Women with celiac disease (CD) may be at decreased risk of female hormone-related cancers given the observed reduction in breast cancer seen in some cohorts. Using biopsy data from all 28 pathology departments in Sweden, we identified 17,852 women with CD who were diagnosed between 1969 and 2007. We used Cox regression model to estimate their risk of breast, endometrial and ovarian cancer and then compared them with 88,400 age- and sex-matched controls. The results indicate that individuals with CD were at a lower risk for all three outcomes: breast cancer (hazard ratio, HR = 0.85; 95% CI = 0.72–1.01), endometrial cancer (HR = 0.60; 95% CI =0.41–0.86) and ovarian cancer (HR = 0.89; 95% CI =0.59–1.34). This inverse relationship was strengthened when we excluded the first year of follow-up beyond CD diagnosis (breast: HR = 0.82; 95% CI =0.68–0.99; endometrial: HR = 0.58; 0.39–0.87; ovarian cancer: HR = 0.72; 0.45–1.15). In conclusion, CD seems to be inversely related not only to breast cancer but also to endometrial and ovarian cancer. Potential explanations include shared risk factors and early menopause.
There has long been interest in whether people with celiac disease (CD) are at increased risk of certain malignancies compared with the general population. Indeed, it would appear that this increased risk is the case for enteropathy associated T-cell lymphoma (EATL) of the small intestine and non-Hodgkin's lymphoma (NHL).1, 2 As a by-product of these observational investigations, the risks of other common malignancies have been reported, and of note, there has been a reasonably consistent finding suggesting that CD is inversely associated with breast cancer,3–11 which is the leading cause of cancer in women in the developed world. Although the reported relative risks vary five-fold (0.2–1.0), they all indicate a potentially negative association between CD and breast cancer.
The reduction in breast cancer risk among women with CD has been explained by malnutrition and weight loss associated with clinical or subclinical nutrient deficiency.12 Consequently, women with CD may suffer from disturbances in reproduction, including delayed menarche, early menopause and ovulatory dysfunction,13–15 indicating a lower life time exposure to estrogen. Sex hormones (e.g., estrogen and progesterone) clearly play an important role in breast cancer aetiology16 but estrogens in particular are also closely linked to both endometrial and ovarian cancer.17 This study therefore aimed to estimate the risk for breast, endometrial and ovarian cancer in women with CD using nationwide data from CD patients undergoing biopsy.
Subjects and Methods
We examined the risk of breast, endometrial and ovarian cancer among patients with biopsy-verified CD. Data on cancer were obtained from the Swedish Cancer Register18 and linked to biopsy data through the unique personal identity number assigned to each Swedish resident.19
We performed computerized searches of the 28 Swedish regional pathology departments to identify individuals with CD [equal to villous atrophy (VA) in the small intestine; equal to Marsh 3].20 We refer to our earlier paper for a detailed description of the data collection procedure, including the morphology codes used to identify patients.21 We also collected data from pathology departments on individuals with inflammation without VA (equal to Marsh 1–2) and normal small intestinal mucosa (equals Marsh 0). In all, our searches yielded a total of 351,403 unique biopsy reports in 287,586 unique individuals (unique individuals with CD: 29,148; with inflammation: 13,446).21 We then matched a regional subset of data on normal biopsy performed at any of 10 university hospitals with positive CD serology of the corresponding biochemistry departments from these hospitals. The purpose of this matching was to identify a group of individuals with normal mucosa but positive CD serology (defined as having an IgA or IgG antigliadin, endomysial, or tissue transglutaminase test within 180 days before biopsy and 30 days after biopsy) (for details, see our earlier paper22). We chose to include individuals with a small intestinal biopsy without VA in this study to examine if the negative association between CD and reproductive cancers was prevalent also in individuals with milder mucosal lesions (no VA) and not specific for CD.
Each individual undergoing small intestinal biopsy (CD, inflammation, normal mucosa but positive CD serology) was then matched on age, sex, county and calendar year with up to five reference individuals from the Total Population Register.23 This matching was performed by the national agency for population statistics, Statistics Sweden. We then excluded 174 individuals in which the biopsy could potentially have originated from the ileum and not the duodenum/jejunum, and another 35 individuals in which Statistics Sweden had failed to identify an age- and sex-matched reference individual or was unable to link to data on vital status. At this stage, our data set was identical to that of our earlier paper on mortality in CD.24 We then restricted our study to women and added data from the Swedish Cancer Registry.18 Relevant cancers were defined according to the international classification of disease (ICD) system, version 7 (breast: 170; endometrium: 172 and 174; ovary, 175).
