intake compared with the lowest quintile (RR = 1.44, 95% CI = 1.13–1.84), neither the top versus bottom category comparison nor the trend test for total energy intake was significant. No increased risk of endometrial cancer was observed with higher intakes of total fat or vegetable fat or specific fatty acids. In fact, after adjusting for other risk factors, intakes of total and animal fat were associated with slightly lower risk. The RR in the top versus bottom quintile RR of total and animal fat were 0.78 (95% CI = 0.60–0.99; P_trend = 0.18) and 0.84 (95% CI = 0.65–1.08; P_trend = 0.07; Table 3). Saturated and monounsaturated fats, which are highly correlated with animal fat, had RRs close to those of animal fat but the trend tests were not significant. However, when we looked at the absolute values of animal fat without adjusting for total energy (Table 3, last column), we observed a nonsignificant increased risk. After adjusting for total energy in the multivariate model, nonenergy-adjusted animal fat intake had RRs similar to those observed for energy-adjusted values (data not shown).

We did not observe the hypothesized decreased risk associated with fiber intake (Table 4). Energy-adjusted and nonenergy-adjusted values of dietary fiber had similar RRs. In fact, high intakes of total fiber and several types of fiber, particularly cereal fiber, were associated with an elevated risk of endometrial cancer. The RR in the top (vs. bottom) quintile of total, vegetable, and cereal fiber were 1.21 (95% CI = 0.94–1.57), 1.21 95% CI = 0.94–1.55), and 1.38 (95% CI = 1.07–1.79), respectively. Similar borderline positive associations were observed with intakes of either total carbohydrates or glycemic load. The RR in the highest (vs. lowest) quintile of carbohydrates and glycemic load were 1.29 (95% CI = 1.00–1.67) and 1.29 (95% CI = 0.99–1.67; Table 4). As expected, given the positive correlation between fiber and total energy intake, fiber intake that was not energy adjusted had stronger positive associations with endometrial cancer (top vs. bottom quintile RR for total fiber = 1.54, 95% CI = 1.19–1.99; P_trend = 0.003). This association was attenuated by further adjusting for carbohydrate intake (RR for the top vs. bottom quintile decreased to 1.24; 95% CI = 0.88–1.75), and in the same model, the RR of nonenergy-adjusted carbohydrate for the top versus bottom quintile decreased from 1.51 to 1.36 (95% CI = 0.96–1.92).

When we assessed baseline and recent intake, instead of cumulative average intake, results were similar but slightly attenuated. Findings for latency analyses also were very similar. Therefore only results for the cumulative average intakes are shown.

When stratified by menopausal status, the association between energy adjusted animal fat and endometrial cancer was seen primarily in the small group of premenopausal women (top vs. bottom quintile RR = 0.36, 95% CI = 0.17–0.76; P_trend = 0.0009; Table 5). Total and monounsaturated fats had similar but weaker associations (data not shown). No association was observed among postmenopausal women (P_{heterogeneity} = 0.0008). Carbohydrate intake was positively associated with risk among premenopausal women (top vs. bottom quintile RR = 2.87, 95% CI = 1.38–6.08) but not among postmenopausal women, although the test for heterogeneity was not significant (P_{heterogeneity} = 0.94; Table 5). Glycemic load had similar but weaker associations (data not shown). We also evaluated carbohydrates and animal fat in the same model to assess their independent association with risk. However, because these macronutrients were highly inversely correlated, CIs became wider and were no longer statistically significant. The RRs for the top versus bottom quintiles of animal fat and carbohydrates were 0.47 (95% CI = 0.18–1.23) and 1.57 (95% CI = 0.60–4.10), respectively. When assessed by menopausal status, the positive association between cereal fiber was observed primarily among postmenopausal women, but this difference was marginally significant (P_{heterogeneity} = 0.05; Table 5). We also stratified by BMI (BMI < 30 vs. BMI ≥ 30) and smoking (never vs. ever), but our findings for nutrients and endometrial cancer did not vary substantially across these exposures.

