A Meta-Analysis of Maternal Smoking during Pregnancy and Autism Spectrum Disorder Risk in Offspring

The association between maternal smoking during pregnancy and autism spectrum disorder (ASD) risk in offspring has been investigated in several studies, but the evidence is not conclusive. We, therefore, conducted this meta-analysis to explore whether an association exists between maternal smoking during pregnancy and ASD risk in offspring. We searched PubMed, Embase, Web of Science, and the Cochrane Library for studies of maternal smoking during pregnancy and ASD risk in offspring up to 10 June 2015. The random-effects model was used to combine results from individual studies. 15 observational studies (6 cohort studies and 9 case-control studies), with 17,890 ASD cases and 1,810,258 participants were included for analysis. The pooled odds ratio (OR) was 1.02 (95% confidence interval (CI): 0.93–1.13) comparing mothers who smoked during pregnancy with those who did not. Subgroup and sensitivity analysis suggested the overall result of this analysis was robust. Results from this meta-analysis indicate that maternal smoking during pregnancy is not associated with ASD risk in offspring. Further well-designed cohort studies are needed to confirm the present findings.


Introduction
Autism Spectrum Disorders (ASDs) are a group of heterogeneous neurodevelopmental conditions characterized by social and communication difficulties, and by restricted and repetitive interests and behaviors [1]. The ASDs include autistic disorder, Asperger syndrome, atypical autism, and other or unspecified pervasive developmental disorders [2]. The global prevalence of ASDs is estimated to be 1 person in 132, accounting for more than 7.7 million disability-adjusted life years worldwide in 2010 [3]. Although the cause of ASDs still remains unknown, they are regarded as multifactorial, with many risk factors acting together to produce the phenotype [4]. The identification of modifiable environmental risk factors is of great importance for the primary prevention of ASDs.
Smoking during pregnancy may lead to numerous adverse outcomes in the offspring, such as low birth weight, placenta abruption, and sudden infant death syndrome [5]. Maternal smoking is also a risk factor for several neurodevelopmental disorders, including attention-deficit, conduct disorder, and antisocial behavior [6]. Several studies have explored the relationship of maternal smoking during pregnancy and ASDs risk, but the results were inconclusive. A recent meta-analysis by Roson et al. concluded that there was no association between maternal prenatal smoking and ASD in offspring based on 15 observational studies [7]. However, Roson et al. [7] only searched two databases which missed several eligible studies. In addition, Rosen et al. misunderstood the difference between the period of "prenatal" and "pregnancy", and they inappropriately included studies that only reported smoking during pregnancy but not specified whether it was during the prenatal period or not. We therefore performed this study to systematically assess the association of maternal smoking during pregnancy (including both prenatal and perinatal smoking) and ASDs risk.

Data Sources and Searches
This meta-analysis was performed according to the Meta-Analysis of Observational Studies in Epidemiology guidelines [8]. A systematic search of PubMed, EMBASE, Web of Science, and Cochrane Library up to 10 June 2015 was conducted to identify relevant studies regarding the association between maternal smoking during pregnancy and ASDs risk. A reference list of the related studies were further screened for any additional literatures. The search terms for PubMed were: (smoke OR smoking OR tobacco OR risk factors) AND (autism spectrum disorder or autism) AND (pregnancy OR prenatal OR perinatal OR maternal). Similar search strategies were used in other databases. We applied no language restriction in the process of literature search and selection.

Study Selection
Studies were included if they met the following criteria: (i) had a cohort or case-control design; (ii) reported the association between maternal smoking during pregnancy (including prenatal or perinatal smoking) and ASD risk in their offspring; and (iii) reported risk estimates and 95% confidence interval (CI). Only the study with the most complete data was included if data was reported more than once in different literatures. Non-human studies, letters, case reports, conference abstract, review, and studies that with insufficient information were excluded.

Data Extraction and Quality Assessment
Two reviewers independently extracted data, any disagreements were resolved by discussion. The following data were extracted: last name of the first author, publication year, study design, study location, age, sex, number of cases and total population, risk estimates adjusted for the maximum number of confounding variables with corresponding 95% CIs, and confounding factors that were being adjusted for in the analysis. We also extracted any dose-relationship data for smoking and ASDs risk.
The methodological quality was assessed using the Newcastle-Ottawa Scale (NOS) [9]. The NOS has been widely used for quality assessment of cohort and case-control studies, a maximum of nine scores were assigned to eight items which indicated the methodological quality of each study. We divided the study quality into three categories (1) high quality (scored 7-9); (2) moderate quality (scored 4-6); and (3) low quality (scored 0-3).

