Risk of Endometrial Cancer in Relation to Medical Conditions and Medication Use

Endometrial cancer is the fourth most common cancer among women of westernized countries and the most common malignancy of the female genital tract (1). Obesity is strongly related to endometrial cancer, probably through the increased production of unopposed endogenous estrogens. Type 2 diabetes mellitus is also related to increased risk, although its independence from body mass index (BMI) remains somewhat unclear. Both obesity and type 2 diabetes mellitus facilitate a proinflammatory cellular milieu (2). Experimental evidence has shown that nuclear factor-κB, cyclooxygenase 2, and prostaglandin E 2 are overexpressed in endometrial cancer cells, further suggesting an implication of inflammation in endometrial carcinogenesis (3-5). The relation of other obesity-related conditions such as hypertension and dyslipidemia with endometrial cancer is less clear, and findings could be related to the uncontrolled effect of obesity (6-8). Consistent with the inflammatory theory, two epidemiologic studies found that regular aspirin use—a common anti-inflammatory—was related to a decreased risk of endometrial cancer among obese women (9, 10).

In a population-based case-control study among U.S. women, we assessed the relation of medical conditions and their treatments with endometrial cancer.

The Estrogen, Diet, Genetics, and Endometrial Cancer (EDGE) Study is a population-based case-control study conducted in six counties in northern New Jersey. Cases were eligible if they were ages 21 y and over, lived in one of the six counties (Bergen, Essex, Hudson, Middlesex, Morris, and Union) at the time of diagnosis, spoke English or Spanish, and were physically and cognitively able to take part. They were diagnosed with invasive epithelial endometrial cancer from July 1, 2001 to June 30, 2005. Cases were identified by the New Jersey Department of Health and Senior Services using rapid case ascertainment supplemented with a review of state Cancer Registry data to identify cases diagnosed out of the area. Pathology reports and slides were obtained for cases who consented to be in the study and sent to the study pathologist for review. All epithelial histologic types are included in this analysis; 90% had tumors classified as endometrioid or predominantly endometrioid. During the four years of the study, 1,559 eligible women were identified, of whom 1,108 could be contacted within 1 y of diagnosis. A total of 469 completed the interview. The major reason for not completing the interview was refusal. In comparison with all women diagnosed with endometrial cancer in these counties in this time period, the cases included were younger and more likely to have localized disease. The mean age of those included was 61.7 y (median, 61 y; range, 33-88 y), compared with 63.6 y in all women. Our case group included 11% who were age ≥75 y, a smaller proportion than the 20% in all cases. Seventeen percent of the women included in EDGE had Surveillance, Epidemiology and End Results (SEER) stage classified as regional or distant, compared with 26% in all cases. These imbalances are likely to be due to refusal, language barriers, illness, or death among the older and more seriously ill women.

Controls had the same eligibility requirements as cases and, in addition, could not have had a hysterectomy. Three methods were used to locate controls. Women age <65 y were located through random digit dialing. Using names and phone numbers from a commercial research service that we contracted, we reached 355 eligible women, of whom 175 (49%) completed the interview and the remainder declined. Initially, women age ≥65 y were identified through lists purchased from the Centers for Medicare and Medicaid (CMS); they were contacted by letter, followed by telephone calls when we were able to locate the numbers. We identified 316 women, of whom 68 (22%) completed the interview, whereas the remainder declined. Because many telephone numbers were not available from CMS, in August 2003 we began using area sampling to reach older controls. Initially we sought women age ≥65 y; later we included women age ≥55 y. In randomly selected areas within the six counties, we identified 30 consecutive households according to postal route. Each selected household was sent a letter by mail introducing the study, then visited by interviewers to ascertain eligibility and interest in participating. We identified 524 eligible women, of whom 224 (43%) completed the interview. In total, 467 controls completed the interview, including 175 from random digit dialing, 68 from CMS, and 224 from area sampling. The study was approved by the Institutional Review Boards at Memorial Sloan-Kettering Cancer Center and the New Jersey Department of Health and Senior Services.

For random digit dialing and CMS controls, we frequency-matched controls to the expected distribution of cases by 5-year age groups; in area sampling, however, this would have required a complex sampling scheme that would have been difficult to execute in the field. Instead, we included all age-eligible women who were willing to take part in the study. We were able to enroll more older controls, leading to an imbalance in age between cases and controls. Area sampling has the potential to introduce bias because women living in the same neighborhood are likely to have similar characteristics. The 224 women recruited through area sampling lived in 150 neighborhoods, with the number from any one neighborhood ranging from 1 (in 90 neighborhoods) to 7 (in one neighborhood).