Finally, we assessed percentage of energy from total fat and from carbohydrates, as well as total fiber (g/1,000 kcal) in relation to endometrial cancer risk (Figs. 1–3). The percentage of energy from total fat was associated with a nonsignificantly lower risk of endometrial cancer (top vs. bottom category RR = 0.87, 95% CI = 0.65–1.15, P_trend = 0.13). Animal fat had similar associations (data not shown). The percentage of energy from carbohydrates was positively associated with risk (top vs. bottom category RR = 1.30, 95% CI = 0.98–1.73, P_trend = 0.03). Total fiber was associated with a significantly increased risk (top vs. bottom category RR = 1.26, 95% CI = 0.98–1.62, P_trend = 0.02), although this association was attenuated and became nonsignificant after further adjustment for percentage of energy from carbohydrates. After stratification, similar to what was seen in Table 5, the associations of cancer risk with fat and carbohydrates were stronger among premenopausal women (data not shown).

In this prospective study, intakes of most dietary fats and carbohydrates were not clearly related to the risk of endometrial cancer overall, although a possible positive association with carbohydrate intake was suggested. In contrast to our hypothesis, total fiber and cereal fiber were modestly positively associated with risk, but these associations were no longer significant after controlling for carbohydrate intake. Our data also suggest that some of these associations may vary by menopausal status.

In the majority of case–control studies, dietary fat, or animal food intake has been positively associated with risk of endometrial cancer (11–19, 35, 36). Several case–control studies reported no association (22–24) or a weak inverse association with monounsaturated fat intake (24). In 1 of 2 other cohort studies to assess these associations, the Canadian National Breast Screening Study, most macronutrients were not related to the risk of endometrial cancer, but a decreased risk was observed with animal fat intake (RR in highest quartile = 0.60, 95% CI = 0.40–0.90; 25) among postmenopausal women. In another U.S. cohort study among postmenopausal women with 216 cases, dietary intake of most animal foods was not related to or was only weakly related to risk (26). In a recent meta-analysis, a positive association with total, saturated, and animal fat was observed in case–control studies, however, the limited available cohort data did not support these associations (12).

Although our study found little association overall between glycemic load/carbohydrates and cancer risk, carbohydrate intake was significantly positively associated with risk among premenopausal women. Several other studies have reported modest positive associations (27–31) as well. In the prospective Iowa Women's Health Study, a positive association was observed between glycemic load and risk (top vs. bottom quintile RR = 1.46, 95% CI = 1.02–2.08; P_trend = 0.02) among nondiabetic postmenopausal women (29). The Canadian National Screening Study reported that glycemic load, but neither carbohydrates nor the glycemic index, was positively associated with risk among premenopausal (RR in highest quartile = 1.55, 95% CI = 1.05–2.29) and obese women (RR in highest quartile = 1.88, 95% CI = 1.08–3.29; ref. 31). The Swedish Mammography Cohort Study found RRs comparing extreme quartiles of 1.90 (95% CI = 0.84–4.31) for carbohydrate intake and 2.99 (95% CI = 1.17–7.67) for glycemic load among overweight women with low physical activity (30). In the European Prospective Investigation into Cancer and Nutrition cohort study, although total carbohydrates and glycemic index/load were not associated with endometrial cancer overall, total carbohydrate intake was associated with increased risk among postmenopausal women (RR in highest quartile = 1.26, 95% CI = 0.95–1.65; ref. 28). In an Italian hospital-based case–control study, glycemic index and glycemic load were significantly associated with elevated risk of endometrial cancer and the associations were stronger in older women, in those with higher BMI, and in PMH users (27). Although several other studies found no associations between carbohydrates and endometrial cancer (18, 25), results from the above studies suggested glycemic load/carbohydrates might have modest positive associations with endometrial cancer risk which are generally consistent with the positive association we observed among premenopausal women.

