Case Study 28.1 — Nesiritide: Approved on Surrogates, Answered by Outcomes

What it costs to find out late

The setup

By the late 1990s the natriuretic peptide system had been understood for nearly two decades, and the inference sitting in front of the field was almost irresistible.

The heart, when overloaded, secretes a peptide that tells the kidney to shed sodium and water, tells the blood vessels to relax, and tells the renin-angiotensin-aldosterone system to stand down. That is a precise description of what a patient in acute decompensated heart failure needs: too much volume, too much vascular tone, and a compensatory hormonal system that has become part of the problem.

A patient arriving in that state is drowning slowly. Fluid in the lungs, breathless lying flat, often frightened. The available treatments — loop diuretics, nitrates, occasionally inotropes — work but are old, imperfect, and in the case of inotropes carry known mortality concerns.

The reasoning was: the body already makes the right drug. It is just not making enough of it, or not enough of it is getting through. Give more.

Nesiritide was recombinant human BNP — not an analog, not a modified sequence, the actual 32-amino-acid human peptide produced in bacteria. There was no question about whether it was the right molecule. It was, by definition, the molecule.

What the approval rested on

The pivotal evidence was hemodynamic and symptomatic.

Infused nesiritide lowered pulmonary capillary wedge pressure — a catheter-derived measurement of left-sided filling pressure, and about as objective a number as cardiology produces. It is not a questionnaire, not a subjective assessment, not a proxy invented by a marketing department. It is a pressure, measured in millimeters of mercury, in a chamber of the heart. It went down.

Patients also reported less breathlessness at a few hours compared with placebo.

The drug was approved in the United States in 2001 for acute decompensated heart failure with dyspnea at rest.

Nothing in that sequence was fraudulent, careless, or unusual. The measurement was real. The symptom improvement was real. The regulatory pathway that accepts such evidence exists for good reasons: it is faster, it is cheaper, and for a condition where patients are acutely miserable, "they breathed more easily" is not a trivial finding.

But note what the evidence base did not contain. Nobody had shown that patients given nesiritide were less likely to die, or less likely to be back in the hospital in a month. The question had not been asked, because asking it requires a trial an order of magnitude larger than the trials that supported approval.

In Chapter 5 §5.6 vocabulary: the approval rested on a surrogate endpoint and a short-term symptom score. The hard endpoints were unexamined.

The doubt

Nesiritide entered practice and was used widely. It was expensive relative to the alternatives, and its use grew quickly.

Within a few years, other researchers — not the sponsor — pooled the randomized data that existed and published re-analyses raising two concerns. One suggested an association with worsening renal function. Another suggested a possible increase in short-term mortality.

These analyses had real limitations, and the limitations were argued about vigorously. Pooling small randomized trials retrospectively, with endpoints those trials were not designed to measure, is a method that generates hypotheses rather than conclusions. The signals were not large and not definitive.

But they were not dismissible either. A widely used drug, approved on surrogates, now carried a credible question about whether it might be causing harm on the outcomes nobody had measured. There was no way to answer that question with the existing data, because the existing data had never been collected to answer it.

The manufacturer commissioned a definitive trial. That decision is worth stating plainly: it was expensive, it was not legally compelled, and the sponsor had more to lose from the answer than anyone. They did it anyway.

The trial

A randomized, double-blind, placebo-controlled trial in roughly 7,100 patients hospitalized with acute decompensated heart failure across many countries. Nesiritide or placebo was added on top of standard care — the only ethical design, and also the only design that answers the practical question, which was never "nesiritide or nothing" but "does adding this help?"

The endpoint structure deserves attention because it was well designed. Co-primary endpoints: the change in patient-reported dyspnea at 6 and 24 hours, and the composite of rehospitalization for heart failure or death at 30 days.

They kept the symptom endpoint that had supported the approval and added the hard endpoint that had never been tested. That lets one trial answer both the original question and the new one, and it makes the result far harder to argue with in either direction.

The results:

• Dyspnea improved slightly more with nesiritide. The difference did not meet the prespecified threshold for statistical significance.
• The 30-day composite of death or rehospitalization was essentially identical between groups.
• Hypotension was more common with nesiritide.
• The mortality and renal function signals from the earlier pooled analyses were not confirmed.

Read the last two bullets together, because the story is frequently told wrong. The trial did not show that nesiritide was dangerous. It showed that nesiritide did not help.

What it cost

Use of nesiritide declined sharply after the trial reported. It is no longer a meaningful part of practice.

The ledger, stated without melodrama:

A decade of use. Patients received an expensive drug that, on the outcomes that matter, did nothing for them. They were not harmed by it in any way the definitive trial could detect, but they were not helped either, and they and their insurers paid for it.

Opportunity cost in the field. Attention, enrollment capacity, and money went to a strategy that did not work. Acute heart failure remains a condition with strikingly few outcome-improving therapies.

A decade of uncertainty. Between the first pooled re-analyses and the definitive trial, patients and clinicians made decisions in a fog of unresolved suspicion — the worst possible epistemic state, in which a drug is neither trusted nor abandoned.

And a genuine gain, which should not be overlooked. The field learned something durable about a strategy, not just about a molecule. When a second natriuretic peptide was later tested in acute heart failure in its own large randomized trial, it too improved hemodynamics and it too failed to improve cardiovascular mortality. Two independent negative results on the same strategy is a finding about the strategy — and it is a finding that would not exist if the nesiritide question had been left open.

Discussion questions

1. The evidence supporting nesiritide's approval included a catheter-measured pressure — about as objective a physiological measurement as medicine offers. Does objectivity make a surrogate endpoint more trustworthy as a predictor of clinical benefit? Argue both sides, then state your position and what would change it.

2. The sponsor commissioned the definitive trial voluntarily, at great expense, knowing the answer could end the product. Design an incentive structure that would make that behavior routine rather than admirable. Where does your structure create new problems?

3. The re-analyses that raised the alarm were performed by researchers who had not run the original trials, using data those trials had generated. What conditions have to hold for that kind of external scrutiny to be possible at all? Which of those conditions is most fragile?

4. "The trial did not show harm; it showed absence of benefit." Explain why that distinction matters — for how the drug should have been described afterward, for what patients who received it should have been told, and for how the ❌ rating in §28.5 should be read. Then explain why the same distinction is often unavailable for the compounds in Part III of this book.

5. Section 28.4's ⚠️ Hype Check argues that a better mechanism should make you more careful, not less. Apply that argument to nesiritide specifically: identify the three features of the mechanistic story that made it especially persuasive, and for each, say what a skeptic could have pointed at before the trial. Then state clearly why none of those skeptical points would have counted as evidence.

6. Suppose you are advising a regulator considering approval of a new peptide for acute heart failure on the basis of improved hemodynamics and symptoms — the exact evidence package nesiritide had. Knowing this history, what would you require? Be specific about the trade-off: your requirement delays access for patients who are acutely ill now, in exchange for information that arrives years later. Defend the trade you choose.