Introduction

An eating disorder is a psychological disorder characterized by a pathological way of eating affecting the body and the mind, resulting in an impaired physical and psychosocial functioning. Most well-known subtypes are anorexia nervosa (AN), bulimia nervosa (BN) and binge eating disorder (BED). AN is a psychological disorder characterized by a restriction of energy intake and an extreme weight loss, induced and sustained by the patient. It is mostly, but not exclusively, seen in adolescent girls, but boys and men may also be affected [1]. BN is a syndrome characterized by repeated episodes of overeating followed by vomiting, due to an excessive preoccupation with the control of body weight. The repeated vomiting gives rise to disturbances of electrolytes and physical complications. BN has a lot of psychological features in common with AN and a history of AN can be seen in a lot of BN cases [1]. If the inappropriate compensatory behavior ceases to exist, the patients meet criteria for BED [2]. Eating disorders are relatively common among adolescent girls and young women. Of European women, < 1–4% suffer from AN, < 1–2% from BN and < 1–4% of BED [3].

An eating disorder can make individuals vulnerable to disease. It is widely known that excessive weight measured as BMI is positively associated with the incidence and fatality of approximately all cancers [4]. The potential effects, both positive and negative, of excessive leanness, however, have received far less attention. A longitudinal study showed no increased cancer risk for low BMI [5]. The same lack of evidence can be found for eating disorders in general. Furthermore, energy restriction is a fundamental aspect of eating disorders like AN. Several animal studies have shown a protective effect of energy restriction without malnutrition on cancer risk [6]. Although energy restriction is more difficult to study in humans, some human studies have been conducted [7,8,9], with still limited research supporting the hypotheses that energy restriction without malnutrition might result in a lower risk of developing cancer in humans. However, malnutrition might be the result of excessive energy restriction as in AN, which can increase cancer vulnerability via immunodeficiency [10]. On the contrary, BED is often associated with excess energy intake. Also other cancer risk factors differ between eating disorders such as compensatory vomiting or laxative use, hormonal changes, mental health issues, seeking medical advice, tobacco/alcohol abuse, and sexual activity [11].

We have conducted a systematic review in which we evaluate the relationship between eating disorders and cancer risk by summarizing the existing epidemiological literature. Unique is that the focus is not on AN alone  since different eating disorders have different cancer risk factors and that also case reports were included. This can provide insight into human cancer etiology and ideas for cancer prevention for individuals with eating disorders and the general population.

Methods

The literature for human observational studies on eating disorders and cancer risk was reviewed following the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement [13]. The protocol was submitted to Prospero with number CRD42020158333.

Literature search

The search strategy was developped  by three independent collaborators with complementary expertise. We systematically searched the electronic databases Medline (via PubMed) and Embase for articles published from database conception up to 28th April 2020 of human observational studies or case reports about eating disorders and cancer risk, using Medical Subject Headings (MeSH), Emtree terms and text words. Tailored search strings containing keywords and database-specific terms for the two major topics eating disorders and cancer and risk were developed, combining multiple variations of search terms to produce the final search strategy (see Online Resource).

Selection criteria

Two independent reviewers (KM and MD) from our multidisciplinary team (medical and dietician degrees) screened titles and abstracts. When information was not sufficient to determine eligibility or in case of disagreement between the two reviewers, the full text report was obtained for further assessment. Eligibility criteria were applied by the two reviewers obtaining a shortlist of relevant articles.

The following selection criteria were applied.

Population or participants

We included observational studies performed in the general population and/or cases of ED. No exclusions were made based on socio-demographic characteristics (e.g., sex, age, origin) or for other subgroups.

Exposure

Eating disorder as defined by The International Classification of Diseases 11th Revision (ICD-11): Feeding and Eating Disorders as abnormal eating or feeding behaviors that are not explained by another health condition and are not developmentally appropriate or culturally sanctioned. Both diagnoses and self-reports were considered.

Comparison

A comparison was made with individuals without a history of eating disorder.

Outcome

The outcome was the incidence and/or mortality of overall cancer or site-specific cancer as defined by The International Classification of Diseases 10th Revision (ICD-10: C00-C97), tested in relation to eating disorder.

