The Drug Didn’t Fail. The Trial Did.

Every so often, a drug fails in a way that doesn’t quite add up. One well-documented case started, of all places, as a heart study.

In the CANTOS trial, an anti-inflammatory antibody called canakinumab was being tested to prevent second heart attacks. It did something no one was looking for: patients on the drug had markedly lower rates of lung cancer, and lower lung cancer mortality. An inflammation drug, tested in cardiology, had produced a striking anti-cancer signal.

Novartis took the signal seriously and launched three Phase III trials to repurpose canakinumab for non-small cell lung cancer. One after another, they missed their primary endpoints. The standard reading is simple: the signal didn’t hold, and the drug doesn’t work in cancer.

But that reading skips something. When Novartis looked more closely at who responded, one group stood out: patients with high levels of inflammation did noticeably better on the drug. And it turned out the trials that failed had enrolled very different patients from the original heart study, where the signal first appeared, people with much lower inflammation. The drug may not have failed everywhere. It may just have been tested, mostly, in the wrong patients.

I’m not arguing that canakinumab is a secret cancer cure. It may genuinely not work here. But that’s exactly the point: looking at those trials, it is honestly hard to say whether the drug failed, or whether the trials failed to find the patients it could have helped. Those are different conclusions, and the field almost always files them under the same word. Failed.

The Pattern Behind the One Case

What makes canakinumab useful is how well documented it is. But the basic story is common. A drug is tested in a big, mixed group of patients, it doesn’t work well enough overall, and development stops, even though a smaller group inside the trial may have genuinely benefited. The headline becomes “the drug failed.” What actually happened is narrower: it failed for most of the patients it was tested on, which is not the same thing.

It’s worth being careful here, because this idea is easy to oversell. Not every failure is a hidden success. Sometimes a drug doesn’t work, full stop, and the honest answer after a failed trial is that there was nothing to find. Canakinumab might be one of those cases. It failed several large trials, and a promising subgroup is only a clue, not proof. The point isn’t that failures are secretly wins. It’s that we usually can’t tell, and we rarely check.

So the real difficulty isn’t noticing that failures contain information. Most researchers know that. The difficulty is telling which failures are worth a second look and which aren’t, separating a real signal from noise and wishful thinking. That’s a genuinely hard call, closer to clinical judgment than to data processing. And mostly, no one makes it systematically. A failed drug gets shelved, the team moves on, and whatever was in the data stays in the data.

Why Failures Rarely Get a Second Look

Going back through failed drugs to see which ones were misread mostly doesn’t happen. There are a few reasons.

One is the data itself. As I described in an earlier piece, the information you’d need is scattered and incomplete, spread across registries, filings and internal records. Working out why a drug really failed means pulling all of that together first.

The second reason runs deeper. Even once you have the evidence, reading it well takes two different kinds of expertise. One is technical: making sense of the data, checking it, and rebuilding what a drug actually did and in which patients. The other is clinical: looking at a promising result in a small group of patients and judging whether it points to real biology or is just a false signal. Someone strong at the first isn’t always strong at the second. That mix is rare, and without it, revisiting a failure is slow and easy to get wrong.

What’s new is that both are becoming more possible. AI can now pull scattered evidence together and analyse it in days rather than months, drawing on the accumulated record of decades of research. And there have never been more failed drugs to go back to: late-stage failures are, if anything, increasing, leaving a growing pile of shelved drugs no one has revisited.

This Is What R2H Is Building

It doesn’t have to be left to chance anymore. Rescuing a failed drug isn’t new, but it has usually happened by accident. Sildenafil was being tested for heart disease when it revealed a very different use. What’s new is that it no longer has to be an accident: with the right evidence, you can take a shelved drug on purpose, rebuild what actually happened to it, and decide, honestly, whether it truly failed or was simply misread.

That is what R2H built NeoRevive, its drug rescue platform, to do. It runs on the two kinds of expertise that make this hard. Clarisyn, the system underneath it, handles the technical side: it works across published literature and biomedical data, and its engine doesn’t just retrieve evidence but analyses it, cross-checks it, and reconstructs what can actually be established about why a drug failed and in whom. Alongside it, a clinical team does the part no model can, reading that reconstruction with real clinical experience and asking whether a promising signal reflects recoverable biology or a finding that looks convincing and leads nowhere. One side rebuilds the evidence; the other decides what it means.

Clarisyn came first, built for this problem, and it turned out useful enough on its own that others now use it as a research tool. But its real purpose is NeoRevive. None of this means every shelved drug can be saved; most can’t. The point is narrower, and more interesting: some real treatments are sitting in the failure pile right now, filed under the wrong verdict, and almost no one has the mix of technical and clinical judgment it takes to find them.

Which brings it back to canakinumab. Somewhere in that trial data may be a group of patients the drug actually helped. Maybe not. The honest answer is that no one has done the work to find out. That is the opportunity hiding in plain sight: not only in the molecules still being designed, but in the ones already given up on.