Association among obesity, overweight and autism spectrum disorder: a systematic review and meta-analysis

Obesity, overweight and autism spectrum disorder (ASD) remain serious public health problems. Although lots of studies have recently explored the association among obesity, overweight and ASD, the findings are inconsistent. Thus, we conducted a meta-analysis of epidemiological studies to examine the association among obesity, overweight and ASD. PubMed, Embase, and the Cochrane Library were used for literature searches to identify eligible studies published in English before November 15, 2016. Relevant studies estimating the association among obesity, overweight and ASD were included. Fifteen studies encompassing 49,937,078 participants and 1,045,538 individuals with ASD were included in this study. A random effects model was chosen to synthesize the effect sizes of individual studies. The prevalence of obesity was significantly higher in individuals with ASD than in controls (OR = 1.84, 95% confidence interval [CI]: 1.37–2.48, P < 0.001). However, the prevalence of overweight in individuals with ASD was not significantly different from that in controls (OR = 1.07, 95% CI: 0.83–1.38, P = 0.62). Both sensitivity analysis and publication bias testing revealed that the findings were robust. The meta-analysis showed a significant association between obesity and ASD. However, no significant association was identified between overweight and ASD.

is scarce. ASD begins at an early age and lasts throughout the lifespan, which increases the economic burden on both family and society. Thus, ASD is also a significant global public health problem.
Individuals with ASD have several issues, including food selectivity, gastrointestinal symptoms, less physical activity, and medication use, which may result in abnormal anthropometric measurements [20][21][22][23] . Recently, some studies have focused on the association among obesity, overweight and ASD 9,[24][25][26] . However, the findings are inconsistent. Some scientists have reported that BMI in individuals with ASD is not different from that in controls 9,24 , while other scientists have reported the prevalence of obesity and overweight is higher in individuals with ASD than in controls 25,26 . Therefore, we performed a meta-analysis to examine the association among obesity, overweight and ASD. We believe that a comprehensive evaluation of this critical public health problem may promote efforts to develop adequate interventional strategies in this population.

Results
Literature search. We initially identified a total of 2,720 potential articles, of which 936 were from PubMed, 1,670 were from Embase, 109 were from the Cochrane Library and 5 were from screening the references in the included articles. We identified 72 articles after excluding 421 duplicate articles, 2,177 irrelevant articles, 6 reviews and 44 letters/meetings. Then, those 72 articles were screened carefully. Eleven articles were excluded because their topics were irrelevant. Twenty-three articles were excluded because they lacked a control comparison group. Seventeen articles were excluded because they did not provide sufficient data and 6 articles were excluded because of overlap. Ultimately, 15 articles encompassing 49,937,078 participants and 1,045,538 individuals with ASD were included in this study. The flow diagram of the literature search is shown in Fig. 1  A total of 2,720 potential studies were initially identified in databases. 72 studies were identified after excluding 421 duplicate studies, 2,177 irrelevant studies, 6 reviews and 44 letters/meetings. Then, 57 studies were removed after a careful screening. Ultimately, 15 studies were adopted in this meta-analysis. individuals with ASD. Ten studies were carried out in the United States [24][25][26][27]29,30,[34][35][36][37] , two in Brazil 9,31 , one in Turkey 32 , one in Spain 33 , and one in China 28 . The sample sizes ranged from 40 33 to 986,352 30 individuals with ASD and from 19 9 to 48,598,466 30 controls. Additionally, the mean ages of the samples ranged from 2 years old 25,27,34 to 29.4 ± 12.1 years old 35 .

Quality evaluation. As presented in
were considered moderate quality. The mean score of all studies was 7.1.