In our study, most fat subtypes had no significant associations with endometrial cancer except for total and animal fat which were associated with a possible nonlinear decreased risk in the small group of premenopausal women. We also observed a nonlinear increased risk associated with carbohydrate intake among premenopausal women. Because of the strong inverse correlation between these 2 macronutrients, it is hard to distinguish their effects and thus these associations need to be assessed in even larger studies. The biological mechanisms for these possible associations are unclear. High-glycemic-index diets are thought to be associated with hyperinsulinemia and decreased insulin sensitivity (6). Higher circulating plasma insulin levels have been associated with lower sex hormone-binding globulin production, leading to higher free estrogen levels (37–41). Among premenopausal women, insulin provides a key stimulus to ovarian androgen synthesis, especially among women with polycystic ovary syndrome, and therefore, by inducing anovulation and progesterone deficiency (6), may increase risk of endometrial cancer. To our knowledge, no prior prospective studies have assessed insulin levels and risk of endometrial cancer in premenopausal women. A case–control study nested within the European Prospective Investigation into Cancer and Nutrition found that women with elevated serum levels of C-peptide had a modestly increased risk of endometrial cancer, suggesting hyperinsulinaemia may influence endometrial cancer risk among both pre- and postmenopausal women (42). In a prospective case–cohort study, a positive association was found between serum insulin level and risk of endometrial cancer among postmenopausal women (43).

Dietary fiber was proposed to have an inverse association with endometrial cancer, possibly by increasing the fecal excretion of estrogens and thereby reducing plasma estrogen levels (44, 45). However, in our study, we did not observe inverse associations with total dietary fiber or fiber subtypes. Cereal fiber had a stronger positive association with endometrial cancer risk among postmenopausal women, but there was no clear trend. The results from previous case–control studies and cohort studies are inconsistent. Most of the case–control studies supported the hypothesis that dietary fiber decreased the risk of endometrial cancer (13, 14, 17, 21, 23). In a metaanalysis, an inverse association between fiber intake and risk was observed (RR = 0.71, 95% CI = 0.59–0.85) comparing the top versus bottom dietary fiber intake) from eight case–control studies (46). In contrast, in the only prospective study that evaluated this association, no association was noted (25, 46).

In our secondary analysis, associations of percent of energy from fat/carbohydrates and grams of fiber per 1,000 kcal with risk were similar to those of energy-adjusted factors. The nonsignificant or borderline significant associations of nonenergy-adjusted intakes were attenuated after adjusting for total energy in the multivariate models, suggesting that if there is any association of nonenergy-adjusted intakes and cancer risk, it is likely because of total energy intake.

In our large prospective cohort study, we collected repeated dietary assessments over 2 decades. The prospective nature and high follow-up rate minimized recall and selection biases. The repeated dietary assessments reduce measurement error, which should result in more accurate estimates, although some nondifferential misclassification is unavoidable. Residual confounding is minimized in this study, as potential risk factors are tightly controlled, and adjusting for covariates changed the results only slightly. However, caution is warranted in interpreting these findings, because of the small number of premenopausal cases (n = 93). In multivariable models, we adjusted for BMI, hypertension, and diabetes, factors that could be on the causal pathway linking dietary factors with endometrial cancer, thus raising the possibility of statistical over adjustment. However, the composition of the diet, independent of its influence on energy balance, was of primary interest in this analysis hence accounting for other factors related to energy balance was appropriate. Also of note, in our previous analyses, the percentage of calories from fat was only weakly associated with weight gain (47) and, in several randomized trials of weight loss, those assigned a low-fat diet did not lose more weight than those assigned other types of reduced calorie diets (48, 49). Also, adjusting for hypertension and diabetes did not substantially alter our results.

In summary, the findings in this prospective study did not support the hypothesis that dietary fat is an important cause of endometrial cancer. Although we found no overall associations between dietary fat, fiber, or carbohydrates with endometrial cancer, several of the associations seemed to vary by menopausal status and these warrant evaluation in other prospective studies.

No potential conflicts of interest were disclosed.

The work was supported by NIH (grant number CA87969).

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.