The interview covered established and possible risk factors for endometrial cancer. These included reproductive history, menstrual history, and oral contraceptive and hormone replacement therapy use. BMI [wt (kg)/ht (m)²] was ascertained at the reference date (date of diagnosis for cases and 6 mo before the interview for controls). Participants were specifically asked if they had ever been told by a health professional that they had a number of medical conditions, including those related to BMI (type 2 diabetes mellitus, high cholesterol, hypertension, osteoporosis or thinning bones, and fibroids in the uterus) and other conditions (bladder or urinary infections, low blood count or anemia, depression, migraine headaches, ovarian cysts, endometriosis, rheumatoid arthritis, hyperthyroidism or Graves' disease, and Crohn's disease). Age at first diagnosis was recorded, as well as what medications, if any, were taken for the conditions reported. Participants were also asked about other medications they may have used continuously for at least 6 mo. Total duration of use was recorded for each medication. We considered only conditions and medications used before the reference date.

The effect of duration of the disease was assessed by categorizing duration as shorter or longer than the median time since diagnosis for each particular disease among the controls.

Medications were categorized as follows. Type 2 diabetes mellitus–related medications were biguanides (e.g., metformin), insulin, and sulphonylureas (e.g., glibenclamide, repaglinide). Hypercholesterolemia medications were statins (e.g., atorvastatin, simvastatin) and fibrates (e.g., bezafibrate). Hypertension medications were angiotensin-converting enzyme inhibitors (e.g., ramipril, captopril), angiotensin 2 receptor antagonists (e.g., candesartan, telmisartan), β blockers (e.g., atenolol, bisoprolol), calcium channel antagonists (e.g., verapamil, nifedipine), thiazide diuretics (e.g., hydrochlorothiazide, indapamide), loop diuretics (e.g., furosemide, torasemide), and potassium sparing diuretics (e.g., spironolactone). Osteoporosis medications were biphosphonates (e.g., pamidronic acid, alendronic acid) and calcitonin. Pain and inflammation medications were nonsteroidal anti-inflammatory drugs (NSAID; e.g., aspirin, ibuprofen, diclofenac), analgesics (e.g., paracetamol, phenacetin), and corticosteroids (e.g., cortisone, prednisone). Mood disorder medications were serotonin selective reuptake inhibitors (e.g., paroxetine, fluoxetine) and benzodiazepines (e.g., diazepam, lorazepam). Gastrointestinal medications were proton pump inhibitors (e.g., omeprazole, pantoprazole). Duration of medication use was categorized based on years of use among the controls, using the median or tertiles, depending on the number of women who used each medication.

Unconditional logistic regression was used to estimate odds ratios (OR) and 95% confidence intervals (95% CI) as the measure of association between specific variables and the occurrence of endometrial cancer. In logistic regression, we used a spline technique that modeled the association between age and case-control status separately according to three age categories: <65, 65 to <80, and ≥80 y (11). To investigate the possible effects of clustering by neighborhood in area sampling on the results, we compared results from the logistic models with those obtained by fitting generalized estimating equations (12), which accounts for the clustering of subjects. Because differences in parameter estimates and standard errors were trivial, indicating that the clustering did not affect the results, we used logistic models.

In the results, ORs adjusted for age and BMI are shown. In one analysis, we considered conditions related to obesity (type 2 diabetes mellitus, hypercholesterolemia, hypertension, osteoporosis, and uterine fibroids) and medications used to treat them. For this analysis, we present results from a second model, adjusted for other factors that have been related to increased risk of endometrial cancer: age at menarche, parity, menopausal status at index date, oral contraceptive use, hormone replacement therapy use, and family history of endometrial cancer. This model (the “fully adjusted model”) was also adjusted for the conditions and medications.

Whenever numbers allowed, confounding by indication was specifically addressed by restricting the analyses to individuals with the condition for which the drug is generally prescribed. Additionally, direct comparison of ORs found for different medications usually prescribed for the same indication (e.g., two different types of diuretics for hypertension) was also used to assess the possibility of confounding by indication. Stratification for BMI (<30 and ≥30) was conducted for obesity-related conditions and medications, and for NSAIDs, to assess possible effect modification by obesity. These models were also fully adjusted and included BMI as a continuous variable.