In our main analysis, we excluded women with a breast cancer diagnosis before biopsy and study entry (or prevalent cases of ovarian cancer in the ovarian cancer analysis etc.). Hence, the main analysis of CD and risk of breast cancer was based on 17,852 women with CD and 88,400 age- and sex-matched reference women. Secondary cohorts consisted of 7,354 women with inflammation and 2,283 individuals with normal mucosa but positive CD serology and their age- and sex-matched controls. The number of women with CD and controls in the analyses of endometrial and ovarian cancer is listed in the legend of Table 1.
Table 1. Characteristics of CD patients and controls in the analysis of breast cancer
Statistical methods and analyses
We modeled the risk of cancer using Cox's regression model. Analyses were internally stratified in the sense that the risk of cancer was calculated for each stratum separately (a stratum consists of one individual who had undergone biopsy and up to five reference individuals) before stratum-specific risks were summarized to one overall risk estimate. The proportional hazards assumption was tested by means of log-minus-log curves and was not violated (Supplementary Material, figure). We have reported 95% confidence25 intervals and p values based upon an arbitrary cutoff of alpha = 0.05 for the purposes of completeness, however, in this analysis we have focused upon the point estimates and confidence intervals for the purposes of interpretation.25, 26
Follow-up time started at first biopsy with VA and ended with first breast cancer, death, emigration or end of study period (December 31, 2007), whichever occurred first. Follow-up in controls also ended if the control underwent small intestinal biopsy showing VA, inflammation, or normal mucosa at the time of positive CD serology. In analyses of ovarian and endometrial cancers follow-up ended if ovarian or endometrial cancer occurred.
One a priori analysis included stratifications for age (0–19; 20–39; 40–59 and 60+ years) and calendar period (-1989, 1990–1999 and 2000-), while a second a priori analysis involved follow-up time (<1 year, 1–5 years and >5 years). The latter analysis was planned because in other cohorts an increased risk of malignancies has been observed about the time of CD diagnosis, and it is difficult to know whether these malignancies are associated with CD or merely due to ascertainment bias. Given that our hypothesis is that there would be a reduced risk of female reproductive cancers, ascertainment bias could lead to a potentially spurious null finding if we were to include cancers diagnosed in the first year.
Because breast cancer is the most common reproductive cancer, we then went on to focus on breast cancer risk in CD. In subanalyses we adjusted for the following potential confounders: education, country of birth, type 1 diabetes, rheumatoid arthritis and autoimmune thyroid disease (see Supplementary Material for definitions and classifications). Not considering these factors may increase residual confounding and lead to over- or underestimation of risk estimates.
In a subset of women, we were able to obtain data on body mass index (BMI) and smoking from the national Medical Birth Register.27 We adjusted for BMI and smoking in this subset of women to determine whether these potential confounders would explain a negative association between CD and breast cancer. We also examined the risk of breast cancer according to the age at CD diagnosis before or after assumed menopause (=50 years; =51 years).28 In a post hoc analysis we examined the risk of breast cancer in CD when using age instead of years of follow-up since biopsy as the time scale.
Breast cancer in biopsied patients without VA
Because breast cancer is the most common female cancer, we also examined incident breast cancer in individuals with inflammation or normal mucosa but positive CD serology. These data are presented in the Supplementary Material. Statistical significance was defined as 95% confidence intervals (CIs) for risk estimates not including 1.0. We used SPSS 18.0 software to perform the analyses.
This study was approved by the Regional Ethical Review Board in Stockholm (2006/633-31/4). All data were anonymized before inclusion in the analysis.
Most study participants were adults at first biopsy (Table 1). Study participants had been biopsied between 1969 and 2007, with 1999 being the median year of biopsy and study entry.
Reproductive cancer in CD
Individuals with CD were at generally lower risk for all three outcomes: breast cancer (HR = 0.85; 95% CI = 0.72–1.01), endometrial cancer (HR = 0.60; 95% CI = 0.41–0.86) and ovarian cancer (HR = 0.89; 95% CI = 0.59–1.34) (Table 2). Risk estimates decreased when the first year after diagnosis was removed from the analysis with point estimates of 0.82, 0.58 and 0.72 for breast, endometrial and ovarian cancer, respectively. Fully stratified time after the follow-up analyses is shown in Table 2. The absolute risk of breast cancer in women with CD was 90/100,000 person-years, which can be compared with 106/100,000 person-years in controls (see Supplementary Material for additional data on absolute risks according to follow-up). Risk estimates for breast cancer did not differ by age at biopsy (p-value for interaction = 0.529), or according to calendar period (p-value for interaction = 0.278) (Table 3).
