We conducted a rapid systematic search of recent clinical and observational studies published from the date of publication of the review conducted by Coury et al. (01/08/2012)4 to present in PubMed via MEDLINE. The search strategy was the following: (Asperger’s OR Autism OR Autism spectrum disorder OR Autistic OR Pervasive developmental disorder OR PDD-NOS) AND (Abdominal pain/abdomen OR bloating OR Celiac OR Colitis OR Constipation OR Diarrhea OR Digestion OR Digestive disorders OR Disaccharidase OR Endoscopy OR colonoscopy OR Esophagitis OR Functional gastrointestinal disorder OR Gastroenterology OR Gastritis OR Gastrointestinal OR Gluten OR Gastroesophageal reflux OR Irritable Bowel Disease OR nausea OR reflux OR Vomiting).

We selected original articles that (1) reported data on the prevalence of GI symptoms or sufficient data to calculate it, (2) were conducted in a representative sample of children or adolescents with ASD (3) whose diagnosis of ASD was confirmed with a validated instrument, (4) published in English or Spanish and (5) for which the full text was available.

We excluded studies that preselected children and adolescents with GI comorbidities, whose samples did not include children or adolescents with ASD (e.g. coeliac disease, paediatric eating disorder, eosinophilic oesophagitis), or with samples that were not representative of children and adolescents with ASD for the purpose of estimating the prevalence of GI disorders (e.g. children with ASD who visited the emergency department due to constipation or who had received treatment before their GI symptoms were assessed).

We collected data on a form designed for the purpose, including the following fields: country, sample size, proportion of female patients, mean age and age range, instruments used to confirm the diagnosis of ASD and methods used for assessment of GI function.

To assess the quality of the studies, we used a tool developed for the critical appraisal of prevalence studies.13 We assessed quality based on 8 criteria, each rated on a scale from 0 to 1. In this critical appraisal system, 1 point was given for each of the following criteria that were fulfilled: (1) random sample or whole population, (2) unbiased sampling frame (i.e., census data), (3) adequate sample size (>30 subjects), (4) measures were the standard, (5) outcomes measured by unbiased assessors, (6) adequate response rate (>70%) refusers described, (7) confidence intervals and subgroup analysis, and (8) study subjects described. The possible total score ranged from 0 (poor quality) to 8 (high quality). Based on the total score, we classified studies as having a low (6–8), moderate (4–5) or high (0–3) risk of bias.

We pooled the data with generic inverse-variance weighting in an effects model,14 applying double-arcsine transformation to proportions to take into account the variability and heterogeneity in the prevalence rates reported in the included studies.15 Several studies16 have demonstrated that this adjustment may reduce the likelihood of false positives, especially when the number of studies is small. We present the main outcomes as proportions with the corresponding 95% confidence intervals (CIs) and the obtained heterogeneity statistic values.17

We assessed between-study heterogeneity with the Hedges g, considering results statistically significant if the P value was less than 0.10, and quantified it with the I2 index and the corresponding 95% CI.18 Index values of 25%–50%, 50%–75% and 75% or greater are considered indicative of low, moderate and high heterogeneity, respectively,19 indicating the degree to which the variance in the results obtained for a given exposure-disease association can be explained by sampling error. We carried out a subgroup analysis to explore the expected sources of heterogeneity in the meta-analysis of observational studies19 and a sensitivity analysis to determine the impact of individual studies on the overall results by omitting the data of each study one at a time.17 We did not perform a meta-regression, as the meta-analysis included fewer than 10 studies and the statistical power was insufficient.20

To assess the risk of publication bias, we calculated the fail-safe N, which is more precise than conventional funnel plots for meta-analysis of proportion studies21 and recommended when the analysis includes fewer than 10 studies.22 The fail-safe N indicates the number of nonsignificant, unpublished (or missing) studies that would need to be added to the meta-analysis to reduce an overall statistically significant result to nonsignificance, so that if the N is large relative to the number of observed studies, one can feel fairly confident in the conclusions.21 To make a graphical analysis of publication bias, we used the Doi plot and the Luis Furuya-Kanamori (LFK) index, which are particularly suitable for meta-analyses of proportions and offer a higher sensitivity and power than the funnel plot or Egger regression.23 As regards their interpretation, asymmetry in the Doi plot indicates potential publication bias, and the LFK index provides a numerical measure, with values ±1 considered to represent no asymmetry, values greater than ±1 but within ±2, minor asymmetry, and values greater than ±2, major asymmetry.

All the statistical analyses were performed with the R software, using the metaprop, metafor and dmetar packages for meta-analyses (https://www.r-project.org). We calculated two-tailed P values and considered values of less than 0.05 statistically significant, unless otherwise noted.

Fig. 1 presents a flowchart of the article selection process. The initial search yielded 1666 articles, which were reduced to 91 after applying the following filters: Clinical Study, Clinical Trial, Comparative Study, Controlled Clinical Trial, Multicenter Study, Observational Study, Pragmatic Clinical Trial, Randomized Controlled Trial, 2012/08/01–2022/3/14.

Figure 1.

Flow chart of article selection.

GI, gastrointestinal.

(0.28MB).