In analyses for other conditions and medications that are not strongly related to obesity, two models were used, one adjusting for age and BMI and another adjusting for these variables and other endometrial cancer risk factors as listed above.

The categories for variables included in logistic regression were as follows: race [white, black, Hispanic (any race), other]; education (high school graduate or less, some college, graduate school); smoking (never smoked at least one cigarette per day for 1 y, past, current); BMI (<25, 25-<30, 30-<35, ≥35); age at menarche (<12 versus ≥12 y); parity (0-1, 2, ≥3); menopausal status [premenopausal, early menopause (<50), late menopause (≥50), post menopause, age unknown]; oral contraceptive use (ever, never); hormone replacement therapy use (never, unopposed estrogen only, ever combined therapy); and history of endometrial cancer in first-degree relative (yes, no). Women were considered postmenopausal if their last menstrual period occurred more than 1 y before the reference date. Women who had a “natural” menstrual period (i.e., had not taken hormones or oral contraceptives) within the year were considered premenopausal, whereas others were considered premenopausal if age <55 and postmenopausal if older.

The data were analyzed using SAS version 9.1 (SAS Institute). Due to the exploratory nature of the analysis, we did not adjust for multiple comparisons. The results were evaluated at a level of significance α = 0.05.

Table 1 shows the sociodemographic characteristics and endometrial cancer risk factors of the 469 cases and 467 controls. Several differences between cases and controls existed. For instance, cases were younger than controls. As expected, cases were also more obese, had had menarche earlier, and had fewer offspring after adjustment for age.

Table 2 shows the associations of endometrial cancer with BMI, conditions related to obesity (type 2 diabetes mellitus, hypercholesterolemia, hypertension, osteoporosis, and fibroids) and their treatments. For these conditions and medications, the results of two models are shown: ORs adjusted for age and BMI; and ORs adjusted for age, BMI, risk factors shown in Table 1, and other conditions and drugs shown in Table 2 (fully-adjusted model). Obesity was strongly related to increased risk of endometrial cancer in both models (P for trend < 0.001). The age- and BMI-adjusted risks for type 2 diabetes mellitus and hypertension were only marginally significant (OR type 2 diabetes mellitus, 1.4; 95% CI, 0.9-2.3; OR hypertension, 1.2; 95% CI, 0.9-1.6). After full adjustment, no association was seen for hypertension (fully adjusted OR, 0.9; 95% CI, 0.6-1.5) but a marginally significant relationship still existed for type 2 diabetes mellitus (fully adjusted OR, 1.9; 95% CI, 0.9-4.1). Hypercholesterolemia was also associated with a marginally significant decreased risk of endometrial cancer (OR, 0.7; 95% CI, 0.4-1.0). Osteoporosis was related to a significantly lower risk of endometrial cancer in the age- and BMI-adjusted models, and to a somewhat lower risk in the fully adjusted models (OR, 0.7; 95% CI, 0.4-1.2). Analysis of type 2 diabetes mellitus stratified by BMI (<30 and ≥30) was carried out to assess the possibility of effect modification by obesity. Although no statistically significant interaction with BMI was seen, a previous history of diabetes was associated with a somewhat larger risk of endometrial cancer among women with BMI ≥30 (OR, 2.2; 95% CI, 0.7-7.2) than among women with a BMI<30 (OR, 1.8; 95% CI, 0.6-5.6; data not shown).

Endometrial fibroids were significantly associated with an increased risk of endometrial cancer in all models (fully adjusted OR, 1.8; 95% CI, 1.2-2.5). Women who had been diagnosed more than 15 years ago showed a significantly increased endometrial cancer risk in the fully adjusted model (OR, 2.0; 95% CI, 1.2-3.2; data not shown).

Two medications used for hypertension, angiotensin 2-converting enzyme inhibitors and thiazide diuretics, were related to endometrial cancer after adjustment for age and BMI (OR, 1.6; 95% CI, 1.1-2.5; and OR, 1.7; 95% CI, 1.2-2.5, respectively; Table 2). After full adjustment, only use of thiazides remained a statistically significant risk factor (OR, 1.8; 95% CI, 1.1-3.0). Duration of thiazide use was also assessed: only the longest duration of use (i.e., >6 years) was significantly related to an increased risk of endometrial cancer (fully adjusted OR, 2.4; 95% CI, 1.2-4.8). Long-term use of angiotensin 2-converting enzyme inhibitors (i.e., >5 years) was not significantly associated with risk (data not shown). An alternative analysis restricted to subjects with hypertension (221 cases, 173 controls), the condition for which thiazides are primarily used, consistently showed a significantly increased risk among users of thiazides (OR, 1.7; 95% CI, 1.0-3.0 for fully adjusted model; data not shown). The other medications used to treat hypertension (i.e., angiotensin 2-converting enzyme inhibitors, β blockers, calcium channel blockers, angiotensin 2 receptor antagonists, loop diuretics, and K sparing diuretics) were not associated with an increased risk of cancer in this restricted analysis. In analysis stratified by BMI (<30 and ≥30), thiazide diuretic use seemed to be more strongly related to increased risk of endometrial cancer among those with lower BMI (OR, 2.2; 95% CI, 1.1-4.5; and OR, 1.2; 95% CI, 0.5-2.9 for those with lower and higher BMI, respectively), although differences were not statistically significant (data not shown).

