Eating Disorders and Breast Cancer

Eating disorders such as anorexia and bulimia nervosa are relatively rare (1). In addition to being associated with adverse mental health outcomes (1–4), eating disorders can have long-lasting effects on overall and reproductive health through a variety of mechanisms, including low body mass index (BMI; ref. 1), altered menstrual function (2, 5–7), delayed child bearing or infertility (3, 8–10), and pregnancy complications (11–15). Many of these long-term health consequences are themselves considered established risk factors for breast cancer (16–23). The relationship between anorexia nervosa requiring hospitalization and incident breast cancer was investigated in several Nordic population registry studies, all of which reported inverse associations (24–28).

We used a large cohort to explore the association between self-reported history of anorexia nervosa or bulimia nervosa and breast cancer risk. The U.S.-based Sister Study cohort collects extensive data on women's general health, reproductive histories, and breast cancer incidence. With this wealth of data, we provide a more comprehensive picture of the association between eating disorders and breast cancer risk than has previously been available for the U.S. population.

The Sister Study comprises 50,884 women from the United States and Puerto Rico who had a sister with breast cancer, but had never been diagnosed with breast cancer themselves, and were 35–74 years of age when they enrolled between 2003 and 2009. Baseline information was collected via a computer-assisted telephone interview. Participants provided information on demographic and lifestyle characteristics from childhood to present, personal and family history of medical conditions, and occupational and environmental exposures. Study examiners collected body measurements during in-home interviews. Extensive follow-up interviews are conducted every 2–3 years.

We asked women to self-report any breast cancer diagnoses in annual health updates. Whenever possible, we validated these diagnoses using medical records. As agreement between self-reports and medical records was better than 99% for total and invasive breast cancer, we included self-reported cases even when medical records were not available (18%). Although we did not test participants for BRCA1/2 mutations, we did ask them if they had ever been tested, and if so, for their results. All participants provided informed consent and the study was approved by National Institute of Environmental Health Sciences' and the Copernicus Group's Institutional Review Boards.

In addition to the computer-assisted telephone interview, participants were mailed a diet questionnaire that included the question: “Have you ever had anorexia or bulimia?” Those who responded “Yes, currently” or “Yes, in the past” were then asked “How old were you when you first had this?” and “For how long did you have this?” Participants could respond with any value for age when the eating disorder began, and were given three possible responses for eating disorder duration: <1 year, 1–2 years, or >2 years.

Five percent of participants (n = 2,294) skipped the entire eating disorder section and were excluded from further analyses. Women who said that they had had an eating disorder currently or in the past were counted as having had an eating disorder, unless they provided an age that was younger than seemed plausible (<9, n = 9). Some women did not respond to the initial eating disorder question but provided an age at onset or duration (n = 194). Such responders were categorized as having had an eating disorder unless the provided age was less than 9 (n = 4) or if they also answered yes to the immediately preceding question about lactose intolerance/sensitivity, as it was deemed likely that these participants thought they were providing information about age at onset for that condition instead (n = 102; leaving 88 that counted as having had an eating disorder). We then excluded participants missing data for race/ethnicity, childhood socioeconomic status (SES) or baseline education level (n = 556), and women who were diagnosed with breast cancer before completing the enrollment process or who had no available follow-up information (n = 112). The final sample included 47,813 women.

As part of a follow-up study, we resurveyed some of the women who reported having had an eating disorder. We targeted women who had responded to their most recent follow-up questionnaire and who either had eating disorders that started during adolescence or early adulthood (defined as between the ages of 9 and 22; n = 907), or who said they had an eating disorder but did not provide an age (n = 99). Because of limited resources and competing research priorities, we were not able to resurvey women who had later-onset eating disorders or conduct a true validation study. Nevertheless, this follow-up survey provided valuable insight into the types of eating disorders our participants experienced.

In the follow-up questionnaire, we repeated the original eating disorder–related questions and asked whether the participants had anorexia, bulimia, or both, as well as other questions that allowed us to more accurately classify the timing, characteristics, and severity of any reported eating disorders. Criteria for severity were based on the Diagnostic and Statistical Manual of Mental Disorders, 5th edition (29). Severe anorexia nervosa was defined as (i) self-reported hospitalization or institutionalization for anorexia; (ii) at lowest weight, a weight less than 85% of the median weight for age (ages 9–19; ref. 30) or BMI ≤ 17 kg/m² (age ≥20); or (iii) postmenarche, nonpregnant, nonlactating amenorrhea for 3 months or longer. Severe bulimia nervosa was defined as: (i) binge eating combined with compensatory behavior (e.g., vomiting or laxative use) at least once a week for 3 months; or (ii) self-reported hospitalization or institutionalization for bulimia.

