In 1989, the drugs were working exactly as designed, and that turned out to be the problem. The logic was clean enough to teach a medical student: patients who survive a heart attack often develop extra, irregular heartbeats, those beats predict sudden cardiac death, so suppress the beats and you should prevent the deaths. Two drugs, encainide and flecainide, suppressed them beautifully. Then the Cardiac Arrhythmia Suppression Trial was halted ahead of schedule, because the patients taking the drugs were dying at roughly 2.5 times the rate of the patients on placebo. The therapy hit its target and killed people anyway.
Cardiology never forgot the lesson, and it has a name: a treatment can pass the exact test it was built to pass, the surrogate endpoint, and still leave you worse off than doing nothing. Hold onto that, because it is the hinge of an essay Vinay Prasad published on September 28. Prasad is the oncologist and epidemiologist who ran the FDA’s Center for Biologics Evaluation and Research until he stepped down this spring and returned to his faculty post at UC San Francisco. His argument is not that a specific drug is dangerous. It is that American public health is structurally incapable of seeing the next CAST coming, because it is looking somewhere else entirely.
The indictment is that the apparatus does not set its own agenda. It “only talks about issues in response to what the Rs are doing,” he writes, calling it “nakedly liberal and reactionary,” an institution that “craves the daily political news cycle” and “can’t seem to formulate ideas on its own.” Measles outbreaks, a contaminated-lettuce recall, a Hantavirus case: real events, worth a press release, and above all safe to talk about because a political opponent has already made them a fight. The trillion-dollar questions go unasked, in his telling, precisely because no one is scoring points on them.
Consider the test now moving toward Medicare coverage. Multi-cancer early detection tests, the best known being Grail’s Galleri, promise to find dozens of cancers from a single tube of blood. It sounds like pure upside, and that is exactly the framing Prasad distrusts. His question is whether the people screened will “experience net benefit or net harm” once Grail is “approved and later covered by Medicare.” The real-world data give the question teeth. In the PATHFINDER study the test’s specificity looked reassuring at 99.1 percent, and yet only about 38 percent of the people who got a positive signal actually had cancer. For the rest, a false alarm, the workup to rule out a tumor that was not there dragged on for a median of 162 days, against 57 days when the signal was real. Months of scans and specialist visits, chasing a cancer that had never been present.
| Measure | Share |
|---|---|
| of Galleri positives | 38 percent |
| Group | Value (days) |
|---|---|
| True positives | 57 |
| False positives | 162 |
And “finding it early” is itself a surrogate. A test can shift more cancers into the early-stage column by catching indolent tumors that would never have killed anyone, generating treatment, cost, and fear without adding a day of life. The largest trial of the idea lands right on that worry. In NHS-Galleri, screening more than 140,000 adults did not meet its primary endpoint: the incidence rate ratio for stage III and IV cancers together was 1.03, with a confidence interval running from 0.92 to 1.14, sitting squarely across no effect. A prespecified secondary signal, a 14 percent drop in stage IV diagnoses, was offset by a rise in stage III cancers the test pulled forward. That is a stage-shift the screen can document and not, on these numbers, a demonstrated gain or loss in lives. Coverage is coming anyway, on a slower and narrower track than the hype suggests. A 2026 law, Public Law 119-75, built a Medicare pathway for these tests, but it does not open until January 1, 2029, and even then it pays for nothing until a test first clears the FDA – a 510(k) clearance, De Novo classification, or approval – and then earns a separate national coverage determination from CMS. Its payment limits also draw a narrow band: Medicare will not pay when the test is furnished to someone under 50, or to someone who has already reached 65 as of January 1, 2029, with that upper cutoff climbing a year in each later calendar year, so a beneficiary who only turns 65 during 2029 is not swept out – and those age limits lift entirely for a test the U.S. Preventive Services Task Force grades A or B and that Medicare then covers as a preventive service. Conditional, phased, and years out. The door is being built before anyone has answered whether walking through it helps.
| Result | Value (IRR) | Lower bound | Upper bound |
|---|---|---|---|
| Combined stage III and IV | 1.03 | 0.92 | 1.14 |
The weight-loss gold rush has the same shape. Prasad asks which of the new GLP-1 drugs will “save the most lives,” which are “dangerous,” and whether the raised heart rate seen with retatrutide, Eli Lilly’s triple-hormone agonist, makes it worse than tirzepatide. The heart-rate signal is documented: in its phase 2 trial retatrutide raised resting heart rate in a dose-dependent way, a rise that climbed through the first six months before easing. Lilly is now running the two drugs head to head in a trial called TRIUMPH-5, but it is built to compare weight loss in people with obesity, not to score which drug lets more patients live longer. “Running a head to head trial of these agents powered for hard endpoints would be a boon,” Prasad writes, and the phase 2 data he is working from cannot answer that question in its place.
Then Prasad names the fear directly. “What if: a widely used prescription medicine that is supported by disease specific regulatory endpoints increases mortality. This could be catastrophic.” Read it again with 1989 in mind. That is not a hypothetical. That is CAST, written in the future tense, describing a disaster the system already produced once when it trusted a surrogate and only counted the bodies afterward. The machinery that would have to catch such a thing in advance, the horizon-scanning arm of public health, is the very thing Prasad says does not functionally exist.
Here the establishment’s instinct is telling. A recent paper in Frontiers in Public Health diagnoses the collapse of public trust and prescribes teaching citizens “critical media engagement,” better media literacy so readers can sort credible guidance from noise. Notice the direction of the fix: train the audience’s eyes, not the institution’s. It is a revealing reflex from an apparatus that spent the COVID years overselling what it did not know and dismissing what it could not control, and then decided the failure was that the public misread it. Prasad, a man who sat in the regulator’s chair and left it, is pointing the other way. The blind spot is not in the audience. It is in the agenda.
So watch a calendar and two decisions. Medicare’s door for multi-cancer tests does not open until 2029, and nothing is covered until the FDA authorizes a test and CMS issues its national coverage determination; those are the moments that will convert Prasad’s net-benefit question into a live experiment on real patients. The class is not running on surrogates alone – semaglutide’s SELECT trial already cut a composite of cardiovascular death, nonfatal heart attack, and nonfatal stroke, even if that is not the same as proving people live longer in every population. What stays unresolved is Prasad’s narrower challenge: whether retatrutide, still an investigational drug, buys any survival edge over tirzepatide – a question the phase 2 weight-loss data he points to simply cannot settle. The next CAST, if it is out there, will not announce itself. It will surface the way the last one did, in a death count someone tallies after the fact, and the only open question is whether anyone in public health is looking in time to see it.
Sources
- Vinay Prasad, “Public health fails in ways it can’t even see” (2026)
- PATHFINDER, The Lancet (2023)
- GRAIL NHS-Galleri full results (2026)
- Public Law 119-75, section 6221, Medicare MCED coverage framework
- CAST, NEJM (1991)
- American College of Cardiology, CAST and the surrogate-endpoint lesson
- NEJM, CAST preliminary report on encainide and flecainide (1989)
- NEJM, retatrutide phase 2 obesity trial
- TRIUMPH-5, retatrutide versus tirzepatide in obesity (Eli Lilly)
- STAT, Vinay Prasad to depart FDA’s CBER
- Frontiers in Public Health, critical media engagement and public-health trust (2026)