Highlights From The Comments On Aducanumab


These are highlights from the comments of Adumbrations Of Aducanumab, Details Of The Infant Fish Oil Story, and discussion of those posts elsewhere.

C_B writes:

I agree with this post's overall point that the FDA is not, on average, too lax, and that the Atlantic article's take that the aducanumab approval is a sign of them being too lax is a bad take.

That said, I think the beginning of this article really undersells how uniquely bad the aducanumab approval is. It's not just "pretty unclear whether it actually treats Alzheimers." Nobody in the field thinks there's any serious possibility that it treats Alzheimers.

- Here's Derek Lowe talking about it: https://blogs.sciencemag.org/pipeline/archives/2021/06/08/the-aducanumab-approval

- The FDA's advisory committee doesn't think it treats Alzheimers: https://alzheimersnewstoday.com/2020/11/11/fda-committee-votes-aducanumab-trial-data-fail-support-alzhimers-treatment-benefit/

- The trial was halted for futility: https://www.reuters.com/article/us-biogen-alzheimers/biogen-eisai-scrap-alzheimer-drug-trials-idUSKCN1R213G

- The details of the "positive results" are textbook p-hacking of exactly the sort that the whole replication crisis has been about. It's a post-hoc subgroup analysis where the subgroup was selected based on similarity to the patients who had the most positive results; i.e., trivially guaranteed to show "positive" results via group selection. You can read more details in the statistical reviewer's comments in the advisory committee's document (PDF, starting on p. 174): https://www.fda.gov/media/143502/download

- Anecdotally, as someone who used to work in the Alzheimer's research space, nobody in the field seriously thinks the amyloid hypothesis is true. It's been tried, over and over again, every drug that's ever aimed at that mechanism has failed, it's busted.

It's fair to use this Atlantic article as a jumping-off point about how the FDA's approvals system is bad and lots of things get held up there for bad reasons. But I think you're really not engaging with just how bad the aducanumab approval is, and why.

This is completely fair! I used noncommital language like “aducanumab might not work” because I hadn’t studied this specific drug in depth. I was getting my impression of a consensus that it was ineffective from the media. And since I’m part of the media, if I were to condemn it based on the media, this would snowball into a sort of false consensus effect. Readers could be left with the impression that 100 different trustworthy media sources think aducanumab doesn’t work, whereas in fact maybe one trustworthy media source came to that conclusion and 99 others thought “well that seems trustworthy” and repeated it.

But enough commenters whom I trust berated me for this that I edited some of the language to much more strongly suggest that aducanumab really really doesn’t work. Bio blogger and ACX commenter Metacelsus makes the full argument here, starting with:

Yesterday the FDA approved aducanumab, an anti-amyloid antibody developed by Biogen, for the treatment of Alzheimer’s disease. This was based on post hoc interpretation of clinical trials that were stopped early due to futility, and against the strong recommendation of the FDA’s own advisory committee. Aducanumab will be priced at $56,000 per patient per year. Since Alzheimer’s patients are usually covered by Medicare, this cost will be paid by the American public. I estimate that it could be over $100 billion per year.1 This is excluding indirect costs, such as those of monitoring for brain swelling that is a known effect of this drug.

Putting this in context, for $56,000/year you could hire a graduate student or postdoc to work on Alzheimer’s research full-time. Imagine the opportunity cost!

Writing at In the Pipeline, Derek Lowe called this a tragedy.3 I want to go further. This is not just a tragedy, but a travesty. The FDA has become a sadistically distorted mockery of what medical regulation should look like. There can be no excuses for the level of statistical incompetence required to approve aducanumab based on the flimsy efficacy data from the trials.

But tell us what you really think! (he does, at great length, in the rest of his post)

I hadn’t seen the number “$100 billion/year” before, and it’s pretty crazy. For context, there are about a million doctors in the US, and if each one makes on average $200K/year, that’s $200 billion. So Metacelsus is claiming aducanumab might cost the US health system half as much money as all doctors combined. This is based on his calculation that ~30% of 6 million Alzheimers patients get the drug for $56K each, which sounds like all reasonable assumptions, but I still bet somebody stops this long before it goes that far.

