Jun 26 2014

Pesticides and Autism

A study has been making the rounds on social media claiming an association between prenatal exposure to pesticides and the risk of autism and developmental delay. This means that I am getting asked by many people what the study actually shows. Spoiler alert – not much. But let’s break it down.

Autism spectrum disorder (ASD) is a neurological disorder involving brain development resulting in decreased communications among neurons in the brain and characterized by reduced social ability. Our current scientific understanding is that ASD is largely a genetic disorder. While environmental factors cannot be ruled out, it seems that genes are the primary factor. It’s reasonable to search for environmental risk factors, but so far none have been clearly established.

Those who feel there likely is an environmental factor also tend to believe that there is an autism epidemic – that the incidence of autism is increasing in a way that is not easily explained by genetics, and therefore suggests and environmental factor. While it is uncontroversial that the number of ASD diagnoses has been increasing over the last two decades, this does not necessarily mean that the true incidence of ASD has been increasing.

The evidence actually shows that diagnostic substitution, broadening of the definition of ASD, and increased surveillance account for much of the increased recorded incidence. It’s possible that changes in diagnostic behavior entirely accounts for the apparent increase. It’s also possible that a subset is due to a true increase, but that has not been clearly established.

This still leaves us with the conclusion that an environmental factor is possible in ASD, but not necessary.

What does this current study show? The study in question is a case-control study using data from the CHARGE study and data about pesticide use in California. A case-control study is a retrospective epidemiological study. It looks at two or more populations based upon whether or not they have a condition, in this study there are three groups – ASD, developmental delay (DD) and typical. The groups are then compared based on exposure to a potential risk factor to see if it is correlated with the condition.

In this study the authors looked at proximity to pesticide use prior to conception and during each trimester of pregnancy. They concluded:

“This study of ASD strengthens the evidence linking neurodevelopmental disorders with gestational pesticide exposures, and particularly, organophosphates and provides novel results of ASD and DD associations with, respectively, pyrethroids and carbamates.”

Orac has already reviewed this study and I agree with his assessment – this conclusion is not justified by the data presented.

First, there is a fatal flaw in the study design: the authors used proximity to pesticide use (either within 1.25, 1.5, or 1.75 km of the use) to estimate pesticide exposure. However, they made no attempt to justify their core assumption – that this proximity significantly affects pesticide exposure. There were no measurements of specific pesticides in the blood or urine of mothers, and no measurement of pesticides in the local environment where they lived. Proximity was the only measure.

If living 1.5 vs 3 km away from a farm using a certain pesticide has no significant affect on exposure, then the entire study evaporates.

Also, the data itself are far from convincing. The three groups in the study were relatively small for a case-controls study – ASD = 486, DD = 168, and typically developing referents = 316. Keep in mind these break down further into smaller and smaller groups based upon type of pesticide and distance from use.

If you look at table 3 in the study, you will see a summary of the data. This is not impressive – many of the confidence intervals overlap with 1, meaning the difference is not statistically significant. For those that are, there is no clear relationship to distance or to trimester of exposure. The authors claim there is a statistical distance relationship, but it is inconsistent. Overall effects sizes and statistical significant are also low.

My overall impression of the data is that we have a relatively small study with barely significant results, small effect sizes, and an inconsistent pattern of effects. This is all compatible with random noise, maybe with a little p-hacking thrown in. With such a complex data analysis, it is certainly possible to inadvertently extract apparently significant results by making numerous comparisons and using various methods and then choosing the one that “looks” the best.

I also noted that males overall had a higher pesticide exposure than females. The authors note:

Overall, exposure to pesticides during gestation was slightly more common for male children than female children (31% vs. 26%, p=0.004).”

I couldn’t find anywhere else a note that this was accounted for in the data analysis. It’s possible I just missed it, but if not, it should be noted that male children have a higher incidence of ASD than female, and so this could be a confounding factor.

Conclusion

Overall this study is small and underwhelming. The data are not convincing, and in my opinion the authors overstate their conclusions. At best this is a preliminary study, of the kind that often does not pan with larger and more rigorous follow up studies.

Having said that, I would note that the default position is for pregnant women to minimize their exposure to any potential toxin. Pregnancy certainly is a vulnerable period, and I am not saying the there are no concerns with pesticide exposure during pregnancy.

This study does not establish, however, that living within 1.5 km of pesticide use contains any health risk.

