For decades, the Food and Drug Administration has been making the wrong mistakes when approving medical treatments. Any FDA mistake is very bad, of course, often directly killing people. But some mistakes are inevitable—every so often, a clinical trial will give incorrect results from random chance or bad actors, or miss a good or bad effect because it takes 11 years to show up. The best they could do is to try to err on the side that causes fewer casualties. They’re not doing their best.

According to cost/benefit analyses, with the most respected published by Isakov, Lob, and Montazerhodjat in 2019, the FDA is much too conservative in their approval process for treatments of otherwise-terminal illness, and somewhat too lenient for treatments of other serious diseases. The FDA typically requires a study design that, in effect, targets a 2.5% false positive rate—an efficacy trial of a treatment that doesn’t work will appear to succeed anyway 2.5% of the time, due to bad luck (or, from the perspective of the patients in the treatment arm of the trial, good luck). Isakov et al. present an evidence-based model arguing that the optimal false positive rate for life-saving treatments is about ten times greater—when testing stuff that doesn’t work, we should be approving it about a quarter of the time. That’s for diseases, like some forms of pancreatic cancer, that are almost always fatal even if you catch them early. In these cases, the patient has little to lose and much to gain from an experimental treatment. For what they term “relatively less deadly conditions,” the optimal threshold is a bit smaller—we should only erroneously approve a treatment about half as often, or 1.25% of the time. Here, there’s a more significant opportunity cost—people taking bad treatments when good ones are available—and more concern about side effects, since the patient is more likely to live long enough to experience them. An analysis published by Samuel DeCanio in 2024 found that the approval of Vioxx to treat arthritis symptoms, in ignoring data suggesting that it also caused heart attacks, cost us more than we benefited from all the other drugs approved under the same process combined.1

This is not simple to fix. If the FDA lowered their standards, so would drug companies. Individuals in the FDA have incentives to be more conservative than official policy dictates—they get in more trouble for a bad approval than a bad rejection. Pharma giants like that it’s so hard to get drugs approved because it locks smaller companies out of the market, so nobody with deep pockets is lobbying to fix the FDA.

I think it is fixable, but I’m not going to propose a solution here. There’s a more pressing issue: the current FDA leadership is steering us in the opposite direction. The FDA does have a patchwork collection of quasi-ad-hoc procedures for accelerated approval, meant to be used in the sort of cases Isakov et al. are highlighting. Those procedures are now being paused or dismantled. Under the Trump administration, we’re on track for fewer approvals this year than last year. This is partly because the FDA leadership has been changing a lot. Companies that conducted clinical trials using a design the FDA signed off on are now being rejected because the new heads have different standards. That’ll calm down eventually, probably. But we’re still going to be stuck with stricter standards.

The thought leader, and intermittently2 formal leader, of the move towards rejecting more medicine, is Vinay Prasad, current director of the Center for Biologics Evaluation and Research and author of the book Ending Medical Reversal. “Medical reversal” is a term he coined—a reversal is any time the medical community starts recommending a treatment and then stops recommending it in response to new evidence. Prasad doesn’t, so far as I’ve seen, present his own cost/benefit model or critique the existing ones. Rather, he makes a qualitative argument— “medical reversals erode trust,” alongside some statistics that don’t particularly demonstrate anything.

Coining the phrase “medical reversal” was a clever move. By its definition, it excludes the other kind of error, where the medical community initially rejects a new treatment, but eventually starts recommending it. This asymmetry means it’s only an appealing phrase to his faction, so they’re the only ones that use it. The other side uses the generic and objectively terrible terminology “Type 1 vs Type 2 errors,” where in this case a medical reversal is typically Type 1 and rejecting an effective treatment is Type 2.

So if you google “medical reversal,” you will only get results where the authors implicitly assume that we have too many of them, and that they’re the result of certain trial methodologies Prasad dislikes. But now, I hope, you’ll also get this article.

I’ve discussed before how when you name one side of a binary, but not the other, the result is often an exaggerated fear of the named side. (Consider, for example, a hypothetical language that has a word that means “murder,” but also the word “sinicide,” which means “murder committed by a left-handed person.” Or a media diet where you hear about “crimes” and also “crimes committed by illegal immigrants.”) Naming “medical reversals” allows Prasad and his allies to smuggle in assumptions without ever having to make the case for them.

As Alex Tabarrok wrote in 2015, “when the FDA fails to approve a good drug, people die but the bodies are buried in an invisible graveyard.” When another country approves a treatment years before the U.S. does, we sometimes call it “drug lag.” But when R&D stops on a treatment we’ve mistakenly dubbed ineffective, we don’t call it anything at all.

So, clearly, we need a name. I propose “missing medicine.” Some usage examples:

In 1990, the FDA failed to approve the anti-cancer drug Interleukin 2, which was already approved in nine European countries. For the next three years, Interleukin 2 was missing medicine in the U.S., causing an estimated 3,500 deaths from kidney cancer.
David J. Stewart, oncologist and author of Why Cancer Still Sucks, estimated in 2018 that missing medicine has cost U.S. cancer patients a cumulative 19 million years of life, by looking at cancer drugs that took more than 5 years to approve and their effects on life expectancy.
Due to recent vaguely-defined changes in FDA standards that invalidated several clinical trials, we’re missing promising treatments for, among other things, two incurable, deadly genetic conditions: Huntington’s disease and spinocerebellar ataxia. These treatments slow, but do not reverse, the progress of these diseases, meaning every day of delay is doing irreparable harm to tens of thousands of people.3

Note that a list of bullet points is not an argument, just rhetoric. You can find plenty of lists like this of “medical reversals.” They don’t show anything, in isolation, about the relative importance of medical reversals and missing medicine. My main hope in writing this article is just that anybody looking up medical reversals is alerted that a controversy exists.

But yes, I am more or less On A Side here. Academics who talk about both medical reversals and missing medicine, who try to quantify their respective impacts, seem to fall overwhelmingly in the “missing medicine is a bigger problem” camp. The people on the other side are just quietly assuming away the question. That doesn’t necessarily mean the former side is right. But it’s a strong preliminary result.

The Serpent Steals the Herb of Immortality from Gilgamesh (illustration by Ludmila Zeman) — Illustration from The Last Quest of Gilgamesh, artist Ludmila Zaman. In the ancient Epic of Gilgamesh, our hero finds the plant granting immortality, but takes too long setting up a test of its efficacy. A snake eats it while he’s sleeping, shedding its skin and becoming a younger, healthier-looking snake. This isn’t valid evidence by modern standards—the experiment was unintentional and lacked a control snake.

1

I’m not sure that particular analysis is going to be borne out—there’s some evidence that Vioxx wasn’t any worse, heart-wise, than less selective NSAIDs like ibuprofen, and therefore wasn’t causing excess deaths when prescribed in their place. But the general idea that a quality-of-life treatment needs stricter standards seems solid.

2

Laura Loomer got him fired for a few weeks by faking evidence that he was a left-wing saboteur.

3

The FDA notified uniQure on November 3rd that they no longer considered their trial valid basis for approval. They promised to explain why within the next 30 days (during which over a hundred people would die of Huntington’s, but hey, they’re busy). As of this writing, it’s been 29 days with no public announcement from either the FDA or uniQure. My best guess is that they objected to the way patient outcomes were being compared—instead of creating a new control group, the researchers compared their patients to similar ones in the Enroll-HD program, which monitors volunteers who aren’t receiving any novel treatment. Prasad considers that design unacceptable in most cases because it’s too easy to end up with a flawed control group, e.g. people not receiving the current standard of care.

We Should Have More Medical Reversals