But Chebky is willing to go on record defending aducanumab (sort of):

I want to add another point: IMO, Aducanumab isn't shit, it's just that Alzheimer's is really really hard.

Personal background: We spent a couple of lectures in a molecular neuroscience class I took in the spring to review the clinical trials of aducanumab, including the latest trials (EMERGE and something else, iirc), so I dove quite deeply into them.

Aducanumab is not only excellent at doing its direct job - removing amyloid plaques - but also pretty good at reducing secondary biomarkers like amyloid PET signals and ARIAs (small hemorrhages and edemas that show up in an MRI). The problem is with the final proof of slowing cognitive decline, where most cognitive tests in the different trials failed to show statistically significant difference from placebo.

It is also worth noting that amab has an *excellent* safety profile with a wide therapeutic window and few side effects even given the age of the target population.

So what's going on? You can say that amab IS actually completely useless in slowing cognitive decline, but then you need to explain why the different biomarkers, all with well-established connections to neurodegeneration, fail to predict the (non)effect. The amyloid model is not the only one out there, but it's a central one for very good reasons.

Explaining away this connection is not impossible - you can say that by the time these biomarkers are fixed the neurodegeneration is already well underway and irreversible - but it adds a layer of improbability (and would mean that the drug may work as a prophylactic).

What is more likely in my opinion (specifically intuition from what I learned and going over the clinical data) is that Alzheimer's is pretty wide combination of different and very entangled neurodegenerative processes, with different contributions in different patients. Thus, the drug only works well on a subset of patients (maybe, like I mentioned, only those at the very onset of degeneration), but we don't know enough to predict which, so for the entire population of patients you fail to reach significance. Scott shows repeatedly how we see this in psychiatric drugs, which sometimes only barely squeeze by placebo in global trials but doctors have good experience in finding the one that works for a specific patient (and not knowing why).

This approval will mean (a) Biogen, previously on the cusp of failure, now have a lifeline to investigate which kinds of patients the drug is good for, (b) it is released into the market, which means we will soon be getting efficacy data in big numbers, combined with genetic data.

Other avenues are now also more easy to explore - maybe you need combinations with other treatments, maybe the cognitive tests we use aren't the best (it's not like we had a positive control of a perfect treatment to benchmark them against!).

Overall, the scientific advisory committee was absolutely right in saying the clinical evidence is kinda crappy. But the cost-benefit analysis is still heavily in favor of the drug, as it is very safe, molecularly effective, and with no alternatives, in a very widespread and destructive disease. The only cost being some people paying money for a drug that won't help them, and the benefits are huge both short term (for the people which it will help) and long term (for better understanding and getting closer to more global treatments).

In class we placed bets on what would happen. I said the committee will say no, and the FDA won't approve. I've rarely been so glad to be wrong.

Will counters:

Strong disagree. There are 10 anti-amyloid drugs that failed clinical trials in the past, some of which succeeded in removing amyloid. There should be a presumption that amyloid reduction does NOT cause a clinical benefit.

Your hypothesis that the drug will help a subsection of patients is plausible, but Biogen had 2 trials to find that group and failed to do so. They had every motivation to. To me, that signals the couldn't find a biologically plausible sub-group that had good results, that they were confident would replicate.

And "final proof of slowing cognitive decline" should be the only thing that matters. I don't care if a AD patient dies with a brain entirely free of amyloid. I (and you and taxpayers and doctors) care about cognitive performance. If Biogen, with billions on the line, couldn't devise a rigorous cognitive test that found evidence of clinical benefit, it isn't there.