We examined the relationship between eating disorders and invasive breast cancer using Cox proportional hazards models. Age was the primary time scale and we used generalized estimating equation (GEE) methods to account for within-family clustering (as some families were represented by multiple sisters). Women were followed until death, most recent follow-up, or diagnosis with breast cancer. For analysis of invasive breast cancer, a diagnosis of ductal carcinoma in situ (DCIS) was treated as a censoring event. We also conducted analyses to estimate the effects of eating disorders on postmenopausal invasive breast cancer, estrogen receptor (ER)-positive invasive breast cancers, and a grouped outcome including either invasive breast cancer or DCIS. In addition, we re-ran analyses after excluding individuals who either tested positive for BRCA1/2, or who had a sister who reported a positive test but were not tested themselves (334 individuals from 313 families). For analyses of postmenopausal risk time, participants were left-truncated by entering them into follow-up at study entry or age of menopause, whichever occurred later. There were too few ER⁻ and premenopausal breast cancer cases with eating disorders to conduct informative analyses.

When assessing the relationship between eating disorders and breast cancer, we adjusted for race/ethnicity, childhood SES (as defined by the education level of the head of the participant's household when she was 13 years old), baseline SES (as defined by the participant's own education level), and birth year in our main analyses. For the main analyses, we did not adjust for covariates that potentially acted as causal mediators between eating disorders and breast cancer (e.g., age at first-term pregnancy, alcohol use, physical activity, or BMI), although we did assess the impact of adjusting for or stratifying by BMI in sensitivity analyses. Ideally, we would have conducted more comprehensive mediation analyses to isolate the effects of individual causal pathways, but in reality the covariates of interest are too intertwined for this to be an informative and valid approach (31). The timing of eating disorder initiation relative to other covariate measures made these relationships especially difficult to disentangle.

We examined the proportional hazards assumption by testing the coefficient for a time interaction term against the null (P < 0.05 was considered evidence of violation). We examined HR modification by age at onset, duration, and timing of eating disorder using likelihood ratio tests. All analyses were carried out using Sister Study Data Release 3.0 (April 2014) and SAS software, version 9.3 (SAS Institute, Inc.).

Three percent (n = 1,565) of participants reported ever having had an eating disorder (Table 1). Women with a history of an eating disorder were more likely to have high childhood SES (39% vs. 26% of those without a history of an eating disorder) and were more likely to be nonHispanic white (90% vs. 85%). Those with eating disorder histories were also younger, more educated, more likely to be nulliparous, and more likely to be underweight at ages 30–39 years or at baseline. For example, 22% and 62% of participants without eating disorder histories were overweight or obese at ages 30–39 years and baseline, respectively, versus 17% and 45% of those who reported having had an eating disorder. Women who were excluded due to missing data (n = 3,071) had lower adult and childhood SES, higher BMI currently or at ages 30–39, and were less likely to be nonHispanic white.

After a median of 5.4 years of follow-up, 1,475 women had developed invasive breast cancer. There was an inverse association between self-reported history of an eating disorder and overall risk of breast cancer [HR = 0.70; 95% confidence interval (CI), 0.49–0.99; Table 2]. The association did not seem to vary by age at onset of the eating disorder (9–22 vs. >22, P_{heterogeneity} = 0.18) or duration of the eating disorder (<2 years vs. ≥2 years, P_{heterogeneity} = 0.24). Compared with those with no history of eating disorders, those with eating disorders in the past had a reduced risk of breast cancer (HR = 0.62; 95% CI, 0.42–0.92). The breast cancer HR for having a current eating disorder was substantially different (HR = 3.81; 95% CI, 1.69–8.60; P_{heterogeneity} = 0.001), but only 6 women with current eating disorders developed breast cancer. This number did considerably exceed the expected number of 1.7.

Women whose self-reported average weight corresponded to a BMI during their thirties of <18.5 kg/m² had a reduced risk of breast cancer, relative to those with normal weight (HR = 0.67; 95% CI, 0.45–1.00). The effect was closer to the null when we instead considered current, examiner-measured BMI. As previously reported (32), women who were overweight or obese at baseline had an increased risk of breast cancer, with a stronger effect seen in postmenopausal women (Supplementary Table S1). All of the other observed associations were fairly similar when we restricted to postmenopausal women or women without known BRCA1/2 mutations, but were generally weaker for analyses that included only ER⁺ cases or that included both invasive and DCIS cases. We observed no violations of the proportional hazards assumption.

