Systematic Review With Meta-analysis

The Growing Incidence and Prevalence of Eosinophilic Oesophagitis in Children and Adults in Population-based Studies

Pilar Navarro; Ángel Arias; Laura Arias-González; Emilio J. Laserna-Mendieta; Miriam Ruiz-Ponce; Alfredo J. Lucendo


Aliment Pharmacol Ther. 2019;49(9):1116-1125. 

In This Article

Materials and Methods

This systematic review has been registered in the PROSPERO International prospective register of systematic reviews (; register no. CRD42018108756), and has been reported in accordance with the PRISMA statements.[11]

Selection of Studies

A systematic literature search was performed independently by two researchers (AJL and AA) in three major bibliographic databases (PubMed, EMBASE, and Scopus) for the period up to September 2018. The search was not restricted with regard to date or language of publication. A predetermined protocol was used in accordance with the quality standards for reporting meta-analyses of observational studies in epidemiology.[12] Comprehensive search criteria were used to identify articles dealing with the epidemiology of EoE in children and adults. The following search strategy was used to consult the thesauri for MEDLINE (MESH) and EMBASE (EMTREE): ("eosinophilic esophagitis" OR "eosinophilic oesophagitis") AND ("epidemiology" OR "incidence" OR "prevalence" OR "demography"). As for the SCOPUS database, only free text searches with truncations were carried out. Reference lists from retrieved articles and abstracts of conference proceedings (taken from abstract books from the annual Digestive Diseases Week, American College of Gastroenterology Meetings and the United European Gastroenterology Week for the period between 2014 and 2018) were also examined to identify additional, relevant studies. Four reviewers (PN, AA, LA-G & AJL) independently screened the database search for titles and abstracts. If any of the reviewers felt that a title or abstract met the study eligibility criteria, the full-text of the study was retrieved.

Inclusion Criteria

A combination of symptoms referred to oesophageal dysfunction and a dense eosinophilic infiltration (≥15 eosinophils per high power field) in oesophageal biopsies was a diagnosis of EoE. Population-based studies including national, provincial/state-wide and local estimations were considered if they provided original data on the prevalence and/or incidence of EoE in children and/or adults, irrespective of the study design or document format.

Exclusion Criteria

Our study excluded clinical guidelines, consensus documents and reviews that did not provide original epidemiological data. We also excluded studies not carried out on humans, papers providing duplicated information (ie repeated abstracts presented at different congresses or abstracts subsequently published as a full-paper), and studies using subsets of patient cohorts from previously published research by the same group of authors.

Risk of Bias Assessment

Retrieved documents were evaluated for risk of bias only if the article described all the patients' demographical data, the diagnostic criteria used for EoE, and the reported prevalence/incidence with its 95% CI. Risk of bias assessment was checked against The Joanna Briggs Institute Critical Appraisal Checklist for Studies Reporting Prevalence Data.[13] A study was considered to be at low risk for bias if each of the bias items could be categorised as low risk. Studies were judged to have a high risk of bias however if any one of the items was deemed high risk. Four investigators (AJL, AA, EJLM and LA) independently gave each eligible study an overall rating of high, low or unclear risk of bias; disagreements were resolved by consensus.

Data Extraction

Four reviewers (AA, AJL, PN and LA) independently extracted relevant information from each eligible study using a standardised data extraction sheet and then proceeded to cross-check the results. The data extracted included the last name of the first author, publication year, study period, study region, level of study (national, state/provincial, local), age and gender of study participants, sample size (total as well as by sex and by number of regional subgroups), reported prevalence and/or incidence with 95% CIs, and prevalence and/or incidence figures by gender and age group, if available. When not directly stated, incidence rates were calculated using the population used to calculate prevalence rates; we estimated the exposure periods, assuming that the reference populations were stable throughout the given study periods.

Methodological design and data indicative of risk of bias assessment for all included studies were also extracted. Disagreements between reviewers regarding data extraction were resolved through discussion.

Statistical Analysis

Proportions provided by source documents were transformed and calculated with the Freeman-Tukey double arcsine method; estimations of both prevalence and incidence were carried out with the aid of random-effects meta-analyses weighted for inverse variance, following DerSimonian and Laird's method.[14] In studies that reported data for more than one-time point, only data reported at the last time point was included in the primary analysis. Summary estimates, along with their 95% CIs, were calculated for the prevalence and incidence rates of EoE among children and adults. The proportions of male and female patients (where reported) were compared using pooled odds ratio (OR) with a 95% CI.

Heterogeneity between studies was assessed by means of a chi-square test (Cochran Q statistic) and quantified with the I2 statistic. Generally, I2 was used to evaluate the level of heterogeneity, assigning the categories low, moderate and high to I2 values of 25%, 50%, and 75%, respectively.[15] Publication bias was evaluated with the aid of a funnel plot, the asymmetry of which was assessed through Begg-Mazumda's rank test[16] and Harbord's bias test.[17]

For the primary outcomes, planned subgroup analyses were performed based on the diagnostic criteria used in each document to define EoE (ie, those provided by either 2007,182 or 2017[1] guidelines), the source of data (hospital-based registries or administrative claims databases) and the year in which studies were carried out.

A subgroup analysis was performed with regard to quality (risk of bias). All calculations were made with StatsDirect statistical software version 2.7.9 (StatsDirect Ltd, Cheshire, UK).