The true tragedy of Aduhelm will be that the Alzheimer's Disease research space, which had begun to move away from the amyloid-centric hypothesis, is being effectively redirected to amyloid once again. Research $$ and healthcare $$ will be redirected towards a drug target that has done nothing but fail in clinical trials for 20+ years. The amount of wasted research and pharma money will be mindboggling.

Magic9Mushroom adds:

If this were the first time someone had tried amyloid-reduction as a therapy for amyloid diseases, I might be on your side. It's not. This has been tried a lot of times before with identical results - less amyloid, same Alzheimer's. Indeed, even outside Alzheimer's they tried this with Huntington's and got "less amyloid, worse Huntington's" for their trouble (the disease effects of Huntington's are from free huntingtin; the amyloid lumps are the brain's partially-successful attempt to sequester it).

The debate goes on another couple of iterations, click here for the whole subthread.

Hook writes:

The FDA is forbidden from considering price during the approval process. They don't exactly say this in https://www.fda.gov/about-fda/center-drug-evaluation-and-research-cder/frequently-asked-questions-about-cder#16 but they strongly hint at the fact that price doesn't enter into the evaluation.

That said, individuals at the FDA have their own opinions about drug pricing, and if they think something is just a copycat of another drug with a 100x mark up, that can have an influence on the approval process, largely through making more demands on the approval seeker.

Not sure if this is good or bad. It seems good that the FDA is only considering the science of a drug and not the economics/politics (eg “sell this for the half the price or we won’t approve you”). But it’s also awkward that they can’t consider price when making a decision which will effectively force the government to buy something.

nuclearspaceheater on Tumblr writes:

> [Texas Governor] Greg Abbott received a third booster dose of the vaccine, something not yet available to the public. Now he’s receiving monoclonal antibody treatment, something the FDA has only authorized for those with “high risk for progression to severe COVID-19.” Must be nice.

It occurs to me that thinking of this as a privileged depends in part on thinking of the FDA process as very excessively long, which many people here do.

But take the perspective of most people, who think of the FDA process as good and justified - and using politicians as test subjects for experimental drugs actually sounds cool as hell.

This is an interesting point. We sometimes see politicians “cutting in line” and managing to get advanced drugs that the FDA hasn’t yet made available to the general public. We usually interpret this as unfair privilege, and I think that’s a completely accurate interpretation. But that requires a sort of covert acknowledgment that these drugs are probably being delayed too long - otherwise, these politicians are heroes for taking potentially-dangerous substances and testing them out for the rest of us.

Peter writes:

I think the lawsuit happy legal environment in the US should get a lot of blame for this also. There are many useless and wasteful things that I have to do in medicine mostly because of the fear of getting sued. I believe many guidelines are developed with this as a background threat also and tend to trigger too many follow-up tests and exams.

That’s fair. I’ve seen some studies by economists saying that “defensive medicine” and fear of malpractice lawsuits doesn’t seem to raise costs too much in the US - but every doctor I know (including myself) feels like they do. Edward Scizorhands writes:

HMOs were extremely effective at keeping costs down by saying "no, that's not worth it." They gradually lost the ability to say "no" by lawsuit after lawsuit.

I don’t know enough history to know how true this is, but I’d like to learn more about it. I think “legislation via lawsuit” is a huge deal in health care and probably everywhere else, and I don’t write about mostly because I don’t know much law and also am in despair about whether it could ever possibly be solved.

Fallon writes:

Strongly agree with most of this article. Work at a biopharma and it's so annoying how expensive and slow the process for bringing anything to market it. What's most shocking is always how much the doctors and even internal employees internalize the FDA mentality that you'd let 10,000 people die from approval delays vs one person go to the hospital from an unintended side effect.