After reading the titles and abstracts, 46 studies were selected for reading of the full text, after which 38 were excluded for the various reasons given in Fig. 1. At the end, 8 articles were included in the quantitative analysis of the prevalence of GI symptoms in children and adolescents with ASD.

Table 1 presents the main characteristics of the 8 studies included in the quantitative review. Of all the studies, 37.5% were conducted in America,24–26 37.5%, in Asia27–29 and the remaining 25% in Europe.30,31 The sample sizes ranged from 12 to 689 children, with a mean size of 134 participants, which decreased to 54.7 with the exclusion of the largest study.31 The mean age was 7.7 years after adjusting for the sample size of each of the 7 studies that provided this information.24,25,27–30 Last of all, the proportion of female participants ranged between 0% and 34.8%, for an overall proportion of girls and female adolescents of 19.85% when the data were pooled.

In 7 of the 8 studies, the diagnosis of ASD was confirmed by physicians, psychiatrists or psychologists at the time of recruitment using different validated diagnostic instruments.25–31 This was done through an in-person interview with the patient or the parents in all studies, except in one in which 2 psychiatrists reviewed the psychiatric records of the children to ensure they met the diagnostic criteria for ASD of the International Classification of Diseases, Ninth Revision (ICD-9).30 We included a single study that did not verify the diagnosis at the time of inclusion because the sample was recruited from a register of children with ASD that had undergone a thorough assessment by a doctor with a broad range of diagnostic tools, including the Autism Diagnostic Observation Schedule (ADOS).24

There was substantial heterogeneity in the methods chosen to assess GI function between studies. Only 37.5% applied the paediatric version of the Rome III questionnaire (QPGS-ROME III),24,28,29 which is widely recommended for diagnosis of functional GI disorders.31 One study used a variation of this questionnaire (12.5%)27 and another analysed the discharge diagnoses of GI disorders in the national health records database, classified with the ICD coding system.30 The remaining 37.5% used different questionnaires that had not been validated to assessed various GI symptoms.25,26,31 The items more frequently included in the assessment across studies were constipation or hard stools (analysed in 87.5% of the studies),24–29,31 abdominal pain (75%),24–29 vomiting or nausea (75%),24–29 abdominal distension/flatulence/aerophagia/excess gas (75%)24,27–29,31 and diarrhoea or soft stools (62.5%).24–27,31 Some studies also included nonspecific categories (“other”).26,30

We ought to mention that the only study that analysed diagnosed GI disorders did not restrict the search to functional disorders,30 and neither did another study that also assessed functional and organic disorders, such as diseases of the pancreas or liver, coeliac disease or lactose intolerance.31

As regards the quality of the studies, 2 were classified as low risk of bias28,30 and 3 as high risk of bias.25,26,31 The main limitation found in all studies was the lack of reporting of confidence intervals for prevalence calculations and subgroup analysis results, in addition to the lack of a population-based sampling frame, with the exception of the study of Mouridsen et al.30 (Table 1, Appendix B, Supplemental Table 2).

The prevalence of GI symptoms ranged from 0%27 to 69%29 (Table 1; Fig. 2). We estimated an overall prevalence of GI symptoms of 33% (95% CI, 13%–57%), with significant heterogeneity between studies (g test: P<.01; I2=96%)

Figure 2.

Forest plot of the prevalence of GI symptoms in children and adolescents with ASD.

(0.14MB).

In the subgroup analysis conducted to identify the sources of heterogeneity, we that the reported prevalence of GI symptoms was greater in the studies conducted in America (38%; 95% CI, 7%–76%) or Asia (33%; 95% CI, 5%–70%) compared to those conducted in Europe (26%; 95% CI, 0%–71%). This difference, however, was not statistically significant (Fig. 3).

Figure 3.

Forest plot of the prevalence of GI symptoms in children and adolescents with ASD based on the continent where the study was conducted.

(0.19MB).

We also found a greater prevalence of GI symptoms in studies that applied the QPGS-Rome III QPGS questionnaire (55%; 95% CI, 21%–87%) compared to those that used other measures (21%; 95% CI, 3%–48%) (Fig. 4).

Figure 4.

Forest plot of the prevalence of GI symptoms in children and adolescents with ASD based on the instrument used for diagnosis.

(0.18MB).

The exclusion from the analysis of each of the studies one by one did not change the overall prevalence of GI symptoms obtained from the pooled data, which ranged from 28% (95% CI, 9%–53%) with the exclusion of the study by Ghodsi and Kheirouri29 to 42% (95% CI, 26%–59%) with the exclusion of the study by Pusponegoro et al.27 (Fig. 5). This indicates that none of the studies had a disproportionate impact on the overall prevalence of GI symptoms.

Figure 5.

Forest plot of the sensitivity analysis for the prevalence of GI symptoms in children and adolescents with ASD.

(0.19MB).

Last of all, Fig. 6 shows the Doi plot and the obtained LFK index of −2.87, indicative of major asymmetry, and therefore of a probable publication bias. However, the value of the fail-safe N was 730, which suggests that 730 studies with null results would be required to reduce the observed overall prevalence to nonsignificance.

Figure 6.

Doi plot for the prevalence of GI symptoms in children and adolescents with ASD.

(0.04MB).