Table 2. CD and risk of subsequent female reproductive cancers
Table 3. CD and risk of subsequent breast cancer by age and calendar year of first biopsy with CD (stratified analyses)
Adjusting for type 1 diabetes, rheumatoid arthritis and autoimmune thyroid disease neither affect the HRs (HR = 0.86; 95% CI = 0.72–1.02) nor did adjustment for country of birth (HR = 0.85; 95% CI =0.82–1.01) or education (HR = 0.87; 0.73–1.04) affect the risk estimates. In a small subset of women with CD (n = 2,525) and their controls (n = 12128), we had data on smoking and BMI from the antenatal check-ups of the Swedish Medical Birth Registry.27 The proportion of smokers was 18.2% in women with CD and 20.8% in controls. In this subset the association between CD and breast cancer (HR = 0.77; 95% CI = 0.33–1.79) did not change more than marginally when adjusting for smoking and BMI (HR = 0.69; 95% CI = 0.29–1.67). Restricting our dataset to women with data on smoking and BMI within 5 years before or after CD diagnosis the adjusted HR for breast cancer was 0.88 (95% CI = 0.14–5.78).
The risk of breast cancer did not differ between women diagnosed with CD before (HR = 0.86; 95% CI = 0.66–1.12) or after (HR = 0.80; 95% CI = 0.63–1.00) assumed menopause (p-value for interaction = 0.669).
When in a post hoc analysis we used age as time scale in the Cox regression model instead of time since study entry, we found similar risk estimates for future breast cancer in women with CD (HR = 0.86; 95% CI = 0.72–1.02).
Biopsied patients without VA
In contrast to CD, inflammation without VA (HR = 1.12; 95% CI = 0.91–1.37) and normal mucosa but positive CD serology (HR = 1.15; 0.70–1.88) did not offer protection against future breast cancer. Additional data on the risk of breast cancer in women without VA are presented in the Supplementary Material.
When we compared women who had all undergone small intestinal biopsy, we found that women with CD were at a statistically significantly lower risk of future breast cancer than women with inflammation (HR adjusted for age and calendar period = 0.71; 95% CI = 0.56–0.90). Women with CD were also at a lower risk for breast cancer than women with normal mucosa but positive CD serology, but this risk estimate failed to attain statistical significance because of low statistical power (adjusted HR = 0.78; 95% CI = 0.49–1.23).
We found that, in comparison with the general population, women with CD have a reduced risk of developing breast, endometrial and ovarian cancer 1 year after diagnosis. Our study is the first that specifically assesses occurrence of all female reproductive cancers in CD and is by far the largest to examine breast cancer risk. Given that increases in risk of some particular cancers are seen in CD (e.g., non-Hodgkin's lymphoma) and that gastrointestinal cancer is more commonly diagnosed, presumably through ascertainment, our findings of an inverse relationship with cancers in the reproductive system are noteworthy.
Only a few studies have examined the risk of female reproductive cancers in women with CD other than breast cancer.4 In one such study Askling et al. reported varying standardized incidence ratios (SIRs) for endometrial (SIR = 0.6; 95% CI = 0.2–1.3) and ovarian (SIR = 1.3; 95% CI = 0.5–2.7) cancers,4 but their risk estimates were based on few cases (endometrial: n = 5; ovarian: n = 7).
In accordance with most studies (Table 4), we found a negative relationship between CD and breast cancer,4–7, 9, 11 but in contrast to those studies, our HR for breast cancer was only marginally decreased. In fact our risk estimate is almost identical to that of the smaller Viljamaa study from Tampere, Finland.10 Awareness of CD is very high in Finland and it is possible that in a region where much of the population has been screened for CD, many patients likely suffer from mild asymptomatic CD that will have little influence on breast cancer risk. Several older studies with relative risks below 0.5 may have been primarily limited to patients with severe disease (patients admitted to hospitals,4, 7 or specialist centres,9 or with a large share of follow-up time in the 1970s and 1980s when malabsorption was a common cause for CD work-up4, 7, 9). Even though the first patients in our CD cohort were biopsied in 1969 (Table 1), the median year of diagnosis occurred in 1999, making this a rather modern cohort. We do not think the moderate decrease in breast cancer in the current study is due to country-specific factors in that an earlier paper from Sweden reported a SIR of 0.3 for breast cancer in a cohort of hospitalized patients with CD.4 Interestingly, the Askling et al. study found a SIR for breast cancer in dermatitis herpetiformis (0.9) that is similar to the CD HR in our study.4
Table 4. Summary of previous studies of CD and breast cancer
Strengths and limitations
This is the largest study on CD and female reproductive cancers to date: as an example of its size, the number of breast cancers is almost four times that of all earlier studies combined.3–11 Our large number of participants allowed for important stratifications, including pre- and postmenopausal CD and breast cancer. Still we had insufficient power to show a statistically significant decrease in breast cancer (at the arbitrary alpha level of 0.05) except for when we removed the first year of follow-up (HR = 0.82; 95% CI = 0.68–0.99). Nevertheless, we argue that CD confers a reduced risk for breast, endometrial and ovarian cancer since the point estimates for these three outcomes were consistently below 1.25
We had extensive data on potential confounders, none of which influenced the risk estimates more than marginally. Although we had data on only BMI and smoking in a subset of individuals, this analysis was nevertheless based on more than 2,500 women with CD and showed that the low HR for breast cancer cannot be explained by smoking and BMI patterns in women with CD.