Publication bias.
The funnel plot showed slight asymmetry (Figs 4 and 5). However, Begg's test (P = 0.913) and Egger's test (P = 0.925) showed no significant publication bias among the studies examining the association between obesity and ASD. Similarly, Begg's test (P = 0.246) and Egger's test (P = 0.773) showed no significant publication bias among the studies examining the association between overweight and ASD.
Subgroup analysis and sensitivity analysis. Subgroup analyses by study design, geographic locations, method of ascertaining of ASD, method of ascertaining BMI, and adjustment factors for age and gender were shown in Table 3. Among the studies examining the association between obesity and ASD, the tests for subgroup differences were not significant in the subgroup analyses by study design, geographic locations, method of ascertaining ASD, method of ascertaining BMI, and adjustment factors for age and gender.
Among the studies examining the association between overweight and ASD, the tests for subgroup differences showed significant different in the subgroup analyses by study design, geographic locations, and method of ascertaining BMI.      The sensitivity analysis was conducted by removing each study individually and comparing the results from the rest of studies with the overall result. The results showed that there was consistently a higher prevalence of obesity in individuals with ASD than in controls, while there was no significant different in the prevalence of overweight between individuals with ASD and controls. The omission of any single study did not change the overall result.

Discussion
This meta-analysis included 15 studies with a total of 49,937,078 participants and 1,045,538 cases of ASD. The results showed that obesity but not overweight was associated with ASD.
In this study, we adopted a random-effects model to synthesize the effect sizes of individual studies because significant heterogeneity between studies existed. Sensitivity analysis revealed the findings were robust. In addition, both Begg's and Egger's tests indicated no significant publication bias existed.
An association between obesity and ASD was confirmed in this study. These findings are supported by previous studies. First, individuals with ASD often have eating problems, such as food selectivity and specialized dietary habits, which make healthy dietary interventions less effective 38,39 . Second, individuals with ASD spend less time on physical activities and more time on sedentary activities 29 . They also have difficulty in participating in standard obesity prevention initiatives due to behavioral problems. Third, individuals with ASD have more chances to be treated with various antipsychotic medications, which are known to be associated with weight gain 40,41 . Fourth, individuals with ASD also have comorbidities associated with obesity, such as sleep problems, gastrointestinal disorders, ADHD 13,23,42,43 . Finally, some individuals with ASD have been reported to have 16p11.2 or 11p14.1 microdeletions, which encompass genes related to obesity susceptibility 44,45 . These results indicated that obesity is related to ASD, which is consistent with our findings.
However, we found significant heterogeneity in this meta-analysis. Thus, we performed subgroup analyses to find the source of heterogeneity. But we could not find the source of heterogeneity among the studies examining the association between obesity and ASD (Table 3). Residual confounding factors such as different ages and genders across studies need to be considered. In addition, we found three factors that may explain the heterogeneity among the studies examining the association between overweight and ASD. First, the overall result indicated that there was no difference in the prevalence of overweight between individuals with ASD and controls in the USA samples, whereas the prevalence of overweight in individuals with ASD was lower than that in controls in the non-USA samples ( Table 3). The result may be because of different genetic backgrounds and lifestyles. Second, different study designs (case-control or cross-section) were a cause of the heterogeneity (Table 3). Third, the different methods of ascertaining BMI (objectively measured, extracted from medical records or parentally reported) used in these studies contributed to heterogeneity ( Table 3). The measures of ascertaining BMI in several studies biased the results because the data were collected from parental reports rather than objectively measured.  Table 3. Summary of meta-analysis results on the association among obesity, overweight and ASD.
Although the heterogeneity is high, the analysis has certain obvious advantages. First, we are the first to conduct a meta-analysis to explore the association among obesity, overweight and ASD. Second, the meta-analysis included a huge amount of samples, making it more likely to draw a reasonable conclusion about obesity and overweight in ASD. Third, the sensitivity analysis showed that removing any study did not change the final results, suggesting that our findings were robust. Finally, publication bias was not found, which increased the reliability of the findings.
However, some limitations existed in this study. First, weight status was reported by parents in three studies. Therefore, the BMI may not be accurate in those studies. Future work should attempt to have the BMI measured objectively by trained clinicians. Second, there was significant heterogeneity across studies, which may reduce the conclusiveness of the results. Finally, adjustments were not made for confounding factors including dietary habits, physical activity participation, family history of obesity and secondary conditions. ASD is a heterogeneous disorder. Secondary conditions, including physical status, mental health status, developmental delay, intellectual and learning disabilities, attention deficit/hyperactivity disorder, IQ and drugs use might influence the association between obesity and ASD 30 . Therefore, future research should systematically adjust for a broad set of possible confounding factors.
In conclusion, this meta-analysis showed a significant association between obesity and ASD. However, no significant association was identified between overweight and ASD. Further prospective studies with more accurate measures of weight status and better control of confounding factors are warranted.