Study design

We included epidemiological studies comprising clinical trials, cohorts or case–control and cross-sectional designs. In addition, we decided to select case reports/series that reported data on defined outcomes (despite having higher risk of bias). In vitro studies, animal studies, editorials and reviews were excluded.

Language

Non-English full text articles were not excluded, but 54 translated titles/abstracts seemed not to match with the other inclusion criteria.

Data extraction and risk of bias assessment

Two independent reviewers (FDB and NM) extracted descriptive data and the summary measures of interest: Standardized Incidence Ratio (SIR) and mortality rate (MR) with 95% confidence interval (95% CI). Where possible, results were stratified by sex and eating disorder. In the case of duplicate studies, only the results of the most complete dataset were considered.

To detect bias potential, the study’s methodological quality was appraised using the Newcastle–Ottawa Scale (NOS) on a scale of 0–9. Studies can be classified as good quality when scoring at least three on four stars in the selection domain and at least one on two stars in the comparability domain (one star for age and sex confounder consideration, one star for alcohol or smoking consideration) and at least two on three stars in the outcome domain. Two reviewers (FDB and NM) independently evaluated all the studies and differences were resolved by consensus.

Results

Results of the search

The search strategy resulted in 990 unique hits on PubMed and EMBASE, checking the reference lists identified four extra case reports (see Fig. 1). Based on title and abstract, 46 full-text articles were further screened. Fifteen publications were included in this review. Publication dates from the included articles ranged from 1985 to 2017. There were six case reports [14,15,16,17,18,19], eight retrospective register-based cohorts, and one prospective register-based cohort [20,21,22,23,24,25,26,27] covering a population from Sweden, Denmark, Finland, Scotland, France, USA and Japan.

Fig. 1
figure 1

PRISMA flow diagram

Full size image

Of the cohort studies (including 47,049 individuals), only four included men (n = 2230, two cohorts did analyses split by sex) and only two included also non-AN eating disorders (n = 1783 individuals). Sample sizes of the cohorts ranged from 186 to 24,332 (median = 1678) eating disorder patients including 1–366 cancer cases (median = 5). Mean follow-up of the patients varied from 5.4 to 15.2 years. All of the studies looked at cancer incidence as the outcome of interest, and two studies also included cancer mortality [23, 26]. Breast cancer was the most often researched cancer site across publications. The key features of the included cohorts are shown in Table 1.

Table 1 Summary of included cohort studies
Full size table

Only six case reports met the inclusion criteria (see Table 2). Five of them reported esophageal cancer (of which two were squamous cell carcinoma and three were adenocarcinoma) and one reported three cases with leukemia. Only one case report included a man and the studies equally covered AN and BN. The time between onset of eating disorder and cancer diagnosis varied between 3 and 44 years.

Table 2 Summary of case reports
Full size table

The case studies did not overlap in population with the cohort studies. Within the cohort studies, some population overlap was noted. The study of Bens [26] was based on exactly the same population as Mellemkjaer 2015 [25], but added mortality analyses. The study of Papadopoulos [22] was based on exactly the same population as Karamanis [23] and only additionally examined parity as risk modifier. Therefore, the SIR of these two duplicates was only considered once when discussing effect sizes. The study of Mellemkjaer 2015 [25] combined registers from three countries and thus partially included populations from two other included cohort [20, 23], although the inclusion period was shorter in the latter two. The study of Michels and Karamanis [21, 23] might also partially overlap as they are based on the same national register, but the inclusion period was different.

The cohorts all had good methodological quality, except one, based on NOS criteria with a total between 5 and 8 on a scale of 9. Only two studies did not score the maximum for the selection component, based on potential non-representative sample [27, 28] and eating disorder classification by self-report [27]. All studies took age and gender as confounder into account. None adjusted for other lifestyle factors like smoking and alcohol, although one study showed that smoking and alcohol-related hospitalization was significantly higher in those with eating disorders [24]. Four studies received the maximum of stars for the outcome component, while three studies received two out of three stars due to unexplained non-response rate [24, 25] or cancer outcome based on self-reporting [27].

Summary of findings

The SIR and MR results by cancer group are discussed, in descending order of the number of cancer cases included.