It's not true that insurance companies cover everything. There's some spaces like obesity where they cover next to nothing (I think semalgutide will be different) and very often they won't cover new medications when there is a cheaper older medication esp if it's generic. Lots of times people also need prior authorization for treatments and insurers can just say this isn't serious enough. I think you could see that happen here. The big issue is though that everyone 65+ qualifies for Medicare, most of those people buy Medicare part D (prescription plan) and Medicare does legally have to cover everything at any price because the pharma lobby owns half of the Democrats and every Republican. This means the government will end up paying biogen billions per year. Also since there aren't really any Alzheimer's drugs it's very hard for an insurer to refuse to cover the only treatment available.

The issue here though isn't the FDA. It does do an amazing job at getting rid of amyloid and it's safe. In an ideal world we should be okay saying doctors are allowed to prescribe medications that are safe and possibly effective. It's actually quite dystopian to say we should be arresting people or banning them from their professional for giving people treatments that are definitely safe and may be effective. Maybe then we should just stop giving out research grants because not every project is successful. The problem isn't with the FDA it's with doctors and moreso insurance companies being spineless and covering everything, which is really a problem with the legal system. No doctor wants to be hung out to dry for not prescribing a well publicized approved medications just because he had some skepticism and then the family goes after you for malpractice. And insurance companies can't afford to be evil boogeymen for refusing to pay $50,000 a year for a drug that may not work and even if it did work have very modest benefit.

The one part of this I want to ask about is the “aren’t really any [other] Alzheimers drugs”. There’s galantamine, rivastigmine, donepezil, and maybe some others I’m forgetting. These don’t work very well (or maybe at all). But aducanumab also doesn’t work very well or maybe at all, so why is everyone treating it as so unprecedented?

The NLRG writes:

“I think the current state of the art is something like Isakov, Lo, and Monterhozedjat , which finds that there are a tiny few disease categories where the FDA might be slightly too aggressive, but that overall the FDA is still much too conservative."

I might have missed something but I think this is a misinterpretation. They say that for severe diseases the FDA is too conservative, but for for more minor ones it's too lax. Since they only ran estimates for the 30 leading causes of premature mortality it shouldn't be surprising that, for most conditions they studied, the FDA is too conservative, because most of those are pretty severe diseases. But I'm not sure you can generalize that to say that there's only a tiny few disease categories where the FDA is too aggressive.

Thanks for this correction. I remember there being an older paper doing cost-benefit analysis on the FDA, does anyone remember what I’m thinking of?

Milt writes:

No discussion of FDA intransigence would be complete without a look at the life of John Nestor, https://www.theatlantic.com/health/archive/2012/05/medicines-missing-measure/257901/. I’m surprised he hasn’t been mentioned yet. He spent his spare time causing massive traffic jams on DC-area highways by insisting on driving 55 in the left lane. If an author tried to create an obstinate bureaucrat character based on his life, he would be panned for being too on-the-nose.

And Ian Argent finishes the story:

Please note: Nestor was transferred to a different role (effectively fired) for not approving ANYTHING in a 4 or 6 year period.

He sued and was restored "with apologies":

"At the FDA, he was a friend and supporter of his colleague Frances Kelsey as she exposed the side effects of the dangerous drug thalidomide. Informed by her experience, Nestor’s tenure at the FDA was marked by obstructionism. In a four-year stretch from 1968 to 1972, he approved zero new drugs for use by the public, viewing the risks and uncertainties still too great. As a result, he was moved to a do-nothing job within the administration. He protested, and was later reinstated and given an apology by the agency."

The willyreads.substack.com blog has its own take on this, Contra SSC On Aduhelm. He summarizes as:

Even though I’m arguing against Scott’s recent post, I want to preface this by saying I don’t really disagree with his broad conclusion all that much, which is that if you could only increase or decrease the strictness of the FDA, broadly defined, you should probably decrease the strictness. However, I think the story is complicated and if you want the best reforms, you need a better sense of where the FDA has gone right and wrong in the past. Also, I completely agree that regarding COVID-19, the FDA has been far too slow on approving testing, vaccines, etc.