It is unfortunately that such studies are often reported with a simplistic bottom line – pesticide exposure is associated with autism, sometimes with a “may” thrown in. This is a problem in general with reporting preliminary data in the lay media. Such reporting is much more likely to be misleading than useful. Most people will just remember the possible association, and either not read or remember the detailed analysis explaining why the study did not establish the association.

Share

16 responses so far

16 Responses to “Pesticides and Autism”

  1. MikeBon 26 Jun 2014 at 4:35 pm

    …the authors used proximity to pesticide use (either within 1.25, 1.5, or 1.75 km of the use) to estimate pesticide exposure. However, they made no attempt to justify their core assumption – that this proximity significantly affects pesticide exposure. There were no measurements of specific pesticides in the blood or urine of mothers, and no measurement of pesticides in the local environment where they lived. Proximity was the only measure.

    Double-take.

    Triple-take.

    What? How does such stuff even get published, let alone circulated so widely?

  2. davezon 26 Jun 2014 at 4:48 pm

    I am confused by why the results were “weighted.” It appears to me from table 2 that without the weighting, autism was slightly less common if living nearer the area of pesticides. They provide the following explanation for the weighting,
    “aThe purpose of the development and utilization of CHARGE survey weights was to correct for the non-sociodemographically representative participation, i.e., the differences in participants vs. non-participants with regard to key sociodemographic factors such as maternal education, insurance payment type, birth regional center, birth place of mother, and child race. Survey weights are based on the probability of participation in the study”
    This appears to me to manipulating the data to get the desired result. I am at a loss for why this weighting needed to be done and how. I am unaware of why insurance type, or birth center should affect autism rates. What am I missing?

  3. Fair Persuasionon 26 Jun 2014 at 6:39 pm

    Pesticide proximity is merely one possible pollutant. California’s air quality should be assessed for all types of contaminants present at the study locale. Personal habits of the pregnant females were ignored in this study. To what degree were the pregnant females inheritors of genetic defects; were some autistic pregnant females participants in this study?

  4. BBBlueon 26 Jun 2014 at 10:40 pm

    Yet another example of a weak, retrospective study misrepresented by weak science journalism as in…

    “UC Davis study finds link between pesticides, autism” http://bit.ly/UO0Cuj

    UC Davis has always been the gold standard for agricultural research, and the CHARGE study will carry a bit of extra weight because of it’s connection to that university. Unfortunately, the UC system in general, and UC Davis in particular is not nearly as discriminating as it once was in terms of the quality of its product and associations.

  5. sonicon 27 Jun 2014 at 12:44 am

    It appears the data being analyzed was not gathered for the purpose of this analysis.
    That makes the analysis a ‘p hack’ by definition, doesn’t it?
    While a study like this might give one an interesting lead for a possible study, the analysis shouldn’t be taken as a conclusion in any case.

  6. airdirtyon 27 Jun 2014 at 12:58 am

    It would have been nice to see blood serum and urine sample data. Although that still could be hard to quantify because that data would have to correlate to other variable as to when the fields were sprayed, wind direction, were they home at the time, concentration levels, etc.

    This statement below by you involves a study of only 38 people. How can you assert that the CHARGE study ‘The three groups in the study were relatively small for a case-controls study – ASD = 486, DD = 168, and typically developing referents = 316″ is small?

    “The evidence actually shows that diagnostic substitution, broadening of the definition of ASD, and increased surveillance account for much of the increased recorded incidence. It’s possible that changes in diagnostic behavior entirely accounts for the apparent increase. It’s also possible that a subset is due to a true increase, but that has not been clearly established.”

    A better study I listed below. But still, they both do not address the remaining ~75% that have an ASD diagnosis. It cannot be argued against that autism is on the rise–sure, better screening methods have helped, but still the rate increases. Why?

    http://www.ncbi.nlm.nih.gov/pubmed/19737791/

  7. steve12on 27 Jun 2014 at 1:09 am

    MikeB:

    Of course, it would be better to see have the physio measure, but the real problem is the inconsistent effects, small effect sizes + small N.

    If the study showed that your risk of having autism increased 10 fold for every 100 meters closer you were to a farm (or some such) and we had x5 the N, this would be compelling w/o the physio measure.

    I think it’s a fair idea to start off with this less expensive step form existing data. The problem is the existing data didn’t justify the conclusion.

