Case Study 1 — The Surrogate That Killed People

CAST, and the trial that had to be stopped

Type: Real, public, historical · Tier 1 facts · Relevance: §5.6, §5.2

Background: a mechanism everyone believed

In the 1980s, cardiologists faced a well-defined problem. People who survive a heart attack are at elevated risk of sudden cardiac death, and many of them show premature ventricular contractions — extra, abnormal heartbeats — on monitoring. The more frequent these abnormal beats, the higher the observed risk of dying suddenly.

The reasoning was straightforward and, on its face, excellent:

  1. Abnormal heartbeats are associated with sudden death.
  2. Sudden death after a heart attack is usually caused by a fatal arrhythmia.
  3. Antiarrhythmic drugs suppress abnormal heartbeats.
  4. Therefore, suppressing abnormal heartbeats should prevent sudden death.

This was not a fringe position. It was standard reasoning, supported by pathophysiology, by observational data, and by drugs that demonstrably did step 3 — they suppressed the abnormal beats reliably and measurably.

Antiarrhythmic drugs were widely prescribed for this purpose. The practice was established.

The trial

The Cardiac Arrhythmia Suppression Trial (CAST) was launched in the late 1980s to confirm the benefit. Randomized, placebo-controlled, in patients who had survived a heart attack and had suppressible ventricular ectopy. The drugs studied included encainide and flecainide.

Note the framing: to confirm. The trial was widely regarded as verifying something already known. There was genuine ethical discussion about whether randomizing patients to placebo was defensible, given that the treatment was believed to work.

The trial was stopped early.

Not for benefit. Mortality was higher in the treated groups. Patients receiving the antiarrhythmic drugs were dying at a greater rate than those receiving placebo.

The drugs did exactly what they were designed to do. They suppressed the abnormal heartbeats. The surrogate improved. And the patients died more.

🔬 Read the Study — CAST

text FIGURE 5.CS1 — "The trial that was supposed to confirm" [real published trial] THE STUDY Randomized, double-blind, placebo-controlled trial of antiarrhythmic therapy (encainide, flecainide) in patients after myocardial infarction with suppressible ventricular ectopy. Publicly funded. Late 1980s; stopped early. THE QUESTION Does suppressing premature ventricular contractions reduce mortality after a heart attack? WHAT IT SHOWS No — the opposite. Mortality was HIGHER in the treated arms. The trial was terminated early for harm. WHAT IT DOESN'T It does not show that all antiarrhythmic drugs are harmful, that the association between ectopy and sudden death was spurious, or that arrhythmia is not a mechanism of sudden death. It shows that suppressing THIS surrogate with THESE drugs did not deliver the outcome. THE VERDICT Definitive for the claim tested. One of the most consequential negative trials in the history of medicine. THE LESSON A surrogate endpoint is only as good as the evidence that changing IT changes the OUTCOME — and that evidence is a separate research program, not an assumption. Here the arrow from surrogate to outcome pointed the wrong way, and thousands of people had been treated on the strength of it.

Why the reasoning failed

Working out where the chain broke is more instructive than the result itself.

Steps 1 and 2 were correct. Ventricular ectopy really is associated with sudden death, and sudden death after myocardial infarction really is often arrhythmic.

Step 3 was correct. The drugs suppressed the abnormal beats. This was measured and reproducible.

Step 4 was the error, and it was an error of a specific kind. The abnormal beats were a marker of an unhealthy, electrically unstable heart — not the mechanism by which that heart killed people. Suppressing the marker did not treat the instability. And the drugs had a separate property: under conditions of ischemia, they could themselves provoke fatal arrhythmias. The benefit was illusory and the harm was real.

Map that onto §5.6's diagram and every one of the three failure modes appears at once:

  • The drug affected the surrogate by a route that did not connect to the outcome
  • The surrogate was a marker of the disease rather than a cause of it
  • The drug had a separate harmful effect that outweighed the benefit

All three, in one case. This is why CAST is taught in every clinical epidemiology course.

What this case teaches about peptides

Nothing in the reasoning was stupid. That is the point. The mechanism was well characterized, the observational data was real, the drugs demonstrably did what they were designed to do, and the clinical community was confident enough to consider a placebo arm ethically questionable.

And it was wrong, in the direction that killed people.

Now consider the surrogates in this book. Raises growth hormone. Increases IGF-1. Increases lean mass. Reduces an inflammatory marker. Improves collagen synthesis in cultured cells. Every one of these is a real, measurable change. Every one is presented, routinely, as though the arrow to a clinical outcome were established.

For most of them, it is not. Chapter 15 will show that "raises growth hormone" and "improves body composition and function in healthy adults" are separated by an unbridged gap. Chapter 16 will show that lean mass and functional capacity can move independently. Chapter 30 will show that an instrument-measured skin change and looking better in a mirror are different claims.

The lesson is not that surrogates are useless. They are essential — a trial measuring hard endpoints in a slowly progressing disease takes years and thousands of people, and without surrogates much of drug development would be impossible. The lesson is that a surrogate carries an assumption, that the assumption is testable, and that it has failed before in ways that cost lives.

The uncomfortable coda

CAST is usually told as a triumph of evidence-based medicine: rigorous trial overturns confident practice, medicine self-corrects, lives saved going forward.

That telling is correct, and it leaves out the timing.

The practice was widespread before the trial. The trial was undertaken to confirm what was believed, and it required substantial effort and some resistance to run at all, precisely because the answer was thought to be known. In the years before it reported, patients were treated on the strength of a mechanistic argument and a surrogate.

The self-correction worked. It worked slowly, and after the fact, and only because someone insisted on running the trial anyway.

Hold that against Part III, where several compounds are in wide use on the strength of mechanism and a surrogate, and where the trial that would settle the question has not been run — not because it was attempted and failed, but because nobody with the resources has an incentive to run it. In CAST's case, the drugs were patented, the practice was mainstream, and public funding existed for the question. For an unpatentable research compound sold by companies with no regulatory obligation, neither the money nor the requirement exists.

Self-correction is not automatic. It is a thing people do, and it requires that somebody be able to afford to.

Discussion questions

  1. Reconstruct the four-step reasoning that led to treating post-heart-attack patients with antiarrhythmics. Identify precisely which step was wrong and why the error was invisible in advance.

  2. There was ethical debate about whether a placebo arm was justified, given the strength of belief in the treatment. In hindsight the placebo arm was the safer arm. What does this suggest about how to weigh confident clinical belief against the value of a trial?

  3. §5.6 lists three ways a surrogate can fail. CAST exhibits all three. Take a peptide surrogate — "raises growth hormone" — and describe what each of the three failure modes would look like for it.

  4. CAST was publicly funded. Would a company selling encainide have run this trial? What does the answer imply about which questions get definitively answered and which do not?

  5. Consider the coda. For a compound with no patent, no regulatory obligation, and an established market, who would fund the trial that could show it does not work? Is there any mechanism that would produce one?

  6. A defender of a peptide compound argues: "CAST shows trials sometimes overturn confident belief — so we should be humble about negative claims too, and not dismiss compounds just because trials haven't been run." Evaluate this argument. Is it a fair application of the lesson, or a misuse of it?