Epidemiology of Autism Spectrum Disorder

 Epidemiology deals in correlations that are subject to confounding. Think tobacco and lung cancer.

Although studies should attempt to account for confounding, confounding can also be exploited to explain away inconvenient data as tobacco company claims that correlations do not prove causation illustrated. This underpinned the famous Doubt is our Product tobacco company strategy, which pharmaceutical companies now apply in debates about the hazards of their products – illustrated perhaps by the Ahlqvist study below.

Confounding can make it difficult to explore the possibility of a link between Acetaminophen – Paracetamol and conditions like Autism Spectrum Disorder (ASD) or ADHD.  Why? Well both conditions are on the rise anyway and part of this rise may be down to fashion.  ASD sounds like a gentler label than mental handicap and it also comes with access to disability benefits and special treatments and this has almost certainly played a part in making it fashionable.

How do we control for things like this?

In the case of states of neurodevelopmental delay in children linked to Acetaminophen taken in pregnancy, there is a growing body of epidemiological studies, including prospective and retrospective studies, case control and cohort studies, studies specifically looking at ADHD or ASD with some noting other indicators of neurodevelopmental delay en passant, as well as studies looking at biomarkers. These studies have been conducted in a number of different countries by independent research groups, some of whom are very highly credentialed.

A list of most of these studies up to the end of last year can be found here ASD Epi Studies.

The key paper perhaps is by Ann Bauer and Colleagues in 2021.  This is not key because it presents more, or new, or better data but because it marks a moment where a group of distinguished researchers, acknowledging there are lots of confounders, pinning their colours to the mast, have come out and stated that, while lots of confounders might make this risk look small, there is enough evidence here that women who are pregnant should be warned.

For any woman to work out what best to do, she needs to know that these Epidemiological studies do not stand on their own. There are also animal studies that show clearly that Paracetamol causes neurodevelopmental delay in offspring, that Acetaminophen has a toxic metabolite – NAPQI – and that this drug has epigenomic effects that overlap with the effects of Valproate which everyone accepts causes ASD.

So while there are confounding fashions, there is also enough solid evidence to point to a signal from something like a drug – a drug that in this case is the commonest drug taken by pregnant women.

Company Strategies

How do pharma companies manage/hide this?

• One strategy companies have used hide the problem is an emphasis on the importance of considering the totality of the epidemiological evidence.  Sounds good?

Except this means – just look at the Epi and not the totality of the evidence such as the epigenomic effects, toxic metabolite effects and asking awkward questions like should we be looking at neurodevelopmental delay rather than ASD.

Diagnoses of ASD and ADHD are intensely fashionable.  They are superficially very different conditions but in a proportion of cases there will be neurodevelopmental delay and it is this that Acetaminophen causes – it does not cause fashions.  Neurodevelopmental delay can be seen in animals – ASD and ADHD can’t.

• Second, insist that only an RCT can establish a link between a drug cause and ASD or whatever and therefore this has link has never been proven.

Companies claim epidemiological studies can only offer correlations whereas RCTs offer cause and effect. This is bananas.  Both RCTs and epidemiology are observational studies, offering correlations. The idea there is a magic ingredient to RCTs that saves them from confounding is gobbledegook.

The major confounder is our ignorance and RCTs do not get rid of this. Deliberately ignoring the evidence of a toxic metabolite, the epigenomic effects and the need to screen for neurodevelopmental delay rather than a fashionable label, is adding to rather than resolving our ignorance and RCTs would not prevent this.

In addition to this, there has been a general consensus that once a serious hazard, like neurodevelopmental delay, has been identified it would be unethical to run an RCT to explore this. This might appear to make it impossible to overcome the first point – the inadequacy of epidemiological correlations.

In practice though, companies working with regulators have  bulldozed through ethical concerns and are now running multiple trials of vaccines in pregnancy – See The Respiratory Syncytial Virus Challenge.

• Third – companies like Johnson and Johnson tell us that for a decade they have been keeping a close eye on possible links between acetaminophen ASD and claim there is insufficient evidence to show a causal link.

The first point to note is that major companies have been aware of this issue for over a decade but have not undertaken any original research to assess the causal association.

Linked into this is a company review strategy that fractures a reviewer’s abilities to consider the evidence in its totality and come to a judgement to a reasonable degree of medical certainty.