Data Analysis
We combined the results using the OR as a measure of the association between smoking and ASD risk. Heterogeneity was investigated by the Cochrane Q statistic (significance level at p < 0.10) and the I 2 statistic [10,11]. The random effect model was used to calculate the pooled OR if p > 0.10 and I 2 ≤ 50%, otherwise the fixed effect model was chosen [12]. Sensitivity analysis was performed to assess whether any individual study significantly affected pooled estimates by omitting one study in each turn. Additionally, subgroup analyses were performed according to study design, study location, and whether they were adjusted for confounding variables to examine the source of potential heterogeneity. We used both Begg's test [13] and Egger's test [14] to detect publication bias. Stata version 11.0 (Stata Corporation) was used for all analysis. All statistical tests in this study were twosided with a significance level of 0.05, unless otherwise specified. Figure 1 shows a flow diagram of the selection process. A total of 1889 literatures were identified by the search strategy. After removal of duplicate literatures, 1156 articles were left for screening. By screening the titles and abstracts, 1140 articles were excluded as they were not clearly relevant. After detailed evaluation of the 16 full texts, one article [15] that reported passive smoking and two articles [16,17] that did not reported risk estimates was excluded. Due to the fact that one article [18] met the inclusion criteria reported results based on two separate cohort studies, 14 articles [6,[18][19][20][21][22][23][24][25][26][27][28][29][30] with 15 studies were included for meta-analysis.  Table 1 listed the characteristics of the included studies. Of the 15 include studies, there were six cohort studies [18,20,21,26,30] and nine case-control studies [6,19,[22][23][24][25][27][28][29], with a total of 17,890 cases and 1,810,258 participants involved. 10 studies [6,18,22,23,[25][26][27][28][29] were from Europe (Denmark, Sweden, Finland, Norway and Poland) while five studies [19][20][21]24,30] were from America (Canada and US). The information about smoking was generally collected during pregnancy, and ASDs were ascertained by medical records in most of the included studies. Overall, the methodological quality was good (average score: 7.3), with 12 studies [6,18,19,[21][22][23][24][25]27,28,30] in high quality and three studies [20,26,29] in moderate quality (Supplemental Table 1).

Main Analysis
The meta-analysis data that explored the effect of maternal smoking during pregnancy on the risk of ASDs is shown in Figure 2. A statistically significant heterogeneity was found across studies (pheterogeneity < 0.01, I 2 = 67.3%). The pooled OR estimates based on 15 studies showed that maternal smoking during pregnancy was not associated with an increased risk of ASDs (OR = 1.02; 95% CI: 0.93-1.13).

Figure 2.
A forest plot of the association between maternal smoking during pregnancy and ASD risk. Table 2 presents the results of subgroup analysis stratified by study design, study location, and adjustment for confounders. Maternal smoking during pregnancy was associated with significantly reduced risk of ASDs in the North American populations (OR = 0.90, 95%CI: 0.83-0.97), but not European populations (OR = 1.11, 95%CI: 0.99-1.24). In subgroup analysis by study design and adjustment for confounders, the difference in pooled ORs across strata was not statistically significant. For the sensitivity analysis, the pooled ORs did not vary materially ranging from 1.00 (95% CI: 0.91-1.10) when omitted Hultman et al. study [23] to 1.05(95% CI: 0.95-1.15) when excluded Dodds et al. study [21].

Publication Bias
Begg's test and Egger's test suggested no significant publication bias among the included studies (all p > 0.05).

Discussion
The present study used a comprehensive search strategy and explored the relationship between maternal smoking during pregnancy and ASDs based on 15 observational studies. However, the overall results did not suggest a significant association between maternal smoking during pregnancy and ASDs risk.
Several hypotheses of association between maternal smoking during pregnancy and ASD are plausible. First, nicotine is the chemical that most probably has adverse effects on brain development among thousands of ingredients of tobacco smoke [31] and its effects are thought to occur via its action on nicotinic acetylcholine receptors [32]. Second, women with lower socioeconomic status (SES) have a higher tendency to smoke during pregnancy compared with those with higher SES [33,34]; nevertheless, low parental SES is associated with an increased risk of ASD [35]. A third explanation is the transmission of genetic risk from mothers to their offspring, as ASD is highly genetic [4] and smoking during pregnancy is more likely to happen in women with a predisposition to mental health and behavioral problems [36]. Another hypothesis was that maternal smoking related high levels of intrauterine testosterone might contribute to the increased ASD risk [37].
Despite all the hypotheses mentioned above, the pooled OR failed to find a positive association between maternal smoking during pregnancy and ASDs. However, in the subgroup analysis, a significant negative association was observed in the North America. Although the association was not statistically significant in the individual studies of this subgroup, the increased sample size narrowed the confidence interval of the pooled result and thus it became statistically significant. Nevertheless, further studies are needed to examine this contrary finding, considering that the results were not fully adjusted and potential bias may exist due to the design of included studies.
Our study has several limitations that deserve consideration. First, because this meta-analysis was based on observational studies, our results were prone to recall and selection bias inherent in the included studies. In addition, the potential confounding from other risk factors could not be ruled out, especially when considering that some studies only reported crude risk estimated without adjustments. Second, potential misclassification of maternal smoking may lead to incorrect estimates of associations. The smoking status was generally collected at early pregnancy and was based on self-report, there was a possibility that participants tend to narrow their smoking status and the smoking habit might change during pregnancy. Third, high statistical heterogeneity was found across the studies that may weaken the strength of our findings; however, the subgroup analysis, as well as sensitivity analysis, showed that results from our meta-analysis was robust. Finally, the included studies of our meta-analysis were mainly conducted in the North Europe and the North America. The findings of this study had limited implications for other populations, like the Africans and Asians, as both the genetic and environmental factors in relation to autism varied among different ethnicities.

Conclusions
This meta-analysis indicates that there is no association between maternal smoking during pregnancy and ASD. Given the limited number of cohort studies included in this study, the present findings should be confirmed in further prospective cohort studies.

Supplemental Material
Supplemental Table 1. Quality assessment based on the Ottawa-Newcastle Scale.