Autism & Tylenol

Warning: I have not read all the papers on this subject. My experience is in these kinds of papers, which is to say, turning correlations into claims of causality. I offer no fixed opinion on whether or in what conditions, if any, acetaminophen causes autism. This is a preliminary post, necessarily incomplete, to help you, dear reader, sort through the evidence.

The claim isn’t even that acetaminophen causes autism, but something called “autism spectrum disorder” (ASD). That “spectrum” ought to have you guarding, and even girding, your mental loins (terrific mixed metaphor). For it can mean anything from “not autism” to actual autism, sometimes called “profound autism”.

Diagnosis rates of ASD, but not necessarily profound autism (though evidence is mixed), have indeed increased over the years. There was a large surge late in last century when (I think it was) 60 Minutes ran a story on it. Which, in the days of network supremacy, had many mothers rushing to the doctor, anxious to get in on it. This is evidence in favor of “awarenesses” being “raised” to account for at least some of the increase in diagnoses. Read that sentence carefully.

And, of course, diagnoses are imperfect, especially with “spectrum” disorders. Especially in a culture with mothers hopefully dragging junior to “doctors” to confirm their sons are “really” “daughters”. Certified credentialed employed quacks make these diagnoses. So why would you trust they would always get ASD right?

Problem is, how do you measure diagnosis rates to separate out increased doctor visits, which reveal both diagnoses of genuine illnesses and false positive diagnoses, with genuine increases? If you are looking at only correlations, which many papers do, you cannot.

I’ve pointed out the site Spurious Correlations innumerable times. Any two time series which are both increasing (or decreasing), for whatever separate causes, will show strong correlation, which can and will and are spuriously called causes because the standard of evidence in medicine is shite. (P-values and parameter-centric analysis. This is what the Class is all about.)

Kale sales and autism diagnoses correlate beautifully. Does kale cause autism? How can you prove it doesn’t if the method of proof you use to say something else, like Tylenol, causes autism is identical? You must take this question quite, quite seriously. If you pass but without answering, how can you hope to show your correlation is “different”?

Acetaminophen use and ASD diagnoses are both increasing. They are correlated. Rotten statistical measures which ought to be banned confirm the “link”. Are true autists, or their mothers, developing a hunger for Tylenol? Or is acetaminophen causing ASD?

Or is the correlation merely spurious?

As a gentle introduction, let’s take a quick look at the 2022 meta-analysis paper “A Systematic Review of the Link Between Autism Spectrum Disorder and Acetaminophen: A Mystery to Resolve” by Khan and others in Cureus.

You probably think, given that title, that they investigated call studies acetaminophen use and ASD. Alas, you’d be wrong. There not being that much information on that subject, they broadened ASD to include any research involving acetaminophen and “IQ”, “isolated language, attention and executive function, communication, behavior, and psychomotor development”, and our old friend ADHD.

Surely ADHD is over-diagnosed. Diagnosed falsely, I mean. Or may not itself be a real disease, whatever Experts might say. But let that pass.

In the end, they whittled the research down to 16 papers which met the criteria of acetaminophen and one of those listed diagnoses. Only 1 (one) of these papers specifically researched ASD, and that paper mixed that in with ADHD. (The remainging papers were mostly questionnaire based, like acetaminophen correlated to the “Strengths and Difficulties” “instrument”; scientists are always wrongly calling lists of quantified questions “instruments.”)

The one paper was “Association of Cord Plasma Biomarkers of In Utero Acetaminophen Exposure With Risk of Attention-Deficit/Hyperactivity Disorder and Autism Spectrum Disorder in Childhood” by Ji and many others in JAMA Psychiatry. This is as good as any in the genre, so let’s examine it closely.

From roughly 1998 to 2018 (it’s important), they gathered 996 mother-kid “dyads” (good grief; out of more than 3,000 initially enrolled) and checked cord blood for acetaminophen and its metabolites. That’s easy, relatively speaking. The idea is maybe moms eating acetaminophen are screwing with their babies, causing ASD.

On the ASD diagnoses: “Beginning at 6 months of age, enrolled infants who continued to receive pediatric primary or specialty care at the BMC were invited to participate in the postnatal follow-up study up to 21 years of age”. ASD and other diagnoses, like ADHD, were made “beginning in January 2004”. Which means we don’t know the timing of the diagnoses. We do see the average age at each diagnosis, including for kids who had no diagnoses (ND, but they say “neurotypical development” because, I assume, normal might be “hurtful”), but the means aren’t that different, about 8 years old for normal and ASD, and the standard deviation is large, about 3.5 years.

