June 27, 2019

solvents in the workplace and risk of autism spectrum disorders

Research published in Occupational & Environmental Medicine reported that maternal occupational exposure to solvents may increase the risk for autism spectrum disorder.

A Round Up accompanied this Before the Headlines. 

Title, Date of Publication & Journal

The CHARGE study: an assessment of parental occupational exposures and autism spectrum disorder by Erin C McCanlies et al. will be published in Occupational & Environmental Medicine at 23:30 UK time on Thursday 27th June

Study’s main claims – and are they supported by the data

This paper does not prove that maternal occupational exposure to solvents increases the risk for autism spectrum disorder (ASD) in children.

The way the study is designed means the authors cannot demonstrate that the association is not simply a false positive result (see limitations).

There is just enough in this paper to favour further research but not enough to make policy decisions or lifestyle changes. The authors do make this point in their press release and paper.

Strengths/Limitations

Strengths

The actual design and conduct of the study is well considered:

1. In determining whether a child has ASD, it would appear that all children were assessed by the same group of experts, which is good practice.
2. In determining maternal and paternal exposure, all exposure estimates were done by a group of experts separate to those who assessed the children for ASD.

• The authors acknowledge the limitations of their small sample size.

Limitations

• A major limitation of the study is that it cannot prove that solvents cause ASD under any circumstances. This is because the study is observational and not experimental. This is discussed in the paper and press release but is a very significant limitation.
• The study was not looking at the association between solvents and ASD – the study was looking at an association between sixteen different pollutants, one of which was solvents (7 pollutants did not have sufficient numbers of parents exposed to run the analysis).  The way the study is designed it will generate a false positive about 50% of the time. There are statistical corrections which can be made to rule out false positives. When the authors conduct these tests, they say “this association did not remain significant after correcting the p-values for multiple comparisons”. Therefore, it is not accurate to report that the authors have found a statistically significant association between solvents and ASD – they can’t rule out having found a false positive. ¹
• The frequency of maternal smokers in the ASD group was twice that of TD group.  No comment was made on this but this is likely a possible confounder and it should have been discussed, as previous studies have reported an association between smoking and ASD.
• Absolute risks should have been reported more. ASD is rare (1 in 68) and therefore even if solvents increased the risk of ASD by 1.5 times as claimed this number would only increase to around 1 in 45. In a school year with 140 children this would be the difference between 2 and 3 children with ASD.
• No analysis or discussion was given about the potential multivariate nature of exposures i.e. can you truly separate the effects of the 9 exposure categories? 

Glossary

False positives are when a statistical test result wrongly indicates that a particular condition or association is present.

ASD (Autism Spectrum Disorder) is a spectrum of developmental disorders which includes autism but is not the same thing as autism (although autism is by far the most common of the autism spectrum disorders, so the confusion is common). ASD includes Autism, Asperger Syndrome, Childhood Disintegrative Disorder (CDD) and Rett Syndrome.

TD means typically developing children

Any specific expertise relevant to studied paper (beyond statistical)?

Neither reviewer has any special expertise relevant to this review

Additional comments on the paper

¹ There are several concerning signals in the paper that indicate the link the authors claim to have found between solvents and ASD may not actually exist. For example: 

• The reported odds ratio between moderate solvent exposure and ASD is 1.85, but the reported odds ratio between high exposure and ASD is only 1.20. If a link between solvents and ASD actually existed, we would expect more exposure to solvents to lead to a greater risk of ASD.
• There is absolutely no link between paternal solvent exposure and ASD, and in fact paternal exposure to solvents actually appeared to protect against ASD, albeit non-significantly (in fairness to the authors, they suggest a reasonable explanation for why this might be). 
• The authors fail to find links between pollutants and ASD which previous studies have established actually do exist (for example, pesticides and metals). Therefore, we should be cautious when they claim to have discovered a brand new association nobody has seen before.
• All 9 maternal exposure categories showed elevated odds ratios with means varying between 1.1 & 1.5 according to figure 1.  Whilst none are statistically significant after allowing for multiple comparisons, if there was no effect at all, I would expect some odds ratios to be less than 1 so all 9 showing elevated odds is somewhat unusual. This could suggest either:

1. 1. these exposures are not independent of each other or
   2. there is hidden bias not accounted for in this study design e.g. the fact that the ASD group had twice as many smokers as the TD group.

Before The Headlines is a service provided to the SMC by volunteer statisticians: members of the Royal Statistical Society (RSS), Statisticians in the Pharmaceutical Industry (PSI) and experienced statisticians in academia and research.  A list of contributors, including affiliations, is available at http://www.sciencemediacentre.org/working-with-us/for-journalists/headlines-for-journalists/