  8. MikeBon 27 Jun 2014 at 6:48 am

    Steve12, thanks. We lay people are truly f***** when it comes to evaluating “studies,” which is why I haunt Neurologica, SBM, and other skeptics sites.

  9. Steven Novellaon 27 Jun 2014 at 9:44 am

    airdirty – I did not link to that one study as a reference. I linked to a previous blog post I wrote on this topic that had multiple references. I was referring to all the evidence taken together – not just that one study.

    The King study you reference was only looking at one factor (diagnostic substitution), so of course it only accounts for a subset (26%).

    Other studies I referenced show that if you apply the same diagnostic methods to different cohorts (from different birth years) you get the same incidence of ASD – strongly implying not true rise in ASD, only changes in diagnostic patterns.

  10. azzerismon 27 Jun 2014 at 11:46 am

    I teach statistics at the first year level of college, and what I noticed was the percentage of data that was not used. Secondly, the sampling method is somewhat flawed. There are families living all over the United States near farmland, why not pick Iowa or Kansas or the Dakota’s to get random sampling of data? This study, if it is flawed can be very easily overturned by taking samples elsewhere in the United States. The reason being that the conclusion is overstated. It seems like someone wanted a strong conclusion, not based on discovery.

  11. steve12on 27 Jun 2014 at 12:04 pm

    YOu’re still right in principle, MikeB – if the next step showed no difference in concentration in the people by farm proximity, out goes that interpretation. This sort of approach could never be the last word.

  12. BBBlueon 27 Jun 2014 at 5:06 pm

    airdirty,

    “It would have been nice to see blood serum and urine sample data.”

    The recent report references a 2006 report by saying the 2006 report should be consulted for more details of the protocol. The protocol described in the 2006 report includes blood and urine sampling. That being the case, it is curious that blood and urine sampling was not done for the recent study, or if it was, why the data were not included in the report.

  13. PerpetualLearneron 28 Jun 2014 at 7:51 am

    Does the study include any multiples? As the mother of twins (fraternal)–one with an ASD and one without–it would seem a helpful population to consider when looking at environmental exposures.

  14. BillyJoe7on 28 Jun 2014 at 6:34 pm

    BBBlue,

    Hey, Badfinger’s “Baby Blue”.
    And, I’m guessing, a “Breaking Bad” fan.

    http://www.youtube.com/watch?v=TkA7xQb6uPk&feature=kp

  15. BBBlueon 29 Jun 2014 at 1:33 pm

    Dr. Novella,

    In your March 27, 2014 post, When Does Autism Begin, you stated that “…autism is a dominantly genetic from of disorders” and cited a 2004 review in support of that claim. Since that information is a bit long in the tooth, I was interested to find more recent information on the subject and found these:

    Non-Inherited Mutations Spotlight Role of Environment

    The Eichler team turned up clues to how environmental factors might influence genetics. The high turnover in a male’s sperm cells across the lifespan increases the chance for errors to occur in the genetic translation process. These can be passed-on to the offspring’s DNA, even though they are not present in the father’s DNA. This risk may worsen with aging. The researchers discovered a four-fold marked paternal bias in the origins of 51 spontaneous mutations in coding areas of genes that was positively correlated with increasing age of the father. So such spontaneous mutations could account for findings of an earlier study that found fathers of boys with autism were six times – and of girls 17 times – more likely to be in their 40’s than their 20’s.

    The Familial Risk of Autism

    Heritability of ASD and autistic disorder were estimated to be approximately 50%.

    In the case of the Eichler study, while it certainly does not support Dr. Hertz-Picciotto’s narrative, if one reads between the lines as the NIH press release seems to do, by implicating de novo mutations, it implies environmental factors can affect the incidence of point mutations in sperm cells, and thereby affect the probability of producing autistic offspring.

    Again, I understand that is not evidence of a link to environmental contaminants, and the focus is on the father, not the mother, as in the CHARGE study, but if point mutations are an important contributing factor, doesn’t that at least add to the plausibility of claims that there may be an environmental component?

  16. BBBlueon 29 Jun 2014 at 1:37 pm

    BillyJoe7,

    Yes, a fan of all of the above, and also Benjamin Buford Blue.

    https://www.youtube.com/watch?v=iLkNPjbaPTk

Trackback URI | Comments RSS

Leave a Reply

You must be logged in to post a comment.