The correlations come at the end of their Table 1. For acetaminophen and two metabolites they report the number of ND and ASD kids at three levels: no detection, below the median level detected, and above. Which is not a keen idea, because that median will vary from study to study. They missed a prime opportunity to show us the raw distribution of measures for ND and ASD kids. Whenever we don’t see raw data like that, my suspicion levels rise, because I wonder if they are not showing us because the signal is weak.

Here are the numbers (DD is any other developmental disability):

There doesn’t seem to be much going on with metabolites, so let’s look only at acetaminophen itself. All of a sudden, we curiously lose the No Detection group, which is replaced by “First tercile”, which supposedly includes that, but it also includes some detection. The second tercile is a cut off less than the median, while the third tercile is one which is more than the median. So none of this is comparable to the metabolites.

There are only 66 kids diagnoses with ASD, and 29 (44%) of them are in the third tercile. But 89 (27%) of the ND kids also had third tercile detection. Here we really miss those distributions. Anyway, one comparison is the 44% versus 27%. A wee P (at far right) confirms the correlation, which means absolutely nothing at all. It would be the most blatant of fallacies—made almost daily, and here, too—to conclude there is therefore a causal relation between acetaminophen and ASD.

There is greater variability in smaller samples, and the ASD group is a lot smaller than the ND group. Especially when the cut offs are not fixed, as they aren’t here.

They do not show us the raw data, as I said. Instead data “were ranked using inverse normal transformation”. Ugh. That gives this picture (just for the cord burden):

Larger numbers mean more acetaminophen. And “density” is a sort of histogram: meaning the data are smoothed, which means we do not see data here, but a model. With this legend:

The dark black line (boy do we miss this raw data here) gives some evidence that acetaminophen was higher in the ASD group sometimes. Let’s suppose this is true.

Why?

Is it because the acetaminophen is causing whatever leads to a diagnosis of ASD? Or, and pay attention, is it because of something happening inside the mothers that drives them to take more acetaminophen than other women?

This could be a malady which needs pain relief or just the desire to be medicated. Not in all ladies, but only a handful. Only a few are needed to make that bump. And if it’s the latter, perhaps those mothers are the ones bringing their kids in more often hence we’d see more ASD diagnoses.

There is some evidence along that line. The ND kids’s mothers were largely not college educated, 67% weren’t. But the ASD moms were about split evenly between college and not. Or is this merely spurious? ASD kids also had a lot more Cesareans: 44% versus 30%. Again, we miss seeing the raw data. Maybe that small handful making the bump are all older moms, taking “measures” to juice pregnancies. Or maybe it’s genetic. We can’t know any of this.

Whatever it is, the signal is weak here. Consider that 27% of the ND (fine) kids had high acetaminophen. Which means we can deduce that (mother’s use of) acetaminophen cannot always be a cause of ASD. There were only 29 women in the third tercile ASD group. Out of 996. Of these 29, because of our knowledge of the third tercile ND group, maybe half of the ASD were caused by something else besides acetaminophen. After all, we just deduced (not estimated) acetaminophen is not always a cause. That means some in the ASD group might have had high acetaminophen, all right, but like in the ND group it didn’t cause ASD.

Stay with me here. This point is almost always missed. Another curse of the dreaded P-value. Because when it’s invoked to say “groups are different”, all minds are drawn to the false notion that thus everybody is, or most are, different in each group. This is not so: it does not follow for the reasons I just gave.

That means, maybe only 15 women out of 996 could have had acetaminophen cause their child’s diagnosis of ASD, or it means some number of these—like I said, a mere handful—just enjoyed being medicated, or had something else wrong with them that led them to take acetaminophen, and that whatever that “something else” was, or “something elses” were, is leading to diagnosis of ASD.

The point is, the verdict is far from clear. In this one paper. In the others? Each has to be read just as critically.

And now you know that if you say acetaminophen causes ASD diagnoses, based on evidence like this, you also have to say kale does

Subscribe or donate to support this site and its wholly independent host using credit card click here. Or use PayPal. Or use the paid subscription at Substack. Cash App: $WilliamMBriggs. For Zelle, use my email: matt@wmbriggs.com, and please include yours so I know who to thank. BUY ME A COFFEE.