The Surrogate Trap: When the Mechanism Works But the Patient Dies
In the late 1980s, cardiologists thought they had cracked the code.
They knew that patients who developed irregular heartbeats, premature ventricular contractions (PVCs), after a heart attack were more likely to die. They had drugs, encainide and flecainide, that reliably suppressed those irregular beats.
The intuition was flawless:
Suppress the arrhythmia → prevent the death.
To confirm it, they ran the CAST trial. What followed did not merely overturn a hypothesis—it permanently damaged medicine’s faith in mechanistic intuition.
The trial was stopped early. Not for success.
Patients taking the drugs were dying at nearly triple the rate of those on placebo. The medications worked exactly as designed. The surrogate endpoint moved beautifully. The patients died anyway—through a different, unanticipated mechanism.
The mechanism worked. The patient didn’t.
CAST is often taught as a historical anomaly, a relic of a less rigorous era. It shouldn’t be.
I’ve watched this failure mode play out in real study team meetings. A biomarker moves, the room gets excited, and slowly—almost imperceptibly—the surrogate becomes the endpoint in everyone’s mind, even though no one ever claimed it was validated.
It is the canonical example of a failure mode that still haunts modern drug development: the surrogate endpoint trap.
The Pattern That Keeps Repeating
CAST was not an outlier. It was an early warning.
Consider a few more recent examples:
- Torcetrapib dramatically raised HDL cholesterol, the so‑called “good cholesterol,” validating decades of epidemiologic association. Mortality went up. The drug was abandoned after a massive Phase III failure.
- Rosiglitazone reliably lowered HbA1c, hitting its glycemic target with textbook precision. Post‑marketing data linked it to increased cardiovascular risk, triggering restrictions and withdrawals.
- BMP (bone morphogenetic protein) trials in spine surgery showed impressive radiographic fusion rates. Patients, however, experienced worse pain, inflammation, and functional outcomes than controls.
In each case, the same story played out:
- A surrogate endpoint that was easy to measure
- A clean, biologically plausible mechanism
- Enormous pressure—scientific, financial, or clinical—to move fast
The surrogate moved in the right direction.
The outcome that actually mattered did not.
Why Smart Teams Fall for This
Surrogates are appealing because they’re fast.
They promise speed. Smaller trials. Earlier readouts. Cleaner narratives for investors and regulators alike. They turn a messy, time‑consuming question—does the patient live longer or feel better?—into a proxy.
And often, surrogates do work.
Blood pressure predicts stroke. Viral load predicts HIV progression. Tumor burden sometimes predicts survival.
The problem is not that surrogates are useless.
The problem is that we routinely treat plausibility as validation.
CAST failed not because the biology was sloppy, but because it was incomplete. The heart is not an isolated electrical circuit. Suppressing arrhythmias did not address the broader physiological consequences of sodium channel blockade.
Mechanisms operate in systems. Surrogates measure one axis and miss the rest.
When Is a Surrogate Trustworthy?
A surrogate earns trust only when it fully captures the causal pathway, predicts outcomes reliably across multiple trials, and the intervention has no meaningful off‑pathway effects.
These conditions are rare. They require meta‑analytic validation—not a single compelling Phase II result—and a willingness to accept that some questions simply take longer to answer.
The Phase II Cliff
This is where companies get into trouble.
Phase II becomes a surrogate‑optimization exercise. A biomarker lights up. A response rate doubles relative to history. The narrative hardens: the drug works.
Capital pours in. Phase III begins.
And then the cliff.
Phase III asks the question Phase II avoided: does this actually help patients?
When the answer is no, the failure looks sudden. It isn’t. It was baked in the moment the surrogate was mistaken for the truth.
The real purpose of Phase II is not to prove efficacy. It is to kill mechanisms that don’t translate, before they consume patients, capital, and time.
Where I Think We’re Still Too Aggressive
Today, surrogate reliance is most dangerous where biology is complex, timelines are long, and pressure is high—oncology, neurodegeneration, cardiometabolic disease.
We increasingly approve drugs that shrink tumors, shift biomarkers, or modify imaging signals without demanding convincing evidence that patients live longer or live better.
Some of these bets will pay off.
Many will look, in retrospect, exactly like CAST.
The Lesson
A working mechanism is not the same as a working therapy.
Surrogates are hypotheses—not truths.
If your development strategy depends on a biomarker being “good enough,” you should be able to answer a brutal question:
What would convince us we’re wrong?
If the answer is “Phase III,” you may already be too late.
The surrogate trap is not a historical curiosity. It is a structural risk.
Ignore it, and the mechanism may succeed—while the patient pays the price.
📬 Want more insights on experimental design across domains? Subscribe to the newsletter or explore the full archive of Evidence in the Wild.
Member discussion