Overall cancer (4 cohorts, 542 cancer cases, AN only)

None of the four cohorts [20, 21, 23, 25] found a significant SIR. One of the two studies found a significantly increased MR 2.0 [95% CI 1.3–2.9] [23]. The SIR effect sizes seem to suggest increased cancer risk in men and decreased cancer risk in women, but CIs were very wide [20].

Breast cancer (5 cohorts, 137 cancer cases, AN and BN)

The SIR ranged from 0.47 to 0.8, with three of them being statistically significant. Breast cancer MR was not significantly different (0.5 [0.1–1.9]) [23] or increased with 2.2 [1.4–3.4] [26] for AN. One study was able to distinguish between AN and BN, but only in a small subsample: the significant SIR could not be found in these split analyses, but AN might have a little bit more chance on reduced SIR [27]. The two studies that stratified on parity, both confirmed the significantly decreased SIR only in parous women and not in nulliparous women [21, 22]. Four studies have split analyses by eating disorder age of onset: one found no difference after adjustments [27], one could not calculate the difference because of zero cases in those < 20 years (but 20–29 years had less reduced SIR than 30–39 years) [21], one found only significantly decreased SIR in the youngest group (SIR: 0.4 [0.1–0.9] in 16-24 years versus 0.8 [0.4–1.5] in 15–40 years) [22], while another only in the older group (SIR: 0.7 [0.4–1.1) for 16–24 years versus 0.6 [0.4–0.7] for 25–40 years) [25]. The three studies that restricted their study population to women with AN diagnosis under 40 years old also found the lowest SIR [21,22,23].

Skin cancer (3 cohorts, 112 cancer cases, AN only)

Both for melanoma and other skin cancers, no significant SIR was found in AN (SIR ranged from 0.6 to 1.8) [20, 23, 25]. One study examining mortality found a statistically significant increase in mortality of melanoma by AN, with an almost tenfold risk of mortality (95% CI 3.1–22) [23].

Digestive organ cancers (4 cohorts, 5 case reports, 88 cancer cases, AN and other eating disorders)

Cohort studies examined esophageal (3 studies, 26 cases) [20, 24, 25], gastric (2 studies, 4 cases) [20, 24], pancreas (2 studies, 8 cases) [20, 25], colorectal (1 study, 27 cases) [25], liver (1 study, 8 cases) [25], gallbladder/bile duct (1 study, 2 cases) [25] and overall digestive organ cancer (1 study, 8 cases) [23]. Two [24, 25] of the three cohorts on esophageal cancer found a significantly increased SIR of 9.96 [4–10.51] and 5.1 [1.8–14.6] and also the study on liver cancer in women found a significantly increased SIR of 5.2 [2.3–12.1] [25]. The MR was not significant (2.3 [0.8–5]) [23]. Split by eating disorder, gastric cancer SIR was not significant in AN and non-AN, while esophageal cancer SIR was significantly increased in AN with 9.96 [4–10.51] and no cases were reported in non-AN [24]. Case studies reported only esophageal cancer, thrice in BN, once in combined AN/BN and once in psychogenic vomiting [14, 16,17,18,19].

Nervous system cancer (3 cohorts, 50 cancer cases, AN only)

None of the three cohort studies found a statistically significant SIR in females (ranging from 0.6 to 1.2) [20, 23, 25] despite a borderline or significant SIR in males (SIR = 12.5 [1.5–45] based on 2 cases; SIR = 2.3 [1.0–5.2] based on 7 cases). The study on mortality found no statistically significant MR (0 [0–3]) [23].

Respiratory cancer (4 cohorts, 50 cancer cases, AN and other eating disorders)

Two studies found a significantly increased SIR: 2.7 [1.0–6.0] for respiratory cancer and 1.6 [1.1–2.4] for lung cancer in women [23, 25]. By stratification, one study could not find different results by eating disorder type [24]. No significant MR was found (2.5 [0.7–6.3]) [23].

Lymphoid and hematopoietic cancer (3 cohorts, 3 case reports, 46 cancer cases, AN and BN)

None of the cohort studies found a significant SIR (ranging from 1.1 to 3.4) or MR (1.4 [0.2–5]) in AN [20, 23, 25]. The three cases in a case series all had leukemia and included two BN and one AN patient [15].