We based our CD diagnosis on histopathology findings from biopsy reports from all Swedish pathology departments (n = 28). Some 95% of patients with VA have CD,21 and this positive predictive value is actually higher than that of a diagnosis of CD in the Swedish Hospital Discharge Register.29 Other explanations for VA than CD are rare in Sweden. When two independent reviewers manually examined more than 1,500 biopsy reports, inflammatory bowel disease (IBD) was the only disease occurring in at least 0.3% of the patients.21 Although we did not require a positive CD serology for the diagnosis of CD, 88% of those with available serology data were positive at the time of biopsy21 while 79% had gastrointestinal symptoms at diagnosis. Biopsy reports also have high sensitivity for diagnosed CD, with 96–100% of all gastroenterologists/pediatricians in Sweden performing a biopsy before a diagnosis of CD.21 Another strength of this article is that we were able to calculate the risk of breast cancer not only in CD but also in individuals undergoing biopsy but without VA.
The validity of a breast cancer diagnosis in the Swedish Cancer Register is close to 100%.30 This partly stems from the routine mammography follow-up offered to all middle-aged women in Sweden. In these age groups the validity is also high for endometrial cancer.
One weakness of our study is the lack of data on dietary adherence in CD. In a random sample of patients with CD indications of low adherence were found in 17% of the patients.21 We accepted positive gliadin antibodies as proof of positive CD serology in our secondary cohort.22 We did so because for much of the study period, other antibodies were not available in Sweden and although we admit that the specificity of gliadin lacks behind that of endomysial and tissue transglutaminase antibodies, the most comprehensive review of antibody specificity so far indicates a specificity of about 93–94% for gliadin antibodies.31
Although we found similar HRs for breast cancer in CD diagnosed before and after menopause, we acknowledge the lack of individual-based data on age of menarche or menopause. Both age of menarche and age of menopause may differ by CD status.13 We were unable to consider parity. Both Swedish32 and British33 data, however, strongly suggest that women with CD have normal overall fertility.
Potential mechanisms of action
The persistent decrease in breast, endometrial and ovarian cancer in women with CD suggests that lower BMI, and potentially lower estrogen exposure, may protect against these cancers in women with CD.17, 34 It has been noted that starvation during adolescence, as in anorexia nervosa,35 and low-growth velocity in childhood36 protect against breast cancer. Both of these features can occur in CD.37
Noteworthy is that the inverse relationship between CD and breast cancer is consistent with data on such other inflammatory disorders as ulcerative colitis (confined to pancolitis)38 and rheumatoid arthritis,39 diseases known to carry a lower risk of breast cancer. We hypothesize that the degree of inflammation is important in carcinogenesis, not only about the increased risk of lymphoproliferative cancer2, 40 but also for the reduced risk of breast, endometrial and ovarian cancer. Of note, the current study found only a negative association between CD (Marsh 3) and breast cancer and not for entities with less mucosal damage (Marsh 1–2 or Marsh 0 with positive CD serology).
Given that women with CD have more exposure to health care professionals than the general population it seems unlikely that a lack of appropriate diagnosis of female cancers explains the inverse relationships we have found. On the contrary, we might have expected both endometrial and ovarian cancers to be diagnosed more often than in our chosen control group because of this access to health care and some cross-over in the symptoms attributable to both diseases (i.e., abdominal pain).41
In conclusion, we have shown that CD seems to be inversely related to breast, endometrial and ovarian cancer. These relationships persist for some time after diagnosis, suggesting that further study of the occurrence of these cancers in women with CD may lead to important insights into their etiology.
J.F.L. was supported by grants from the Örebro University Hospital Research Foundation, Örebro University, Swedish Society of Medicine, the Swedish Research Council, the Sven Jerring Foundation, the Örebro Society of Medicine, the Karolinska Institutet, the Clas Groschinsky Foundation, the Juhlin Foundation, the Majblomman Foundation, the Uppsala-Örebro Regional Research Council and the Swedish Celiac Society. J.W. was funded by a UK National Institute for Health Research Clinician Scientist Fellowship. O.S. was supported by a post-doctoral scholarship from the Swedish Society of Medicine. A.E. was funded by the Stockholm County Council.