Use of biphosphonates or calcitonin was related to somewhat lower risk of endometrial cancer (OR, 0.7; 95% CI, 0.4-1.2) for each medication (fully adjusted model). Longer duration of use of biphosphonates (i.e., >2 years) was related to a somewhat lower risk of endometrial cancer than shorter use (OR >2 years, 0.6; 95% CI, 0.3-1.1; and OR ≤ 2 years, 0.9; 95% CI, 0.4-2.1; fully adjusted models; data not shown). Duration of use of calcitonin was not related to a lower risk of cancer (OR ≤5 years, 0.7; 95% CI, 0.4-1.5; and OR >5 years, 0.6; 95% CI, 0.3-1.3; data not shown). An alternative, fully adjusted analysis restricted to subjects with osteoporosis (51 cases and 102 controls), the condition for which biphosphonates and calcitonin are used, showed no modification of risk for calcitonin (OR, 0.9; 95% CI, 0.3-2.7) or biphosphonates (OR, 0.8; 95% CI, 0.3-2.2), but the numbers of exposed subjects were low. In an analysis stratified by BMI, calcitonin use was significantly associated with a decreased risk of endometrial cancer among subjects with a BMI ≥30 (OR, 0.2; 95% CI, 0.1-0.7) but not among those with a BMI <30 (OR, 1.0; 95% CI, 0.5-1.8; data not shown).

Table 3 shows the risks of endometrial cancer associated with several other medical conditions. Anemia was significantly associated with a decreased risk of endometrial cancer in the BMI- and age-adjusted model (OR, 0.6; 95% CI, 0.5-0.9). Further adjustment for risk factors did not significantly modify this association (data not shown). Reduced risk associated with anemia was the same for those diagnosed ≤34 years ago and those diagnosed more recently (data not shown).

Table 4 shows the associations of endometrial cancer with the reported use of other commonly used medications. Only NSAIDs (OR, 0.7; 95% CI, 0.5-0.97, adjusted for age and BMI) were significantly associated with a decreased risk of endometrial cancer. However, no indication of a duration-response trend was observed after adjustment for age and BMI (P = 0.12; data not shown). When we separately assessed aspirin and non-aspirin NSAIDs, a similar effect on endometrial cancer risk was observed for both types of drugs in the age- and BMI-adjusted model although neither was statistically significant (OR aspirin, 0.7; 95% CI, 0.4-1.0; OR non-aspirin NSAIDs, 0.8; 95% CI, 0.5-1.3). In analyses stratified by BMI, the risk associated with use of any NSAIDs or aspirin was similarly reduced in each stratum. For non-aspirin NSAIDs, reduced risk was somewhat more apparent in women with lower BMI (OR, 0.7; 95% CI, 0.4-1.4) than in those with higher BMI (OR, 0.9; 95% CI, 0.4-1.9), after adjustment for age and BMI (data not shown).