Tl;DR: Aduhelm sucks, biomarkers often suck, approving/paying based on biomarkers creates bad incentives, confirmatory trials often don’t take place, skeptical tiered healthcare will remain tiered in the long-run

A lot of the argument focuses on surrogate endpoints, This is a kind of reasoning where you say “we want fewer heart attacks, we know cholesterol causes heart attacks, so we’ll design a drug that decreases cholesterol, do the easy work of making sure it actually decreases cholesterol, but not do the hard work of making sure it actually prevents heart attacks”. The advantage of this is that it’s much easier to check that a drug decreases cholesterol (which it might do in nearly everyone over a few months) than to check that it prevent heart attacks (which might take years or decades, since most people won’t get heart attacks or will only get them after a long time). The disadvantage is that the drug might just decrease cholesterol but not prevent heart attacks. In theory that shouldn’t happen, assuming that your “cholesterol causes heart attacks” theory is airtight. In practice, this kind of reasoning often fails. Either cholesterol doesn’t cause heart attacks (maybe it was merely correlated with them) or there’s some other disconnect (imagine that total cholesterol is the sum of hitherto unknown Type A and Type B which get lumped together in all our measurements, the drug decreases Type A, and Type B causes heart attacks). Doctors get warned against trusting surrogate endpoints, and the gold standard is always to study what you actually care about.

Still, like every other good idea, doctors are prone to following this one off a cliff. Suppose a study proves that mask mandates decrease coronavirus cases. Can we assume that they also prevent coronavirus hospitalizations and deaths? Seems pretty obvious that they should - but this is the dreaded surrogate endpoint, so purists will insist we continue the study for months longer to get a big enough sample of COVID hospitalizations/deaths to reach significance. Supposing that we prove that mask mandates prevent COVID deaths, can we be sure that the people who die will still be dead next year? Isn’t “dead now” technically just a surrogate endpoint for “dead over the long-term”, which is really what we care about? At some point you have to have faith in Bayes, don’t you? So in the general case I’m kind of split on this. Still, in the specific case of aducanumab we actually have specific, positive evidence that the surrogate endpoint doesn’t capture the real endpoint, so obviously this is bad.

Charlie writes:

It sounds like your qualm is with how American healthcare law is written, not with how the laws are administered by the FDA.

As an example, who are you more mad at: judges who upheld disparate sentencing guidelines created during the war on drugs, or the legislators who wrote the laws?

This is sort of fair, but I think in this example, you would summarize your position as “I am against disparate sentencing guidelines”, not “I am against legislators”. Saying “I am against nearly everything about the way the FDA works” is easily elided into “I am against the FDA”, even though maybe this is unvirtuous. I’ll try to think about if there’s a better way to frame this.

I’m half-flirting with trying a YIMBY analogy by dubbing my position YIMCS (“Yes In My Circulatory System”). It seems to be basically the same problem - governments that overtrained on a history of real abuses made it so easy to veto new developments that nothing ever gets done, and when things do get done they’re by big corporations with enough money and lobbyists to survive a harrowing approval process. But I think my personal credibility with the YIMBYs is pretty low, plus I probably shouldn’t choose an acronym that nobody will understand until I explain it to them.

On the infant fish oil post, Tyler G writes:

Huge fan, but I think Scott's being too easy on himself here. This was THE example that the rest of his initial FDA takedown hung on - the implication was "throw a dart at the FDA and here's what you'll find, it represents a thousand other stories just like it." It turned out to be substantially wrong. Like, AT LEAST as factually wrong than the crappy NYT story about Scott which he (correctly) didn't excuse.

Scott's charging for content now, and is doing important, high profile investigatory journalism. It's not OK to get so many relevant facts wrong. "This story is directionally and emotionally accurate, so don't worry that my supporting arguments were wrong" is exactly the kind of BS I want to see ACT/SSC as fighting against in the rest of media.

I would've liked to see a full retraction here, and most importantly, I'd like to see some of the subscriber money go to an intern who can help fact-check.