Female genital cancers (3 cohorts, 44 cancer cases, AN only)

Overall female genital cancer (12 cases), ovarian cancer (9 cases) and endometrial cancer (7 cases) were each reported by one cohort study [23, 25]. Cervical cancer was the only female genital cancer for which two studies (23 cases) presented SIR [20, 25] but only one study showed significantly decreased SIR 0.7 [0.4–1.0] for cervix [25]. A significantly increased MR of 9.6 [3.1–22] was found in female genital cancers for AN [23].

Cancers with less than 20 cases (AN only)

These studies included cancer of testis (1 study, 1 case), prostate (1 study, 1 case), kidney (1 study, 4 cases), lip/oral (2 studies, 2 cases), thyroid and other endocrine gland (3 studies, 18 cases). None gave statistically significant SIR or MR [20, 23, 25].

Discussion

Some of the seven cohorts on eating disorders found a decreased breast and female genital cancer risk, while an increased digestive and respiratory cancer risk. A recent meta-analysis on AN-specific cases indeed suggested that AN was associated with decreased breast cancer incidence and potentially with increased risk of developing lung and esophageal cancer [29]. In addition, we identified six case studies indicating esophageal cancer and leukemia in eating disorder patients. Also, mortality ratios differed between studies and cancer sites: eating disorder patients might have a higher risk of dying from overall, breast, female genital and skin cancer. We further discuss statistically significant results by cancer types considering the general findings.

Cancer mortality by eating disorders

An increased mortality rate in eating disorder patients was found for overall cancer (one of the two studies [23, 28]), breast cancer (one of the two studies [23, 26]), female genital cancer [23] and melanoma [23]. Possible explanations for a higher cancer mortality may be that AN patients generally have lower treatment adherence, are in poor general health, participate less in screening programs (for breast, cervix, skin cancer) due to poor mental health, and seek medical advice at a later time [26]. The hypo-estrogenic state of AN patients might also promote the development of estrogen-independent tumors. These have a poorer prognosis, higher recurrence rates and fewer treatment options [26]. Future studies might need to include information on tumor characteristics, such as the estrogen receptor status, to support this hypothesis.

Breast cancer risk

Three [21, 25, 27] of the five cohorts that comprised breast cancer patients found a significantly decreased risk for women with an eating disorder. Current knowledge shows that reproductive factors such as early menarche, nulliparity, older age at first birth, and older age at menopause play important roles in the formation of mammary tumors [30]. Due to the caloric restriction in patients with AN, their body undergoes in a stress reaction, leading to hypothalamic dysfunction and eventually a hypo-estrogenic state [30]. Since fat tissue aromatizes androgens into estrogens, the poor amount of body fat in AN plays a leading role in the cause of hypo-estrogenemia [30]. Having AN during a crucial time in mammary gland development, such as adolescence and early adulthood, can affect the risk of breast cancer. The timing of energy restriction may therefore be relevant, but no consistent direction for breast cancer risk difference depending on eating disorder age of onset was found across the studies. As a protective factor against breast cancer, AN is also associated with amenorrhea or delayed menarche and consequently a reduced number of lifetime ovulations. Athletes and women with Turner syndrome share characteristics with women with AN (such as a lower level of estrogens and amenorrhea) and have been found to also have a lower risk of breast cancer [26]. Furthermore, patients with AN tend to have less children or later in life (a reduced SIR was only found in parous, but not in nulliparous women [21, 22]) and have higher physical activity levels [32, 33]. On the other hand, several other factors can have an increasing effect on breast cancer risk and cancer risk in general, such as smoking, alcohol use and diet. However, none of the included studies considered these confounding factors in the analysis.

Female genital cancer risk

One [25] of the two cohort studies [20, 25] found a significantly decreased SIR 0.7 (95% CI 0.4–1.0) for cervical cancer in eating disorders. No significant finding for the endometrial [25] and ovarian [25] cancer appeared, but the number of cases was very low. First, lower circulating levels of estradiol might explain these lower risks of hormone-dependent cancers. Second, eating disorder patients are known to have lower sexual function and engage in little to no dating [34], which may be an underlying explanation for the reduction in cancers of the female genitals [20], e.g., a decreased risk found for cervical cancer due to lower prevalence of HPV infections [35].