We found a strong association between obesity and an increased risk of endometrial cancer, consistent with the conclusion of the recent review by the World Cancer Research Fund/American Institute for Cancer Research that body fatness is a convincing cause of this disease. In that meta-analysis, obesity was shown to increase the risk of endometrial cancer by about 50% per 5 units of BMI (i.e., kg/m2) in both cohort and case-control studies, with a good deal of heterogeneity in the size of the effect (13). On this scale, our results were a 66% increase in risk per 5 units of BMI, higher than reported in most other studies. Our results showed that some other conditions related to obesity also were associated with risk. As in other studies (summarized in Saltzman, ref. 14, reviewed in Friberg, ref. 15), risk estimates for type 2 diabetes mellitus were above unity, although nonsignificantly. The association of hypertension with risk is less consistent and, in our study and others, not independent of obesity (6-8). In our study osteoporosis was related to a somewhat lower risk of endometrial cancer (fully adjusted OR, 0.7; 95% CI, 0.4-1.2), similar to what others have observed (16-18). A previous diagnosis of endometrial fibroid tumors was related to an increased risk of endometrial cancer. A recent paper (19) found increased risk for endometrial cancer in women with fibroid tumors using record linkage methods, although in that study the risk was not significantly increased among women with follow-up of ≥5 years. Because women with hysterectomy were ineligible for our study, and fibroids are the major reason for hysterectomy (20), women in our study were likely to have relatively small or less symptomatic fibroids. There is likely to be confounding by estrogen levels in our study, because both osteoporosis (21) and fibroids (22) are related to estrogen levels. BMI and other variables included in our models do not completely control for estrogen level (23).

The other condition found to be significantly related to risk was anemia, with controls more likely than cases to report this condition. To our knowledge this has not been reported before. In this study we have not been able to rule out the possibility of anemia acting as a marker of irregular menstruation or menorrhagia that could be related to a modified risk of endometrial cancer mediated by alterations in estrogen exposure. Confirmation of this finding in further studies is needed before a relationship can be established.

In studying medications, we found an elevated risk of endometrial cancer for thiazide use after adjustment for age, BMI, risk factors, and other conditions and treatments related to obesity (i.e., in the fully adjusted model). To our knowledge, this association has not been previously reported. An analysis restricted to hypertensive subjects eliminated to a large extent the concern about confounding by indication for thiazides. Also, the fact that other medications used to treat hypertension (i.e., other diuretics, calcium channel blockers, etc) were not related to endometrial cancer in the overall and the restricted analyses suggests a causal relationship for thiazides. Risk increased with longer duration of use. Studies of the prototypic thiazide hydrochlorothiazide in rats showed no carcinogenic effects (24). Some studies have suggested that thiazides could increase the risk of type 2 diabetes mellitus by increasing insulin resistance (reviewed in Sowers, ref. 25), possibly explaining the elevated risk for endometrial cancer, although other, well-designed studies have not confirmed this association (26, 27). Several studies have found chronic use of thiazides to be related to a decreased risk of hip fractures (28) and to a reduced risk of radial bone loss (i.e., osteoporosis; ref. 29). Hip fractures are related to osteoporosis, a hormone-dependent condition caused by a reduction in estrogen levels. In postmenopausal women, administration of estrogens prevents osteoporosis and fractures (30). An elevated risk of endometrial cancer could be explained if thiazides had an estrogen-like activity. Further research is needed to confirm this hypothesis. Thiazides are usually selected as the initial treatment for hypertension, especially among subjects without other risk factors, such as diabetes or previous cardiovascular events (31). Therefore, because we only found an increased risk of endometrial cancer for thiazide diuretics, and not for other antihypertensive drugs, it is unlikely that the results observed in our study could be explained by residual confounding from diabetes or other cardiovascular-related proinflammatory conditions.

The protective effect of osteoporosis was not modified by the medication used to treat it (i.e., calcitonin or biphosphonates), suggesting that the relationship between the condition and the risk of cancer is independent of the treatment used. The effect of potential residual confounding by obesity or estrogen levels remains to be elucidated.

Finally, in our study, NSAID use was related to a marginally significant decreased risk of endometrial cancer (OR, 0.7; 95% CI, 0.5-0.97 after adjustment for age and BMI) whereas other analgesics (i.e., paracetamol and phenacetin), which are prescribed for similar indications as NSAIDs but lack their anti-inflammatory capacity, were not related to a decreased risk of endometrial cancer. Also, the fact that conditions for which NSAIDs are frequently used, such as migraine or rheumatoid arthritis, were not related to a decreased risk of endometrial cancer, argues against confounding by indication. No overall differences were seen between aspirin and non-aspirin NSAIDs. One previous case-control study found that regular use of aspirin was related to a significantly decreased risk of endometrial cancer, although only among obese women (OR, 0.50; 95% CI, 0.27-0.92; ref. 9); other NSAIDs were not studied. A recent cohort study has reported a decreased risk of endometrial cancer among current users of aspirin who were obese or postmenopausal and had not taken postmenopausal hormones, but no risk modification among other NSAID users (10). In contrast, the effect we noted in our study was similar in obese and non-obese women. Interestingly, one recent study has reported that use of NSAIDs is related to lower levels of circulating estrogens, offering a mechanistic hypothesis by which anti-inflammatories could reduce the risk of endometrial cancer (32). Also, factors related to an increased risk of endometrial cancer, such as obesity, insulin resistance, or aging, have been linked to higher inflammation levels, whereas protective factors such as pregnancy or regular menses have been related to lower inflammation levels (2).