In sensitivity analyses, we assessed the influence of adjusting for or stratifying by baseline BMI on the association between eating disorders and breast cancer (Table 3). We found that the HR for ever having an eating disorder was attenuated with adjustment, but still indicative of a protective association (HR = 0.82; 95% CI, 0.62–1.10). The inverse association was somewhat stronger in women with current BMI ≥25 kg/m² (HR = 0.64; 95% CI, 0.40–1.03 vs. HR = 0.96; 95% CI, 0.67–1.38 for women with current BMI <25 kg/m²). The BMI-adjusted HR for the association between past eating disorders and incident breast cancer was also attenuated (HR = 0.70; 95% CI, 0.51–0.97), but the BMI-adjusted HR for current eating disorders and breast cancer was virtually unchanged (HR = 3.81; 95% CI, 1.90–7.64). Effect estimates adjusted for or stratified by self-reported BMI ages 30–39 years were similar, although here the protective association was slightly stronger for women with BMI <25 than BMI ≥25 (HR = 0.77 vs. 0.99; Supplementary Table S2).

Of the 1,006 women with young-onset eating disorders asked to complete the eating disorder validation survey, 62% (n = 627) responded. Of those, 78% (n = 492) confirmed that they had had a young-onset eating disorder. Bulimia nervosa was slightly more common than anorexia nervosa (62% vs. 56% of those with eating disorders, with 89 women reporting both) and 368 (75%) of those with a confirmed eating disorder were classified as severe (202 with anorexia, 207 with bulimia, and 41 with both). Because of small numbers, we were unable to meaningfully evaluate the association between eating disorders of specific types or severities, but based on crude estimates, the HRs for breast cancer were reduced for both types of disorder and for severe eating disorders (Supplementary Table S3).

In this prospective cohort study of women with a first-degree family history of breast cancer, we observed an inverse association between ever having had an eating disorder and breast cancer risk. Those with eating disorders in the past were at particularly low risk of breast cancer, while those with current eating disorders were more likely to be diagnosed with breast cancer over the subsequent follow-up period. Very low average BMI during ages 30–39 years was associated with a reduced risk of breast cancer.

These results are consistent with the results of previous population-based registry studies. In all of the five studies we identified (24–28), the authors compared breast cancer incidence rates for women diagnosed with anorexia nervosa to breast cancer incidence rates in the general population. Anorexia diagnoses were based on hospital records, which could be linked to the nationwide cancer registries maintained throughout Denmark, Sweden, or Finland. Although not all of these studies found statistically significant associations between anorexia diagnoses and breast cancer incidence, all reported standardized incidence ratios less than 1.0 (range 0.6–0.92). Two of the studies found a stronger association among women diagnosed with anorexia at younger ages (27, 28), although none were able to consider past versus current anorexia or expand the study to include bulimia nervosa cases.

A recent population-based, nationally representative household survey of 9,282 U.S. adults (2001–2003) estimated the lifetime prevalence of anorexia nervosa and bulimia nervosa among females to be 0.9% and 1.5%, respectively (1). These estimates are consistent with the estimated lifetime prevalence of eating disorders in our study sample, which was approximately 3%. Any slight increase in prevalence for our sample is expected, given that the eating disorders may be more common in women from more highly educated families (3, 33) and that incidence has been increasing over time (1, 34).

Previous studies have demonstrated that eating disorders are associated with several factors that could affect breast cancer risk. While delayed childbearing and nulliparity would increase breast cancer risk among those with a history of eating disorders (3, 8, 16), many of the other potential health consequences or correlates, such as high physical activity levels (35, 36), infertility (9, 10, 19), delayed menarche or secondary amenorrhea (2, 5–7, 18), having infants with low birth weight (11–15, 21, 23), and low BMI (1, 20, 32), could result in reductions in breast cancer risk. The latter is supported by the evidence that the associations between ever or past eating disorders and incident breast cancer were attenuated when we adjusted for BMI. Other potential mediators of an eating disorder–breast cancer association include breast density, smoking, alcohol use, height, diet, and reproductive outcomes such as miscarriages, induced abortions, or having a preterm baby. On the whole, it appears that the protective factors outweigh the risk factors, resulting in an overall inverse association between eating disorders and risk of invasive breast cancer.