Digestive tract cancer risk

Of the digestive tract cancers, esophageal cancer was the most researched (3 cohorts, 5 case reports) and also showed the highest SIR. Two cohort studies [24, 25] found a significant five- to sixfold increased risk for esophageal cancer in AN.

A first explanation might be vomiting. An international study of eating disorders in families reported a prevalence of regular self-induced vomiting of 97% in purging BN, 94% in AN and BN, 73% in purging AN, 66% in bingeing AN, and 56% in eating disorders not otherwise specified [36]. The more abundant gastro-esophageal reflux seen with obesity leads to Barrett’s esophagus, a metaplastic precursor that can lead to dysplasia and adenocarcinoma of the esophagus [33]. The same pathogenesis has been proposed in eating disorder patients that vomit. The risk of esophageal cancer is expected to be increased by self-induced vomiting causing acidic damage to the esophagus [12]. Three case reports are in agreement with these observations [14, 16, 19], and found adenocarcinoma of the esophagus or stomach in a BN patient that frequently vomited. This factor of self-induced vomiting is hard to measure, since the prevalence of self-induced vomiting differs due to the reliance on self-reporting.

A second explanation might be confounding by established risk factors (smoking, alcohol, nutritional deficiencies). After all, one cohort only found squamous cell carcinoma (SCC) instead of adenocarcinoma and showed that smoking and alcohol-related hospitalization was significantly higher in those with eating disorders [24]. Alcohol use disorder has been associated with SCC, and smoking with both main subtypes of esophageal cancer [24]. Even though eating disorder patients are often not heavy drinkers, they did report having frequent sense of loss of control over drinking [37] and sometimes smoke as a use for weight control [37]. Also two case reports found SCC in eating disorder patients [17, 18], although only one of them had a history of alcohol abuse and smoking.

Respiratory cancer risk

Two of the four cohort studies found a significantly increased SIR of 2.7 (95% CI 1.0–6.0) for respiratory cancer and 1.6 (95% CI 1.1–2.4) for lung cancer in women [23, 25]. The increased risk is probably due to the higher prevalence of smoking among patients with an eating disorder [37]. In addition, emphysematous changes in the lungs of patients with AN have been found, as is the case in other severely malnourished patients [38].

Leukemia risk

None of the three cohorts [20, 23, 25] found a changed risk toward lymphoid and hematopoietic cancer. However, we detected a case series including three young women diagnosed with leukemia in which two had AN and one had BN around the age of 20 years [15]. Apart from the above-mentioned hormonal and lifestyle factors, an underlying mechanism might be immunodeficiency due to malnutrition, which in turn can affect tumor progression by the suppression of tumor-suppressor genes [10, 39]. Nevertheless, a response letter questions this underlying mechanism [40].

Anorexia nervosa versus other eating disorders toward cancer risk

Since 96.2% of the included individuals had AN, making a conclusion on cancer risk differences between several eating disorders is difficult. One cohort study split the sample in AN versus non-AN [24] and another in AN versus BN [27], but no clear differences in cancer risk were detected, probably due to the low number of included non-AN eating disorder cases. In the case reports, both BN and AN were equally presented.

A first problem is thus the low amount of non-AN eating disorder studies in relation to cancer in the literature. Most of the retrieved studies were only interested in AN specific. Although eating disorder prevalence numbers depend on the source used, non-AN eating disorders are not less prevalent than AN [3]. By using clinical registers for population selection, non-AN eating disorder patients might be less represented as the lower presence of life-threatening symptoms (like underweight) prevent them often from contacting a clinician. Indeed, community studies often detect people with eating disorders who did not seek treatment and only about 20% of affected individuals present for treatment [41]. The duration of untreated eating disorder before the start of first treatment seems shorter for AN than for BN or BED [41].

As a second problem, there is considerable overlap of characteristics across the range of eating disorders [11] and thus probably also in health effects. Moreover, the ICD-10 definition of BN states that ‘‘There is often, but not always, a history of an earlier episode of anorexia nervosa’’ [1]. Indeed, AN and BN can co-occur like for example in one of the case studies [14].