In our study no evidence was found for a significant interaction between BMI and the conditions and medications assessed. Nevertheless, we found a suggestion of a greater risk of endometrial cancer among obese diabetics than among non-obese diabetics, something that others have also observed (33, 34).

The EDGE Study has several strengths: a population-based design, relatively large sample size, high-quality exposure and pathology assessment, and expert evaluation and classification of drug use. The use of population controls is especially important for studies assessing conditions and medications to limit selection bias. However, to further assess the possibility of selection bias, we compared the characteristics of our controls with other available data. In our study, 55% of control women were overweight, obese, or very obese at the interview date (i.e., BMI of 25-<30, 30-<35, or ≥35, respectively). This finding is similar to the 59.3% estimation of overweight or obesity prevalence among women older than 45 years old in the United States in 2005 (35). Type 2 diabetes mellitus and hypertension were reported respectively by 8% and 37% of our controls, somewhat lower than the estimated values of 12.5% and 43.5% among women older than 45 years old in the United States in 2005 in the Behavioral Risk Factor Surveillance Survey (35). In our study, 13% of the control women reported having used thiazides for at least 6 months. In the US Nurses' Health Study, regular use of thiazides among participants in 1980 was 8% (36) and in a 2004 Danish study, 22% of the controls reported having ever used a thiazidic diuretic (28). Because differences in the definition of user exist across studies, we consider the prevalence of thiazide use in our study to be in general agreement with the published data. Finally, in our controls we found a prevalence of NSAID use for at least 6 months of 24%, consistent with the 26% prevalence of daily use for at least 2 months in a recent breast cancer study in a population similar to ours (37). We therefore consider the control subjects included in our study to be representative of the similarly aged, general U.S. female population in terms of the studied conditions and medications.

On the other hand, our analysis may suffer from some potential shortcomings. The novel findings of associations with thiazide diuretics and anemia may be due to chance, because we examined a large number of conditions and medications in this study. However, given the exploratory nature of this analysis, which was designed to reveal potential new associations, we consider that all the signals from our data were important to capture. As in most interview-based case-control studies, information on study subjects is based on self-report. Therefore, recall bias is a potential threat to the validity of the study results. In our study, however, we do not expect recall bias to be an issue, as patients are unlikely to relate most of the conditions and medications assessed to a modified risk of endometrial cancer. Also, the fact that we have focused on medications chronically used (i.e., for at least 6 months) should help to enhance general recall of use, regardless of disease status. Recall bias is probably lower for drugs used on a daily basis (i.e., hydrochlorothiazides) than for drugs used sporadically (i.e., NSAIDs) or drugs whose use was suspended by the subject a long time ago. However, a residual degree of misclassification for medication use and presence of medical conditions cannot be completely ruled out. Bias due to an increased probability of diagnosis of the conditions assessed or treatments prescribed because of medical consultation due to an undiagnosed endometrial cancer is possible in this kind of study. However, we think it is unlikely that this bias would differentially affect drugs used to treat the same condition (i.e., thiazide diuretics and loop diuretics for hypertension), which, in our opinion, argues against a strong effect of a diagnostic bias on our data. We did not ask specifically about time since last use of medications, so we could not account for recency in our analyses. In addition, we did not ask about some conditions associated with use of NSAIDs, such as osteoarthritis. Larger numbers of cases and controls are needed to further investigate interactions of conditions and medications with BMI.

In conclusion, obesity, history of fibroids, and thiazide diuretic use were associated with an increased risk of endometrial cancer. A diagnosis of anemia and use of NSAIDs were related to decreased risk of endometrial cancer. The novel findings related to thiazide diuretics and anemia need to be confirmed in other studies.

Joan Fortuny is currently an employee at Novartis Pharmaceutical. The study and manuscript were conducted and prepared before the author's joining Novartis Pharmaceutical. Novartis has not participated in any stage of the study preparation, funding or conduct, and has not had access to the study data, manuscripts in preparation, or the final version of the manuscript. The other authors report no conflict of interest for the present study.

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.

We thank the interviewers and study assistants for their help with all aspects of the EDGE Study.

This work was supported by the National Cancer Institute [R01CA38918].