My Take on Tylenol, Science, and Assigning Value

It’s no mystery that healthcare has a credibility problem. Even before the pandemic, the public’s trust in medicine was being eroded by the relentless commercialization of pharmaceutical innovation, which seemed to infiltrate every corner of American life. Then came the pandemic — and with it, repeated missteps by the CDC and FDA, coupled with the wildfire spread of misinformation supercharged by unchecked social media algorithms.

So, I wasn’t surprised when a wave of outsiders began taking leadership roles at major health institutions. The credibility gap practically invited them to take a swing at shaping health policy. Some of those new voices I admired. Dr. Marty Makary, for example, has done tremendous work exposing malfeasance in the healthcare sector and advocating for commonsense reforms like price transparency. After all, when hospitals charge $90 for a liter of saline, reform seems overdue.

At the top of today’s healthcare debate is RFK — an anti-establishment figure through and through. For over a decade, he has waged a relentless campaign against what he sees as the key drivers of America’s declining health. His targets have ranged from thimerosal and food dyes to the hepatitis B vaccine and the entire childhood immunization schedule. While his intentions may be sincere, his message is often misguided, and thanks to his charisma and platform, the narrative spreads to many that are begging for a fresh perspective.

Even when you live it, science is difficult to wrap your head around. It’s even harder if you aren’t responsible to apply it in real world situations (no skin in the game.) The nature of a bureaucrat is the absolution of accountability.

Dr. Malcom in his prime. (google images)

One of my favorite fictional characters, Dr. Ian Malcolm from Jurassic Park, captures this challenge perfectly. Malcolm isn’t a scientist, per se — he’s a chaos theorist; he deeply understands the implications of science. Looking far out into the horizon he foresees the chaos from introducing discovery into a stable environment. He has this smug, yet brilliant way of explaining how scientific discovery can be dangerous if mishandled. In one iconic scene, he warns about the power of genetic cloning, likening scientists to “children playing with dynamite.”

Discovery is like dynamite: it requires a great deal of respect.

Over the weekend the Trump administration made an announcement that they had made a big discovery on Autism. They set up a press conference for last Monday. Everyone was there. The big bad villain was Tylenol.

A stick of dynamite…

Tylenol is the only safe medication pregnant women can use to help treat pain and fever during pregnancy. I guess they want us to go back to the days of aspirin and Reyes syndrome, or NSAIDS and renal artery atresia.  Or as Trump would put it, “just tough it out!”  As if fever itself is not dangerous.

Immediately after the announcement a back and forth ensued on the validity of the data investigating this question.  Mind you, no new data was offered. Simply historical observational data and their personal interpretation of it.

Back to theory, and dynamite, and discovery…

Robert Persig’s Zen and the Art of Motorcycle Maintenance is a brilliant book. The type of book that ruins your life. It takes your reliance on facts and data, and tips it upside down.  In the book, the protagonist is battling a former version of himself, An alter ego.  He is trying to answer the question “What is Quality?” You would think this would be a silly topic, but it’s a great intellectual battle. The protagonist never really answers this question. What makes him the protagonist is that he gives up, before going insane, for a second time. He resolves to not knowing. He is incapable of knowing because he does not possess a perspective other than his own.

In the book, the subject of science and “facts” is discussed at length.  One fundamental truth about science is that it is inherently valueless. Let that sink in.

Science is numbers on a page. A computation at the end of a statistical equation applied to a dataset. It’s an observation of reality; nothing more. It’s inanimate. As inanimate as the chair you are sitting on. We assign value to it.  It’s our job to look at the number, where the number came from, and decide if it’s “true” or “meaningful.” It’s also our job to look at the data and figure out if it represents our environment. The latter is often more difficult than the former.

It is difficult to assign value to controlled experients.  Observational (uncontrolled) science is worse.  The data regarding acetaminophen (Tylenol) exposure is observation. Meaning, two groups of people were observed, one group had an exposure to acetaminophen, the other didn’t, and the outcome of autism was measured.  However, no other variables could be confidently controlled.

I’m going to give a golf analogy to explain the difficulty of interpreting scientific studies using observational data. Why? Because I like golf.

Typically, I hit a 7 iron about 170 yards. 

But there are many conditions that might affect the ball flight and path that could change that distance.

Let’s say, my ball is sitting on a dry patch of grass, I’m down wind, and it hasn’t rained in over a week.  

I’ve got a 180-yard shot. Since I’m downwind, I decide to hit my 7-iron.

I strike the ball well, it flies off the hard dirt with no spin, lands on a domed part of the green, doesn’t hold, skips off the back, and down a hill 30 yards past the hole.

My playing partner, “Wow! You just hit that thing 220 yards?”

Understanding the conditions in which that took place, I know I am unlikely to repeat that on the next hole if faced with a 180-yard shot again.