I acknowledge I got a lot of the story pretty wrong here, and I’m still trying to figure out what to do about this.

The immediate things I’ve done are

  • Write a paragraph on my Mistakes page admitting the things I uncontroversially got wrong, explaining that some people think I got more things wrong than that, linking to Kevin Drum’s post as an example, and explaining my thought process on the things I think weren’t wrong.

  • Italicize the paragraph in the original post and add underneath it that there are issues with the italicized paragraph, with links to the two subsequent posts explaining/debating the issues, and to the Mistakes page entry.

Things I need to do in the future are . . . man, I don’t know. I wrote the aducanumab post when I was really angry. I don’t want to never write posts when I’m angry, because those tend to be the posts people like the most, and I think they get things done in terms of convincing people of important things. But it does make me cut corners in terms of writing down things I half-remember without checking them. I think in the future, I try more carefully to follow the old “write drunk, edit sober” advice, with emphasis on the sober editing.

When you start writing, you always think “Unlike everyone else who writes stupid false things taken out of context, I will write good true things with lots of context”. But there are obviously reasons why this is so hard for everyone else. I think a big one is that when you get really mad about something, you become pretty desperate to prove that it’s true and worth being mad about, the story grows as you retell it to yourself, and you do the thing where you look for evidence that confirms you rather than disproves you. This is something I warn against in others, and I need to be better at avoiding it myself.

I feel least good about myself when trying to do the pundit thing (where I think I make mistakes comparable to those which other pundits make), and best about myself when I’m trying to figure out better institutions and theories for what rationality, updating, and bias are, and how everyone can do them better. I think one of those institutions is having a great comment section which catches my mistakes and keeps me honest, so thanks.

I’m not going to hire an intern right now because I hate communicating with other people, especially about my work, but if this starts happens more often, I’ll keep the option in mind.

Trevor Klee writes a blog post of his own: From A Small Biotech Startup’s Perspective, the FDA Is Terrifying.

I’ve been following the discussions around the FDA recently with great interest. COVID has made everyone a lot more aware of the FDA, which I think is why Scott Alexander’s recent post about the FDA prompted so much furious discussion.

I wanted to add a different voice to the mix: the voice of small biotech. When people think about who interacts with the FDA, they often think about large biotech, like Biogen and Pfizer. But, as with any industry, there are many more small players than there are large players. And, as with any industry, it is much more intimidating dealing with regulators as a small player than as a large player.

As the president of a very small, very early stage biotech company, this is a matter of personal interest to me. (Side note: if you’re interested in the mission of treating autoimmune diseases by repurposing generics, drop me a line!)

But, this is isn’t about me or my company. This is about a much larger (but still small), much later stage biotech: Axsome Therapeutics. They have a market cap that, until 3 days ago, was hovering around $2 billion, and a drug that, until 3 days ago, looked to be approved very soon for treating major depressive disorder.

What happened 3 days ago? The FDA happened. To put it bluntly, the FDA wrecked their shit.

You see, Axsome thought they were in a good place, as did the stock market. Axome had a drug, bupropion-dextromethorphan, which had done well in their phase 3 trials for major depressive disorder. The next step after that is to submit the drug for approval to the FDA, which, when there are such clear indications of improvement, is usually a layup. An annoying layup, filled with lots of paperwork, but a layup.

But that’s not what happened. Instead, the FDA sent Axsome a letter on July 30 saying that they found “deficiencies” that prevented them from moving forward with the approval. What are those deficiencies? Oh, the FDA doesn’t say. They don’t even give a hint. They just told them to hang tight until August 22.

It’s impossible to overstate how devastating a letter like that can be for a biotech. At this point, Axsome has undoubtedly sunk millions of dollars and around 9 years into this drug. Even worse, in preparation for a commercial launch, Axsome has undoubtedly hired dozens of sales people, marketers, and people who are experienced navigating the incredibly complicated landscape of insurance reimbursement. Every month that those people are on the payroll is probably at least another million burned.