Materials and Methods
Strategy of literature search. We performed a literature search of PubMed, Embase and the Cochrane Library for the potential articles. The search was restricted to articles published before November 15, 2016. The search terms were as follows: ["ASD" OR "autism spectrum disorder" OR "autistic disorder" OR "autism" OR "Asperger syndrome" OR "pervasive developmental disorder"] and ["obesity" OR "obese" OR "overweight" OR "adipose" OR "adiposity" OR "body weight" OR "body mass index" OR "BMI"]. Furthermore, we searched the references of the related articles to attain other potential studies.
Only studies published in English were considered. In addition, we only considered studies performed on human. We reviewed the titles and abstracts to exclude the irrelevant studies. Then, we read the text carefully based on the inclusion criteria. If two authors reached inconsistent conclusions about inclusion/exclusion, a third author would be asked to make a decision.

Study selection criteria.
Any study that reached all the following criteria was included: (1) assessing the association among obesity, overweight and ASD; (2) reporting the raw data or OR with 95% CI; and (3) case-control, cohort or cross-sectional designs.
The following studies were excluded: (1) reviews, non-human studies, case reports, case-only studies, or meetings/letters; and (2) overlapping data. Data extraction. The data were extracted from the studies as follows: first author's surname, year of publication, country, study design, numbers of cases and controls, number of males, age, methods of ascertaining ASD and BMI, outcome variable and adjusted confounders. If multiple studies contained overlapping data from the same population, the one with the largest sample size was adopted.
Quality evaluation. We conducted Newcastle-Ottawa Scale (NOS) to evaluate the quality of the studies.
The NOS evaluated the following aspects: selection of participants, comparability, exposure of participants, and outcome. The score ranged from 0 to 9. Scores of 7-9 indicate high quality, scores of 4-6 indicate medium quality and scores of 0-3 indicate low quality. When raters disagreed, the inconsistencies were settled by discussion.

Statistical analysis.
We combined the effect sizes by using ORs to assess the association among obesity, overweight and ASD. We calculated the ORs by using a fixed-effects model or a random-effects model. If the heterogeneity was low, the fixed-effects model was used. If the heterogeneity was high, the random-effects model was chosen. I 2 and the Q statistic were adopted to evaluate the heterogeneity. The heterogeneity was divided into the following levels according toI 2 value: low (I 2 : 25%-50%), moderate (I 2 : 50-75%) and high (I 2 : >75%). The Q statistic was viewed significant if P < 0.1.
The funnel plot was visually inspected to evaluate publication bias. In addition, both Begg's and Egger's tests were conducted to evaluate publication bias. Publication bias was viewed significant if P < 0.05. Subgroup analyses were conducted to identify the sources of heterogeneity based on study design (case-control or cross-section), geographic locations (USA or non-USA), methods of ascertaining ASD (parental report or standard assessment), methods of ascertaining BMI (objective measurement or other methods), and adjustment factors for age and gender (yes or no). Sensitivity analysis was conducted by individually omitting one study from statistical analysis. Then, the ORs and 95% CIs of the remaining studies were compared to the overall result. Statistical analyses were conducted with Stata 12.0 (Stata Corp, College Station, Texas, USA).