A decade ago GlaxoSmithKline were faced with similar issues in respect of paroxetine. In response to a growing number of epidemiological studies linking ASD to paroxetine taken in pregnancy, GSK set up teratology, pharmacology, epidemiology, pharmacovigilance, psychology and other groups, each tasked to review the evidence from their area of expertise. While the overall picture may be compelling, the chances of any one box containing slam dunk evidence is less so.

In addition, for “the sake of objectivity”, company work in an area like this is likely to be coordinated by a person with no background in the domain who, when asked whether the work-groups considered studies showing neurodevelopmental delay linked to paroxetine, could respond: “I have no expertise to decide whether we should do that or not, my brief was to get the groups to review any published articles that had the term ASD in them”.

What is happening here is that the assessment of a linkage between a medicine and a hazard is viewed by companies as a technical matter, a matter of process, that primarily involves collecting evidence rather than coming to what companies are increasingly likely to term a value judgement. Adhering to process in this manner creates an appearance of objectivity but creates a vacuum.

• Fourth – turn to regulators like FDA.  Beginning in 2014, with the publication of Liew, et. al, and then in 2015, 2016, 2019 and 2022, the mounting evidence linking acetaminophen to ASD led FDA to look at the growing evidence linking acetaminophen use in pregnancy and neurobehavioral outcomes in the published literature.

In 2019, FDA reviewers reported that “available epidemiological data suggest an association between prenatal acetaminophen exposure and urogenital defects. . .”. The review panel suggested the FDA “communicate this message to healthcare providers and pregnant women, considering acetaminophen is so commonly used by pregnant women, some of whom may perceive acetaminophen as risk-free. . .” The reviewers suggested a warning to consumers that use of acetaminophen to reduce fever was appropriate but heavy use for other reasons “may have risks.”

In 2022, FDA reviewers concluded “[i]n general, the functional neurobehavioral outcome studies examined in [their] review along with the review meta-analyses suggest a consistent association between APAP or long durations of prenatal APAP exposure and ADHD.” While they viewed “study limitations” as limiting “causal interpretations,” they suggested additional warnings about APAP use in pregnancy which should be “judicious” and focused on fever reduction.

(The point about study limitations is almost meaningless if the reviewing group were not asked to look at the evolving animal and epigenomic data.  Epidemiological studies on their own are always limited when it comes to causal interpretations).

There are no scientists in FDA – at least no-one better qualified that many readers of this post to work out what the Epi studies add up to.  Besides it is not FDA’s job to decide on cause an effect and warn.  Warnings are a company job – and a company legal requirement.

But companies have managed to give us the impression it would be much better – wouldn’t it if any independent body like FDA came out and told us what the truth is.

While FDA twist themselves into a pretzel over this, companies innocently say to the media – look FDA can’t work out what the truth is, what do you expect us to do? 

For Bettina

This series of posts was triggered by Bettina who listened to the recent podcast – Pregnancy:   Acetaminophen-Paracetamol – and asked for some of the evidence behind the points made.

After last week’s post she emailed:

interesting you gave us this talk and blog and just now July 2, there is a regional (German) mailing list of psychotherapists for children and adolescents reporting the contrary.  This study has been discussed in a medical journal coming to the conclusion that there is no effect of paracetamol in regard to ADHS or autism.  What do you make of this?

Ahlqvist VH, Sjöqvist H, Dalman C, et al. Acetaminophen Use During Pregnancy and Children’s Risk of Autism, ADHD, and Intellectual Disability. JAMA. 2024;331(14):1205-1214

The Reign of Doubt

Unlike SSRIs, in the case of paracetamol – acetaminophen, we do not have a significant ghost-written set of clinical trials to deal with.  But pharmaceutical companies increasingly turn to a set of companies no-one ever hears about, who commission researchers to do studies or write reviews of the epidemiology – articles that can look like they come from apparently independent academics pitching ‘the case is not made’ argument.

We desperately need accounts outlining how companies can seem to get bodies like the American Psychiatric Association or Royal College of Psychiatrists onside.  Getting consensus statements out of these bodies saying there is no good evidence any of the drugs our members use could ever cause a problem is easy-peasy for companies to manage. It often only takes one well-placed insider.

Is Ahlqvist et al one of these maneuvers?

Ahlqvist claims to have done a lot to eliminate typical confounders by using sibling controls and finding that when this is done any link between paracetamol and neurodevelopment problems disappears.

This study looks thorough but closer review suggests otherwise and brings some serious complications into view.

First the rate of paracetamol intake in pregnancy is only 7.6%. The rate of acetaminophen usage in the US is 60% of pregnant women.

