Discussion
This systematic review of 30 publications on the existence of a possible relationship between EoE and CD has revealed a high heterogeneity in the results provided, both with regard to the actual association as well as concerning the quality of the data. Most of the information available in the literature comes from case reports and short series of patients who share both entities; indeed, this appears to be a compelling reason for publication. Our findings show significant publication bias independent of the study area, with the inclusion of source information from three major databases supporting our interpretation. This bias severely affected the results of our meta-analysis. In fact, such a significant publication bias in favour of short case series of patients sharing both conditions could well have influenced the undue conclusion that there is a true association between EoE and CD.
The growing recognition of EoE worldwide in recent years has motivated an increasing amount of research on this disorder along with the proactive search for a diagnosis in patients with upper GI symptoms. In a clinical scenario of the increasing prevalence of CD affecting a higher proportion of the population, the availability of serological screening tests may also have contributed to the identification of EoE patients who were referred for endoscopic exams. In this case, the recognition of oesophageal mucosal abnormalities indicative of EoE in patients undergoing an endoscopic exam to obtain duodenal biopsies after a positive serology for CD may favour increased diagnosis of asymptomatic cases of EoE. Inversely, the systematic procurement of duodenal biopsies to exclude eosinophilic gastroenteritis in patients suspected of having EoE, especially in children (as recommended in current guidelines), also favours the diagnosis of CD, even in patients with no symptoms of the disease.
The prevalence of CD has been established as affecting 1% of the population, based on serological screening methods. However, many authors have repeatedly warned that serological tests underestimate the true prevalence of CD because seronegative cases are not detected. On the other hand, a population-based epidemiological study defined a prevalence of eosinophilic infiltration compatible with EoE (defined as 15 or more eosinophils/hpf) in 1% of randomly selected subjects undergoing endoscopic sampling from the distal oesophagus. With this high prevalence, the identification of patients sharing both disorders cannot be regarded as unusual, even if both disorders are independent from each other.
The study conducted by Ludvigsson et al. had the design with the least risk of bias. It was based on a probabilistic sampling from the general population (external validity), with a high response rate to endoscopic exploration (73%) and carried out independently of patient symptoms. Furthermore, endoscopists were blinded to the ASQ responses and medical history, thus avoiding an important source of information bias. Finally, endoscopic findings were verified by an expert endocopist and a professor of gastrointestinal surgery. In this sample, EoE was diagnosed in 11 (1.1%) patients, whereas coeliac disease was found in none of them (controls: 18/989). These figures show a clear independence between the prevalence of EoE and CD.
Thompson et al. analysed a cohort of patients diagnosed with CD along a 10-year period (2000 to 2009). Routinely (in paediatric cases) and at the discretion of the clinician (in adult patients), oesophageal biopsies were performed and EoE was diagnosed according to standard criteria. The observed number of EoE cases was compared to the theoretically expected number of cases obtained from incidence figures reported in the study conducted in Olmsted County (Minnesota) by Prasad et al. A standardised incidence ratio (SIR) higher than 10 was estimated in all groups, which were stratified by gender and age. However, this study presents bias and drawbacks that explain, in part, the association observed. Firstly, CD patients with EoE could be diagnosed and detected more easily because of an accumulation of symptoms from both diseases. Secondly, the risk period for the development of both diseases should be calculated from the inception of the cohort and not exclusively between the date of diagnosis of CD and EoE. Finally, the reference used to calculate the expected number of cases was extracted from medical records (cases elicited by a patient's demand of consultation), which could have led to an important underestimation of the true incidence of EoE. Thus, if we compare the figures of approximately one case in 10 000 (Prasad) and one in 100 (Ludvigsson), for the latter reference, the SIR in Thompson's study should be approximately one, which is compatible with the hypothesis of no association. Similar objections can be made for the study conducted by Stewart, which also found only three cases of concomitant CD and EoE.
Additional proof of the absence of a relationship between EoE and CD can be found in the limited efficacy of a gluten-free diet in reversing EoE-associated histopathological lesions documented in patients with CD; even when a broad heterogeneity (I = 64.7%) was found, the summarised result was just 32.7%. Wheat has repeatedly been demonstrated as the second most common food trigger for EoE in both children and adults after cow's milk, responsible for symptoms and eosinophilic inflammation in 22–60% of patients. In fact, a positive response to a wheat elimination diet in EoE does not necessarily imply that these patients also have CD.
CD is considered to be a genetically determined disease that affects genetically susceptible individuals who carry the HLA-DQ2 or DQ8 molecules. These HLA heterodimers are known to be the major genetic risk factors for CD, with a negative predictive value of almost 100%; however, the positive predictive value is poor, as approximately 40% of the population carries one or both of these alleles. Indeed, one large, multicentre, observational study demonstrated that the prevalence of HLA heterodimers conferring risk for CD in adult EoE patients was not superior to that found in healthy individuals, thus providing additional evidence for the absence of an association between both diseases.
The strength of our research lies in the fact that it compiles the results of an exhaustive literature search from three major databases, that recovered studies were critically appraised according to their methodological aspects, and that different investigators independently extracted the data from the studies included. The possibility of not recovering all the relevant information published on the putative relationship between EoE and CD has thus been minimised by exhaustive searching; indeed, a significant publication bias has been demonstrated by means of funnel plot analysis, which showed a trend to reporting a positive association between both diseases. In addition, most of the documents retrieved were case reports and short case series, with the most solid data coming from observational studies.
One limitation of this systematic review is that there was a certain amount of small study (or publication) bias. Funnel plot asymmetry might be a consequence of small study bias (often referred to as publication bias) which suggests either selective reporting of positive associations between EoE and CD or poor methodological quality.
In conclusion, given the lack of valid studies, we cannot rule out an association between the two diseases, but the evidence currently available does not unequivocally support this hypothesis. Indeed, the only epidemiological study with sufficient validity suggests the independence of both diseases, with other studies providing evidence against the existence of a link between EoE and CD. However, more well-designed studies are needed to confirm the results gleaned from this systematic review.