Maya Chen: Nathan, hey — can I ask you something before we even really start?
Dr. Nathan Hayes: Sure — though knowing you, 'before we start' is the start.
Maya Chen: Fair. So — when a doctor tells a patient 'your blood sugar is down,' that's good news, right? That's the whole point. Except I've been thinking about rosiglitazone, which did exactly that — lowered blood sugar, durably, in Phase II trials. The FDA approved it on that basis. And then Phase III revealed the drug was increasing heart attack risk.
Dr. Nathan Hayes: Hold on — increased risk, even while blood sugar remained lower?
Maya Chen: Yeah. The measurement was still doing what it was supposed to do. The outcome wasn't.
Dr. Nathan Hayes: Now — this is actually the cleanest demonstration I know of why a surrogate endpoint and a clinical outcome are not the same claim. A surrogate is a substitute measure — blood sugar, cholesterol level, tumor size — used in place of what we actually care about, which is whether the patient lives longer or gets sicker or doesn't. The assumption linking those two things is that biomarker change reliably predicts outcome change. Rosiglitazone is the case where that assumption was not just wrong — it was wrong in a way that caused harm.
Maya Chen: Because moving the measurement through a pathway that isn't causally connected to disease prevention — that's not neutral. That might be actively doing something else.
Dr. Nathan Hayes: Precisely. A drug affecting a biomarker has proven exactly one thing: it can move that number. It has not proven the mechanism responsible for that movement is the same mechanism that protects against disease. And the FDA, sitting in the middle of all of this — the body that separates Phase II biomarker data from Phase III clinical outcome data — grants Accelerated Approval on surrogates that are sometimes nonvalidated. The rosiglitazone case is what that risk looks like when it materializes.
Maya Chen: Mm — and I think what sits with me is the trust dimension. The patient who was told 'your blood sugar is down, this drug is working' wasn't wrong to believe that. The measurement was real. But the meaning we attached to it wasn't.
Dr. Nathan Hayes: Which is the gap this entire conversation is going to live inside — between what a biomarker proves and what we implicitly claim it proves.
Maya Chen: Right — and whether we've ever really been honest about that gap, or whether we've just... relied on people not asking.
Dr. Nathan Hayes: And here's where I want to push on that — because what that gap actually looks like, mechanically, is something most people can picture once you give them the right image. Imagine a smoke detector going off. And instead of finding the fire, someone just fans the detector with a newspaper until the alarm stops. The alarm is fixed. The measurement is fixed. The fire is still there.
Maya Chen: Oh — yeah. The newspaper doesn't touch the fire at all.
Dr. Nathan Hayes: Not at all. And a drug that moves a biomarker without addressing the disease mechanism is doing exactly that. You've silenced the alarm. You haven't changed what the alarm was measuring.
Maya Chen: So the number looks better and the patient feels reassured and the doctor has something concrete to point to — but the actual process causing harm is still running.
Dr. Nathan Hayes: Right. Now — and this is the complication, because I don't want to leave the impression that biomarker movement is always noise — statins. Statins lower LDL and they genuinely reduce cardiovascular events. That evidence is robust. That's how LDL became a validated surrogate in the first place.
Maya Chen: Wait — so statins are actually the reason we trust LDL as a number?
Dr. Nathan Hayes: Essentially, yes. But here's what that means, and this is the part people miss — that validation is specific to statins' mechanism. It is not a property of LDL itself. So another drug comes along, also lowers LDL, and we think — well, LDL is validated, we're fine. Except no. That new drug has to earn the validation separately, through its own mechanism.
Maya Chen: It can't just borrow the statin's track record.
Dr. Nathan Hayes: It cannot. Torcetrapib — the CETP inhibitor — raised HDL, lowered LDL, moved both numbers in the direction we wanted. The ILLUMINATE trial enrolled over fifteen thousand high-risk patients and had to be stopped early because mortality went up. Not flat. Up. And one reason identified was off-target blood pressure elevation — a pathway that had nothing to do with the lipid numbers the drug was moving.