The paper mentions paracetamol dispensed, which likely means prescribed. The authors semi-concede this but claim they have a structured interview process in place that systematically collects data on over-the-counter paracetamol also. They do not say whether this structured interview was in place 25 years ago, in 1995, the year their retrospective data collection starts.

There is then a marked widening of gap in the rates of ASD following acetaminophen exposure between affected and unaffected siblings from 2008, when there is a twofold difference, to 2019 when there is more than a 10-fold difference.  How might a gap like this have arisen?  They don’t say.

The paper also has significant missing data. The tables in the paper show the authors controlling for the effects of Aspirin and other non-steroidal anti-inflammatory drugs (NSAIDs), for anti-migraine drugs, opioids, antidepressants and anticonvulsants. But, while the data for drug groups not linked to ASD are reported at all points, the data for antidepressants and anticonvulsants, which are linked to ASD, go missing at key points.

FDA have very few reports of ASD or developmental delay for Aspirin, NSAIDs, Opioids (unless combined with acetaminophen) or anti-migraine medicines compared to the volume of reports on acetaminophen, which is more than the volume for the other drug groups combined.

Why do antidepressants and anticonvulsants, the drug groups along paracetamol linked to ASD go missing?  These two groups of drugs can be expected to have synergistic effects when combined with acetaminophen.

The result is so strange that Ahlqvist et al don’t spell it out – Aspirin, NSAIDs, opioids and anti-migraine medicines end up in this analysis looking more dangerous than acetaminophen. This strains credulity.

At other points, the authors lump all antidepressants together into one category, neglecting to take into account the fact that it is only the SSRIs that are linked to ASD.

They lump all anticonvulsants into one category, again neglecting the compelling evidence that Valproate is the anticonvulsant primarily linked to ASD.

The authors may be using a move from company playbooks, that let’s them claim there is no difference in the rate of side effects on active treatment compared to placebo in their studies.  This is achieved by adding up all instances of fatigue, for example, in each arm of the study, while neglecting related ratings of severity. The end result of this is that while the active treatment group has been badly affected it appears to be the case that the placebo group has, unbelievably, been equally badly affected, when this has not been the case.

The key point though is about confounding. Confounding is used by companies to claim there may not be a true link to paracetamol. Just the opposite, in fact, can be the case. Taking both acetaminophen and an SSRI or valproate together may greatly enhance the risk of ASD, in the same way as taking acetaminophen and alcohol enhances the risk of liver failure.

There is a RxISK post about this which created a new term that no-one seems to use – maybe Ahlqvist will help change that – The Reverse DODO Effect.

In Alice in Wonderland, unable to decide who has won a race, the Dodo says Everyone has won and all must get Prizes.  When someone drops dead because their heart fails and they are on several drugs that can cause this – Pharma says because we can’t decide who caused this, no-one has.

In case of liver failure, we do not say that, when we are not certain whether alcohol or acetaminophen has done the most damage, we should conclude that neither has caused harm. The same applies when there is concomitant exposure to both acetaminophen and SSRIs or valproate. Combination effects may be synergistic rather than just additive. This is an important point when it comes to warnings but only if we spell out what the hazard is.

Transgenerational?

A number of the parents in this study had prior intellectual disability diagnoses and possibly related psychiatric conditions. The timescale the study covers means that it is possible that some of their data may be transgenerational effects.

The ability of valproate to pass development delays from a parent taking the drug to a grandchild, presumably by virtue of its confirmed effects on chromatin, has been established in recent years (1).  Acetaminophen and SSRIs also have effects on chromatin that overlap with those of valproate. These epigenomic effects might contribute in the cohorts from recent years to both children exposed to acetaminophen in utero and unexposed children being apparently affected.

1. Martin M, Braillon A. Sperm epigenetic mechanisms in autism spectrum disorders. The valproate case illustrates an enduring and systemic failure. Molecular Psychiatry; https://doi.org/10.1038/s41380-023-02401-8. Martin M, Hill C, Bewley S, MacLennan A, Braillon A. Transgenerational adverse effects of valproate? A patient report from 90 affected families. Birth Defects Research, 2022; 114: 13–16.

Footnote

Finally, perhaps a minor note, perhaps not.  One of the authors on the Ahlqvist study has links to one of the well-known companies pharma turn to when they need expertise in the marketing of doubt.

Since the post went live, I have been sent the following materials

• Swann and Bauer Response to Ahlqvist
• Ahlqvist Response to S and B
• Frissel comment on Sibling Analyses
• Ahlqvist Response to Crawford