The word “deficiencies” rings a bell, because I remember the FDA rejected an EpiPen competitor for “certain major deficiencies” without giving more information (at least not to the public). EpiPen took advantage of this to quadruple their prices, although the FDA did eventually approve the competitor a few years later. I was never able to figure out what the “deficiencies” were, which is too bad, because there seems to be something of an epidemic of “deficiencies” at the FDA these days.

Bupropion-dextromethorphan is an interesting case, because both bupropion and dextromethorphan alone are FDA-approved medications (bupropion as an antidepressant, DXM as a cough syrup). Then someone (maybe Axsome) did some preliminary studies suggesting that both of them together worked much better for depression than bupropion alone.

In theory, basically anyone can prescribe these two medications to a depressed person right now. We would have to prescribe them separately, but the patient could just take both at the same time. In practice, everyone (including me) wants to wait for more evidence that the combination is really as good as preliminary results say (I’m not worried about safety, since, again, these are both commonly-used drugs - in theory they could interact, but not enough that anyone would worry about giving them both to a depression patient who also had a cough). Nobody will gather that evidence without a financial incentive, and right now the financial incentive is getting the combination FDA-approved as a special combo medication.

It’s actually weirder than that, because probably Axsome will have to charge really high prices for their combo drug to recoup the cost of the FDA approval process. Doctors will still have the alternative option of prescribing (very cheap) bupropion and dextromethorphan separately and telling the patient to take one of each. And yet Axsome is full of smart businesspeople who have assured their CEO that nobody will do this, and I’m sure those businesspeople are right. Why make patients take two cheap pills instead of one convenient super-expensive pill, when Yagmuk is paying the bills either way?

I plan to make fun of doctors who prescribe this, and maybe to rib on Axsome a little. But realistically they’re doing good work by exploring this potentially promising drug combination.

Dausuul, a commenter on Kevin Drum’s blog, writes:

I have to say, I start giving a lot of side-eye to anybody who criticizes a scientific organization (which the FDA is) with talk about "common sense" and "the average guy on the street." Whoever said "common sense is merely the collection of prejudices acquired by age 18" was dead-on, and the average guy on the street knows f**k-all about science and medicine.

I think that’s the wrong way to look at it. The claim isn't that the man on the street is smarter than some individual scientist or administrator at the FDA. It's that the kind of normal knowledge that normal people use is better than the kind of institutional knowledge the FDA uses.

By analogy, consider criminal trials. Sometimes everybody knows some guy (eg Al Capone) is a crime boss. But the justice system can't arrest or punish him, because the common knowledge doesn't work the same way as institutional knowledge (eg it's hard to bring Capone to trial and prove beyond a reasonable doubt that he committed a specific crime). So even though we men-on-the-street all know he should be in prison, and even though justice system officials like police/prosecutors/detectives are smarter and more expert than we are, the justice system itself is unable to put him in prison, and its decision (keep him free) is stupider than our common-sense one (lock him up).

In this case, there's a really good reason we make the justice system very conservative. We want to have a very strong bias against locking up innocent people: "better ten guilty men go free than one innocent person get convicted." We also want to make it harder for the government/police to lock up people just because they don't like them. All of this is right and virtuous.

The problem is that the FDA uses similar logic. And contra the criminal justice system, where imprisoning an innocent person is a huge rights violation, the FDA is erring on the side of denying people rights, and making us unable to respond to pandemics and diseases. When everyone knows a certain treatment would save lots of babies / prevent COVID, and the FDA bans them from getting it, that imposes a lot of cost without a counterbalancing human-rights-based motive on the other side.

To return to my original point, just as detectives are smarter than you, but the justice system can still fail to implement plans that every common person knows are correct, so FDA doctors are smarter than you, but the FDA can still fail to implement plans that every common person knows are correct. I think it’s good that the justice system stays super-over-cautious. But the FDA could stand to loosen up.