We have published four peer-reviewed papers since December of 2022 evaluating the role of acetaminophen in the development of autism spectrum disorder. Based on that peer-reviewed, published work, twenty-four lines of evidence lead to the conclusion, with no reasonable doubt, that exposure of susceptible babies and children to acetaminophen causes many if not most cases of autism spectrum disorder. Furthermore, the best explanation for all available observations is that exposure of those susceptible children to acetaminophen is responsible for the vast majority of all cases of autism spectrum disorder.
Systematic Review Blind Spots
Yesterday I reviewed a fascinating article (https://pmc.ncbi.nlm.nih.gov/articles/PMC9385573/) from July of 2022 in the journal Cureus from the California Institute of Behavioral Neurosciences & Psychology. That article tells us what some prominent scientists thought about the connection between acetaminophen and autism at about the time we published our paper in December of 2022.
The authors’ concluded, as many other have, that nobody knows if acetaminophen can cause autism. How can one group of scientists come to one conclusion, and another group of scientists come to the opposite conclusion? Don’t we all have access to the same information, and don’t all scientists use the same scientific reasoning to assess scientific information?
The answer is simple, and it can be found in the methods of the Cureus article. The scientists who wrote the Cureus article used a “systematic” approach, which means that they followed very specific rules when they decided what sort of evidence they wanted to consider. They excluded any study older than five years, eliminating several lines of important evidence, including evidence from 2008 showing that acetaminophen given with a vaccine, not the vaccine itself, is associated with autism. Their method excluded some of the most convincing stand-alone evidence ever published, including the finding of more than double the prevalence of autism in circumcised boys compared to uncircumcised boys (https://pmc.ncbi.nlm.nih.gov/articles/PMC4530408/). (Acetaminophen is frequently used with circumcision.) That circumcision study was excluded by two rules they used: it’s older than five years, and it did not assess acetaminophen use. They also excluded the body of evidence showing that acetaminophen impairs social awareness in adults. (Alterations in social awareness are a hallmark of autism.) Their methods inadvertently excluded numerous lines of evidence that require nuanced scientific rationale, including (a) comparisons between autism and fetal alcohol spectrum disorder, (b) otherwise unexplained findings of autism prevalence that are explained by acetaminophen exposure levels, (c) temporal relationships between autism discovery/prevalence and acetaminophen use, and (d) findings by independent laboratories that children with autism are deficient in enzymes needed to safely process (metabolize) acetaminophen.
Out of the 24 lines of evidence we consider (https://pubmed.ncbi.nlm.nih.gov/39202661/), the team at the California Institute of Behavioral Neurosciences & Psychology focused almost exclusively on two of those lines of evidence dealing with studies of healthcare databases. They also made mention of two other lines of evidence (the mechanism is plausible, and strong associations exist between autism and cord blood acetaminophen), for a total of four lines of evidence. We agree with their conclusion, with some modification: IF we only consider one out of every six lines of available evidence, THEN we don’t know if acetaminophen use in susceptible babies and children causes autism.
Why Use a Systematic Review at All?
The team at the California Institute of Behavioral Neurosciences & Psychology actually did a really nice job of collecting information within the constraints of a systematic review. Some groups using the systematic approach consider only one line of evidence, for example analysis of healthcare databases describing acetaminophen use during pregnancy only. Those reviews miss more than 95% of the evidence we consider in our reviews.
The next question is also interesting: why would anybody use a systematic review if it excludes so much useful evidence? My own group has used a systematic review previously. It’s a great way to make sure that you capture EVERYTHING about a HIGHLY SPECIFIC issue. We wanted to know if acetaminophen had ever been proven safe, so we systematically evaluated the thousands of papers that said acetaminophen was safe for babies and children when used as directed. We worked with a bioinformatics professor to employ a technique called “citation tracking”, spending months looking at why people thought acetaminophen was safe.
It worked. We were able to firmly conclude that acetaminophen was thought to be safe ONLY because many people assumed that babies are small adults and that they would react to acetaminophen the same as an adult (https://pmc.ncbi.nlm.nih.gov/articles/PMC9056471/). Nobody checked for brain development before concluding that it was safe, despite the fact that the drug has numerous and complex biochemical effects on the human brain. They only checked for liver function, because, in adults, acetaminophen can be dangerous to the liver.
So, in a nutshell, if the scientific question is suitable for a systematic review, then a systematic review is a great way to make sure that nothing is missed. However, we still need to know how to ask the question, and if the question is more complex, then a systematic review can easily miss lots of information, which could be fatal to the scientific inquiry. In the study from the California Institute of Behavioral Neurosciences & Psychology, their systematic approach caused them to miss more than 80% of the relevant information we have assembled. Again, that’s actually pretty good for a systematic approach. They could have missed much more.
A Second Problem with the Systematic Approach
Another problem with a systematic review is that you still need to be a scientist with the right background to understand the results. We could have concluded that acetaminophen is safe because more than 2000 papers in the medical literature said it is safe, and we found very few people that raised any questions. That conclusion would have been wrong. The correct conclusion from that study was that (a) most people writing articles in the medical literature think acetaminophen is safe when used as directed, and (b) the reason they think it is safe is due to propagation of an error that is well known in the field of pediatric pharmacokinetics.
A great example of a systematic review that drew the wrong conclusion was a recent paper published in the Journal of the American Medical Association (https://pubmed.ncbi.nlm.nih.gov/38592388/). The authors found that acetaminophen use during pregnancy is very strongly associated with autism. The “risk” was about 1.8 under some circumstances, which means that acetaminophen use during pregnancy was associated with 80% more (almost double the amount of) autism. However, in the end, they concluded that acetaminophen was NOT associated with autism, in part because they assumed that autism is caused either by acetaminophen or by the reason for taking acetaminophen. We know this assumption to be wrong. Studies in toxicology and pharmacokinetics tell us that autism is caused by a combination of acetaminophen plus a variety of factors that could lead people to take acetaminophen (https://pubmed.ncbi.nlm.nih.gov/28415925/). The data published in the JAMA paper is consistent with this conclusion. The mistake in the conclusions of the JAMA paper is discussed in detail in our recent assessment of the issue (https://pubmed.ncbi.nlm.nih.gov/39202661/).
Be Systematic. Sometimes.
Does the non-systematic approach we use have any weakness? The main weakness is that we might miss important information that tells us we are wrong. The possibility that our conclusions are wrong grows smaller and smaller as more evidence emerges. At this point, we conclude that the evidence is so strong that it’s literally not reasonable to consider the possibility that our conclusions are wrong. In fact, several lines of our current evidence come from people who tried to prove us wrong, but ended up providing even more evidence to support our conclusion. That’s how science normally works: we try to prove that our current thinking is wrong, and if our thinking holds up, then the thinking is supported. In addition, having a broad, multidisciplinary team of scientists is extremely helpful, and peer review is still a critical double-check. That’s why we work with scientists from a variety of disciplines and publish our work in the peer-reviewed scientific literature.
So, in conclusion… should we use a systematic review? Absolutely not when we are trying to put together the answer to a wildly complex question such as, “Do a variety of genetic, epigenetic, and environmental factors confer susceptibility to acetaminophen-mediated neurodevelopmental injury, leading to a pandemic of autism spectrum disorders in the face of widespread acceptance and use of the drug?” On the other hand, the answer is absolutely yes if we want to ask a simple question such as "Why do people writing articles in the medical literature think that acetaminophen is safe when used as directed?”. Still, we should be careful with interpreting results from systematic reviews. Sometimes a simple question is not as simple as we would hope.