In science we make the mistake, sometimes, of believing what we see, and it fools us. But we don’t know we’re fooled.  So, we try to hit our 7 iron 220 yards on the next hole.  Then, we purchase 100 copies of the same 7 iron, walk around the course handing out 7-irons to the other players with the promise of hitting it 220 yards.

The other players start swinging and it goes all over the place. Some hit it 100 yards, some hit it 150 yards, some hit 220 yards, but the original shot is difficult to repeat.

Eventually after years of swinging for 220 yards some genius says, “Wait a minute, I can only hit my 7 iron 150 yards, and this is how it’s always been.” And we are back to square one.

This is the danger.

If you don’t believe me, here’s a real-world example.

The other week at journal club a resident presented an article investigating hypothermia protocol vs. normothermia in out of hospital cardiac arrest.  The randomized trial showed no difference in outcome.  The authors recommended maintaining normothermia as opposed to purposefully cooling patients to improve neurologic outcomes after cardiac arrest. The trial had 1700 patients in it.

For historical context, in 2002 a randomized control trial investigating hypothermia vs. a control group showed benefit in neurologic outcomes in hypothermia group. The trial had 275 patients in it.  This set off a change in every modern healthcare system to cool patients after cardiac arrest. It was an immense effort, cost, and burden on all parties involved.  Twenty-three years later we find out it was a waste of time. Just prevent them from getting a fever and you’re fine.

Here's a real-world example on assigning value: I was in a discussion with residents regarding the use of pro-calcitonin in patients with sepsis to determine antibiotic use. Some critical care physicians check this level daily ($180ish). Others never check it (free). We searched the study, and it showed anti-biotic adjustment based on pro-calcitonin levels reduced anti-biotic duration by a whopping 8 hours ($50ish). One physician assigned great value to that (and cost). Another assigned no value to it. I suggested the truth is probably somewhere in the middle. 

Examples like this are everywhere in healthcare. Healthcare works; But it's imperfect.

Let’s talk about the issue that started this blog post acetaminophen and autism. The most recent large study involved 2.5 million children in Sweden.  Of those roughly 200,000 were known to be exposed to acetaminophen in utero. The incidence of autism in the exposure group was roughly 3% (comparable to U.S. autism rates). The rate in the non-exposed group was 2.7%. This is an absolute risk of 0.03% leading to a number need to harm of 333.  So, 333 women need to consistently take Tylenol during pregnancy for 1 case of autism. 

The study showed some strange correlations.  The siblings of those exposed to acetaminophen had higher autism rates regardless of acetaminophen exposure than siblings unexposed.  Furthermore, acetaminophen use was consistent throughout the duration of the observation period (20 years), but autism rates slowly declined in the acetaminophen group.  Also, there was a dose dependent relationship between risk and acetaminophen exposure.

The lack of control prevents us from forming a strong relationship. It should also be noted that the 20 years of observation will never be repeated in the history of mankind, since each period is unique and independent of the prior observation period. Like changing holes on a golf course, my score on the current hole is independent of my score on the prior hole.  This is an important concept to understand. You can never relive those conditions in which the study took place again.

There are many plausible explanations that could explain the correlation. Maybe those women that needed to take Tylenol got sick during their pregnancy, and it was the infection, not the medicine that caused the disorder. Maybe it was the fever; Maybe some of them had other medical issues that required acetaminophen to relieve symptoms; Maybe it was genetics; Maybe they were a little bit older; Maybe they were married to older men (side note: when my wife and I were trying for a baby in our mid 30’s I was talking with a friend of mine about the experience and he described my swimmers as being “old and slow.”) Maybe more women are working later into their pregnancy and taking more risk with their health due to concern over their career? The point is, who knows, it could be anything.

I think what the American people are witnessing in real time how hard science can be. Not the actual execution, but the assignment of value. For our leaders to push all their chips in on acetaminophen causing autisim, they took a big risk with their credibility, they assigned big value, and likely missed big.  I’ll take the field against Tylenol in that bet. It reminds me of the COVID days.

Despite their disdain for established healthcare bureaucrats, the new administration is making the same mistakes.  It’s ego. It’s a public promise. It’s a vendetta against the establishment. The tool is data, but the data requires value assignment, which no one person can do, especially if the data source is chaotic. You’re left holding sticks of dynamite, and if you’re impatient, you set some off.

Bad ideas are viewed with skepticism, and good ideas are rare, often decades apart.

If you’re wondering? Yes, you can take Tylenol. Even if you’re pregnant. You don’t have to Mychart me to ask permission. If you’re wondering what the math says; In the current environment there is a 1:33 chance your will give birth to a child with autism, if you take Tylenol during your pregnancy, that may nudge it to 1:32.5 chance of autism, but it may have not been the Tylenol responsible for that nudge.