Theoretically, the cancer risk might differ between eating disorders depending on the frequency of vomiting (increased risk towards esophageal adenocarcinoma) [12], laxative use (increased risk towards colorectal cancer for non-fiber laxatives) [42] and alcohol/smoking abuse (increased risk towards a broad set of digestive organ cancers, lung cancer, …). For the latter, non-AN eating disorder patients might have higher cancer risk. After all, smoking and alcohol-related hospitalization was significantly higher in those with non-AN eating disorders [24] and a meta-analysis concluded that people with BN but not AN were significantly more likely to be lifetime smokers than healthy controls [43]. Similarly, around 15% of BN or BED patients have multiple comorbid impulsive behaviors, including substance abuse, impulse buying, compulsive shopping, and multiple sexual relationships [41]. This difference in sexual relationships might pose non-AN eating disorder patients at higher risk for female genital cancer compared to AN. Also, breast cancer risk might be higher in non-AN eating disorders since the hypo-estrogenic state is less present. As another lifestyle factor, both BN and BEN are often accompanied by, or lead to, obesity and related metabolic disorders [41] which increase the risk for a broad set of cancers like digestive and genital cancers [44]. On the other hand, AN often starts earlier in life than BN or BEN [11], which might rather increase AN’s potential for carcinogenesis.

Recommendations for future high-quality research

Good evidence exists on the relationship between obesity and increased cancer risk [44]. The relationship with underweight has been less explored. These studies show no evidence that supports energy restriction as a protective factor for cancer. Further studies need to be conducted to form a basis for this hypothesis, so that energy restriction can be a solid recommendation for cancer prevention and earlier investigations can prevent cancer incidence in patients with an eating disorder. More studies need to be performed examining energy restriction without malnutrition. Importantly, research should differentiate between eating disorders because of differential cancer risk factors, e.g., BED is rather associated with adiposity instead of caloric restriction, and different levels of compensatory behaviors like vomiting, alcohol/drug/laxative abuse, physical and sexual activity are seen across eating disorder types.

Related to this, a first issue in current literature is confounder/confusing bias. All cohort studies took age and sex into account as confounders, but none adjusted for other lifestyle factors. Several other factors such as physical activity levels, weight evolutions, smoking, alcohol use, nulliparity and diet that affect cancer risk and that might differentiate eating disorder subtypes need to be included in future studies. Since smoking and alcohol abuse are both substantial confounding factors [37], they should also be obtained directly from the patient, instead of only assessing this confounding effect indirectly by looking at cancer of the lungs and esophagus as was done in one cohort [24]. Three out of four case studies found a tobacco or alcohol abuse in the ED patient [16, 17, 19]. Examining the role of psychiatric comorbidities in these eating disorder patients might advance mechanistic insights. One study found that adjusting for BMI reduced the breast cancer-protective SIR to non-significance [27]. Apart from adjusting for confounders, also some socio-demographic groups are underrepresented in current cohorts. For example, there is a huge lack of studies with male patients (only 4.7% of cases in the selected cohorts and only in one case report). Only for brain cancer a significantly higher risk for men with an eating disorder was found in the selected studies [20], but based on two cancer cases only. Future studies could benefit from including extra male participants to gain knowledge about gender differential effects as drivers toward eating disorders and expressed behavior might be different in men, e.g., body image concerns more on muscularity then weight loss [41].

Another issue is selection bias. A substantial number of patients with an eating disorder do not seek medical help [3, 45, 46], and only a minority of eating disorders are detected by health-care providers [3]. Since most of the cohorts relied on clinical registers, studies also need to be conducted in primary health care and the general community with patients who have not yet been diagnosed and hospitalized. This way not only the most severe cases are included, and study results may be more fit to extrapolate to the general population. Herein, it is important not to rely solely on self-reports, but to ensure confirmative diagnoses or at least a structured interview to avoid information bias, which is less an issue in clinical registry cohorts. A related issue is detection bias due to the relatively limited period of follow-up in relation to the time of development of cancer. For example, Bens et al. found a mean period of 18.6 years (range 1–39) between onset of AN and breast cancer diagnosis [26]. Future studies might therefore need to extend their follow-up period (and to also report non-response differences) to include additional cancers. This also decreases the risk of surveillance bias, since this has the greatest impact during the early years of follow-up. Of course, longer follow-up might increase attrition bias.