Maya Chen: Fifteen thousand people. And the biomarker was doing exactly what it was supposed to do the whole time.
Dr. Nathan Hayes: The whole time. The alarm looked better. The — the fire, in this case, was actively worse. Because the drug was operating through a mechanism that was, I mean, orthogonal to disease prevention — and doing something harmful alongside it.
Maya Chen: So we've been assuming — sort of baking in the assumption — that the biomarker is the cause, when it might just be the alarm. And the alarm can be quieted a hundred different ways, and only some of those ways actually put out the fire.
Dr. Nathan Hayes: And that's exactly where torcetrapib stops being a story about one bad molecule — because the class didn't stop there. Dalcetrapib, evacetrapib, anacetrapib, obicetrapib — all CETP inhibitors, all raising HDL, all failing to deliver the cardiovascular benefit the lipid numbers seemed to promise.
Maya Chen: Wait — all of them?
Dr. Nathan Hayes: Every one. Now, torcetrapib had the blood pressure problem — that off-target effect muddied the picture. But dalcetrapib, evacetrapib — no meaningful off-target toxicity identified, and they still didn't move outcomes. Which means the failure wasn't just the drug. It was the assumption that raising HDL through this particular mechanism would protect against cardiovascular disease.
Maya Chen: So the blood pressure thing with torcetrapib — that almost let us off the hook, conceptually. Like, 'well, that one had an extra problem.' And actually the extra problem was covering for a deeper one.
Dr. Nathan Hayes: Correct. The off-target effect was real. But it was also... almost convenient, narratively. It gave the class a reason to keep going.
Maya Chen: Okay, because I keep thinking about — imagine you're a cardiologist in 2006, you're reviewing a patient's labs. Torcetrapib is in the trial. HDL is up, LDL is down, every number is going exactly the direction you trained to want. And somewhere in the ILLUMINATE data, the mortality signal is already accumulating. You just can't see it yet.
Dr. Nathan Hayes: And you would have had no clinical reason for concern. The mechanism you were taught — lower LDL, raise HDL, reduce events — was apparently executing perfectly.
Maya Chen: That's what's sort of — I mean, it's not a failure of the cardiologist. The information they had was genuinely pointing the wrong way.
Dr. Nathan Hayes: Which is why niacin matters as the second data point. Niacin improved HDL and triglyceride biomarkers — and when it was added to effective statin therapy, cardiovascular events didn't budge. No reduction. And here, importantly, there wasn't a blood pressure artifact to blame. The lipid numbers moved. The outcomes didn't follow. The pathway simply wasn't causal.
Maya Chen: Hmm. So niacin is almost the cleaner demonstration, in a way — because you can't explain it away.
Dr. Nathan Hayes: Exactly. And what both cases share is this — the drug was moving a biomarker through a mechanism that was not the mechanism driving the disease. The alarm looked better. The underlying process was indifferent.
Maya Chen: Which makes me wonder — and we'll get into this — what it means that the FDA built an entire Accelerated Approval pathway around surrogates, and the EMA found that the accuracy with which those surrogates predict actual outcomes is, quote, 'not known.' That part gets heavier once you've sat with torcetrapib for a minute.
Dr. Nathan Hayes: It does. And the question is: how many times does the biomarker have to move in the right direction while the patient gets worse before we treat that as a structural problem, not an exception?
Maya Chen: And yet the FDA built a formal pathway around exactly that. Like, not informally — the Accelerated Approval Pathway is the structure that says: nonvalidated surrogate, conditional approval, confirmatory trial to follow. That's... that's the institution endorsing the assumption.
Dr. Nathan Hayes: Now — and here's where the EMA finding lands hardest. Their cross-sectional analysis, 2011 to 2018, looked at how often expedited pathways accepted validated versus nonvalidated surrogates. And the document explicitly states that the frequency and accuracy with which those surrogates correspond to actual clinical outcomes is — quote — 'not known.'