The finding that current eating disorders are associated with an increased risk of breast cancer deserves further exploration. Although this is a prospective study, the presence of an eating disorder could be a marker for undiagnosed breast cancer or other health complications (e.g. something immunologic or infectious) that are also potentially related to breast cancer risk. Alternatively, the association could be driven by stress, nutritional deficiencies, or BMI. With regard to the latter, we observed that although women with any history of an eating disorder were less likely to be obese at baseline than those with no eating disorder history (19% vs. 30%), 35% of women with current eating disorders were obese at baseline. As only 7% of these women reported being obese during their thirties, this may be an indication that eating disorders that are ongoing or newly developed in older women (i.e., ≥40) are very different from eating disorders in younger women. For example, even if they started with anorexia or bulimia, they could now have a binge eating disorder, or could be experiencing severe weight cycling.

We also acknowledge that the positive association between current eating disorders and breast cancer may be a spurious finding, especially as there were only 6 women with both diseases. Four of 6 women with current eating disorder and breast cancer had BMI ≥30 kg/m², putting them at high risk of breast cancer (20, 32). The follow-up study did not provide additional insight into these 6 women, as only 1 of the 6 responded.

The greatest strength of this study is its large sample size, which is especially important given the rarity of anorexia and bulimia. In addition, we had extensive data on our participant's reproductive histories, anthropometric measures, and general health. Most Sister Study participants are well-educated, nonHispanic white women. While this homogeneity limits the generalizability of our results, it also means that we were able to study these disorders in a highly susceptible sample. Because they all have a first-degree family history of breast cancer, our participants are also more susceptible to breast cancer. While this may further limit our generalizability, we previously demonstrated how our cohort has high internal validity and that it is unlikely that known, but rare, highly penetrant genes such as BRCA1 and BRCA2 are highly influential in studies of nongenetic risk factors (37). This is supported by the sensitivity analyses included here, in which we found nearly identical results after excluding families with known BRCA1/2 mutations.

There may be selection or reporting bias if certain characteristics were associated with a woman's decision to participate in this volunteer cohort or to understand or admit that she had an eating disorder. Eating disorder–related fatality could act as such a selection factor, and may be an unmeasured source of bias. We mitigated the influence of race, birth cohort, age, and SES on participation rates by adjusting for them in all of our models.

Although anorexia and bulimia can cooccur and likely share some risk factors and health effects, we were not able to capture differences between the two behaviors within the full cohort. The follow-up survey allowed us some insight into the type and severity of the eating disorder participants experienced, but due to small numbers and incomplete responses, we were unable to closely examine the health consequences of having anorexia versus bulimia or severe versus nonsevere eating disorders.

As we categorized women based on their self-reported eating disorder status rather than clinical diagnoses, some misclassification is expected. Women may have misunderstood the question, misunderstood what qualifies as an eating disorder, or be too embarrassed to disclose a previous eating disorder. On the other hand, our use of self-reported eating disorder status allowed us to capture a more inclusive case definition than has been possible with previous studies, which typically recruit from eating disorder clinics or hospitals (24–28). Given that only roughly half of individuals with eating disorders seek treatment (1), the effect estimates reported here could be considered more representative of the risk factors and health effects of eating disorders as they typically occur.

In the Sister Study, women with historical eating disorders had a decreased risk of developing breast cancer while those with eating disorders ongoing at the time of enrollment had increased breast cancer risk. The apparent association with current eating disorders could be a spurious finding or indicative of other ongoing health problems that are more directly related to disease risk, such as chronic stress, obesity, or nutritional deficiencies. Historical eating disorders, on the other hand, could have directly affected overall and reproductive health during crucial developmental periods in adolescence and early adulthood. Even if some of these health consequences are unfavorable on their own, the evidence suggests that their combined effects may result in a long-term reduction in breast cancer risk.

No potential conflicts of interest were disclosed.

Conception and design: K.M. O'Brien, D.P. Sandler, C.R. Weinberg

Development of methodology: K.M. O'Brien, C.R. Weinberg

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.M. O'Brien, D.P. Sandler, C.R. Weinberg

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.M. O'Brien, D.R. Whelan, C.R. Weinberg

Writing, review, and/or revision of the manuscript: K.M. O'Brien, D.R. Whelan, D.P. Sandler, C.R. Weinberg

Study supervision: C.R. Weinberg

We thank Dr. Donna Baird, Dr. Megan Carnes, and other members of the Epidemiology Branch of the National Institutes of Environmental Health Sciences for helpful comments on this article.

This work was supported by the Intramural Research Program of the National Institutes of Health, the National Institute of Environmental Health Sciences (project Z01-ES044005 to D.P. Sandler and Z01- ES102245 to C.R. Weinberg).

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.