Finally, many cancer sites such as the urinary system, genitals other than cervix, nervous system, thyroid, hematological cancers, lymphatic system and non-esophageal digestive organs have been studied in very small sample sizes with limited power [20, 23, 25]. Results were non-significant and future studies should be conducted to make conclusions about these specific malignancies. Related to this, we cannot exclude the existence of a publication/reporting bias towards certain cancer types or eating disorder types (at least 10 studies are needed to test this).

Recommendations for clinical and public health practice

The low prevalence of eating disorders may mean that the overall cancer incidence among people with an eating disorder adds relatively little to the total pool of cancer cases. Yet, it is still important to those patients that their caretakers are aware of the increased risks. For example, screening of the esophagus should be indicated for eating disorder patients who vomit or with chronic acid reflux symptoms. Physicians tend to choose less aggressive treatment options for patients with an eating disorder, especially in AN. On the other hand, an appropriate treatment should be chosen since patients with a psychiatric condition are known to adhere less to treatments and some treatments have decreased effectiveness in someone with malnutrition. Finally, behavioral therapy might modify distorted lifestyle parameters (diet, drug abuse, physical activity, stress coping,…) that might increase cancer risk.

Strengths and limitations of this review

An important strength of this review is the independent development of an optimal search strategy by three researchers with complementary expertise. This guarantees the completeness of the obtained articles as the final search was optimized based upon the hits found in the three independent searches and by scanning cited/citing references. In addition, the search for this review included two electronic databases that delivered complementary hits. However, consideration of additional databases might still identify other publications. In comparison with a recent meta-analysis of AN cohort studies [29], we included also non-AN eating disorders to discuss potential differences with AN and we included case studies which gave a more detailed idea of confounders and differentiation (although with much lower risk evidence level).

It is important to also consider some limitations of this review when interpreting the results. Animal studies limit the generalizability to humans, therefore we did not include animal studies in this review. There was large heterogeneity in methodology and often very small sample sizes, which withhold conclusive results. Especially the question of differentiation between eating disorders was difficult to answer since the literature mainly focused on AN and the fact that AN and BN impact caloric restriction differently from BED. The sample is also not representative due to participants only from high-income countries and only from hospitalized patients. Since only one-third of people who meet criteria for AN are treated in mental health care, and 6% of people are treated for BN [47], results may not be applicable to the general population or primary health care. Some less severe cases of eating disorders are likely to have been missed. In addition, some studies relied on self-reporting of an eating disorder diagnosis: this introduces the risk of under- or over-reporting and misclassification of eating disorder type.

Conclusion

The literature on the relationship between eating disorders and cancer risk is very limited, with small samples. Most of the seven cohorts focused on women, AN and breast cancer incidence. Five of the case reports were patients with a history of vomiting, diagnosed with esophageal cancer. Despite some contradictive results across the included studies, eating disorder patients seem to have an increased mortality for some cancer sites, an increased digestive and respiratory cancer risk, but a decreased breast and female genital cancer risk. No clear differences in cancer risk were detected depending on eating disorder type, although theoretically non-AN eating disorder patients might be at higher risk (or at least no decreased risk for certain cancers). Hormonal changes, sexual activity, nutritional status, vomiting and concomitant tobacco/alcohol abuse may explain these findings. Concerning public health implications, screening of the esophagus should be indicated for eating disorder patients, they should be stimulated toward cancer screening program participation and health providers should consider comorbidities, adapted treatment and medication compliance monitoring. Future large-scale studies that also focus on male patients, bulimia nervosa and several cancer sites are needed and should foresee longer follow-up time and extensive confounder adjustment.

What is already known on this topic?

Eating disorders are quite prevalent. Some characteristics might be cancer protective, while others cancer-stimulating. Need for systematic review with focus on different eating disorder types.

What this study adds?

There is high cancer mortality, digestive/respiratory cancer risk, but low breast/female genital cancer risk due to eating disorder. Cancer screening (i.a. esophagus) is advised. There is a research gap in men and bulimia.