Maya Chen: Wait — they said 'not known'? That's — that's not 'imprecisely known' or 'partially characterized.' That's just... we don't know.
Dr. Nathan Hayes: Those are their words. And the approvals kept happening anyway. Which means the evidentiary gap isn't a bug they're working to close — it's a documented feature of the operating system.
Maya Chen: Okay but — and I want to push on this because I think it sounds worse than even the pathway defenders would admit — what's the argument for? Like, what is the genuine case for running that risk?
Dr. Nathan Hayes: The genuine case is speed for people who are dying now. And in rare disease, it's not even a trade-off — it's a constraint. Cystic fibrosis. Patient populations are too small to power a phase III trial on clinical endpoints. You cannot enroll enough people. So biomarker-based development isn't a regulatory shortcut, it's the only feasible path.
Maya Chen: Right — that's real. I mean, you can't manufacture a larger patient pool.
Dr. Nathan Hayes: But — and this is the critical move — cardiovascular disease is not cystic fibrosis. Type 2 diabetes is not cystic fibrosis. Those patient pools are enormous. You can absolutely power an outcome trial. The CF logic doesn't migrate. It was born from necessity, not from evidence that biomarkers are sufficient.
Maya Chen: So someone takes the CF exception — which is a genuine, constrained, there's-no-other-way exception — and it becomes... permission, sort of, for diseases where there absolutely is another way.
Dr. Nathan Hayes: That conflation is — I mean, it might be the core problem. Primary biliary cholangitis. Three second-line therapies, conditional approval, all based on liver biochemistry surrogates. And the AASLD-EASL consensus has not accepted those surrogates as validated efficacy endpoints. Not rare enough to plead the CF constraint, not common enough to have the trial infrastructure already built. And the approvals happened anyway.
Maya Chen: Three therapies.
Dr. Nathan Hayes: Three. Conditional. And the confirmatory trials — they're supposed to follow. But when they're delayed, or when they fail, the patients who took the drug in the interim bore that risk without outcome evidence. And there is no clear accounting for who owns that uncertainty. Not the regulator, not the manufacturer — it just... lands on the patient.
Maya Chen: That's — yeah. That's the part that doesn't resolve cleanly. The doctor said 'your liver enzymes are better.' And that was true. And whether it meant anything, no one had actually established. The patient had no way to know they were, sort of, inside an open question.
Dr. Nathan Hayes: And that's — I mean, that's the thing I can't find a clean answer to. If the FDA approves on a surrogate today and the confirmatory Phase III fails three years from now, those patients who took the drug in the gap bore a risk that wasn't measured. Rosiglitazone is the version of that where the Phase III didn't just fail — it revealed active harm. Increased heart attack risk. While blood sugar stayed lower.
Maya Chen: And nobody formally owned that. The harm just... landed.
Dr. Nathan Hayes: No one did. And what stays with me — genuinely, I keep returning to this — is that Phase III trials measuring mortality, actual morbidity, meaningful functional change at scale, those remain the only non-negotiable test we have. There is no Phase II biomarker signal, however clean, however large, that substitutes for that. And the distance between that standard and where we actually operate is... it's not small.
Maya Chen: It's not small. And I don't — I can't resolve who should have to answer for the people in that gap. That's the part I came into this thinking I'd land somewhere on, and I haven't.
Dr. Nathan Hayes: Nor have I. I think that's genuinely open.
Maya Chen: Yeah. The measurement worked. The outcome didn't. And the person sitting in their cardiologist's office hearing 'the numbers look good' — they had no frame for what they didn't know.
Dr. Nathan Hayes: None. And I'm not sure the system currently requires anyone to give them one.
Maya Chen: Mm. That's — that's where I'll just sit for a while, I think.
Dr. Nathan Hayes: Same. Good conversation, even if it didn't close.
