Aprotinin: answered in 1992, tested 52 more times

64 trials asked whether one drug cut bleeding in heart surgery. The answer was in by the 12th, and the trials kept coming for another ten years.

Rohit Pawar · · 6 min read

Key findings

  • Researchers ran 64 randomised trials of aprotinin, a drug that cuts bleeding in heart surgery, between 1987 and 2002, though its effect on transfusions was settled by the 12th, in June 1992.
  • The trial reports cited a median of 20% of the trials published before them, and only 7 of 44 later reports cited the largest trial, which had 1,784 patients.
  • Across 1,523 trials in 227 meta-analyses, reports cited fewer than a quarter of the earlier trials on the same question, a median of 2 however many there were.

I keep telling people that not checking the literature properly costs more than time. I wanted a real case to point to, something documented with numbers rather than a story about a friend of a friend. The best one I found comes from heart surgery, and it is worse than I expected.

What happened with aprotinin

Aprotinin is a drug given during heart surgery to reduce bleeding, so that fewer patients need donated blood. Between 1987 and 2002, researchers ran 64 randomised trials comparing it with a placebo or with no drug at all. Most were small. The median trial had 64 patients, and the biggest had 1,784.

In 2005, Dean Fergusson and colleagues pooled those trials in the order they were published, adding one at a time to see when the answer became clear. It was clear by the 12th trial, in June 1992. By then the pooled data showed aprotinin cut the odds of a transfusion to about a quarter (an odds ratio of 0.25), and the trials that followed barely moved that number. A systematic review pulling the early trials together came out in 1992 as well.

Then came another 52 trials.

They didn't change the answer. What they did do was give some patients a placebo or nothing during surgery, at a point when the drug was already known to help, and use up funding, staff time and years of people's work that could have gone into questions that were still open.

Why did they keep going?

From the outside you can't know what each team knew when it started. Some may have read every earlier trial and gone ahead anyway. But the papers they published do tell you something. A typical trial report cited 20% of the trials that came before it (that's the median). The largest trial, 28 times the size of a typical one, was cited by just 7 of the 44 reports published after it.

So most teams either didn't find most of the earlier work, or didn't treat it as relevant. Fergusson's group said as much: investigators weren't adequately citing earlier research, and that is how so many trials ended up re-asking a question that had already been answered. Their suggestion was that ethics boards and journals should require researchers to search properly for the earlier evidence.

It wasn't just one drug

I'd like to say aprotinin was a one-off. It wasn't.

In 2011, Karen Robinson and Steven Goodman went through 1,523 trials from 227 meta-analyses, published between 1963 and 2004, and checked how many of the earlier trials on the same question each one cited. The median was 21%. Among trials that had at least five earlier trials to cite, about a quarter cited none, and another quarter cited only one. The typical trial cited two earlier trials, and that stayed at two however many there were to cite.

To be fair to the researchers, Robinson and Goodman are clear that they couldn't tell why, and some teams may well have known about trials they didn't cite. Still, those numbers don't look like a field building carefully on its own past.

The question nobody was asking

The part of the aprotinin story that bothers me most is what all those trials didn't do. A trial with a few dozen patients is big enough to show, once again, that a drug reduces bleeding. It isn't anywhere near big enough to show whether more people die.

That question got a proper trial much later. BART, published in 2008, randomised 2,331 high-risk heart surgery patients to aprotinin or one of two other drugs that reduce bleeding. It was stopped early because more patients on aprotinin died, and aprotinin was suspended across Europe. In 2012 the European Medicines Agency decided BART's results were unreliable and allowed it back, but only for bypass patients at high risk of heavy bleeding.

I'm in no position to say who is right about BART. What does seem clear is which question was still open in the 1990s, and it wasn't the one being asked over and over.

A fair objection

Everything looks obvious in hindsight, and I want to be careful about that.

The textbook example of wasted trials is streptokinase for heart attacks. In 1992 Joseph Lau and colleagues showed that pooling the first eight trials, with 2,432 patients, already pointed to fewer deaths by 1973, and yet 25 more trials with another 34,542 patients followed. In 2024, Kim Boesen and Christian Gluud posted a reanalysis arguing that those early trials were too different from each other and too open to bias to be pooled, and that the two very large trials of the 1980s were justified. It's a preprint and hasn't been peer reviewed, but the point is a fair one.

Aprotinin holds up better because the answer wasn't only visible looking back. Systematic reviews had pooled the trials in 1992 and again in 1997, and anyone planning a trial could have read them. For me, both stories point the same way. You can only judge whether a study repeats earlier work by what was known when it began, so the looking has to happen at the beginning.

What this has to do with climate science

Medicine can measure this because its trials are registered and pooled, question by question. I haven't found anything like that for climate and earth science. I doubt that's because it doesn't happen here. It's just harder to see.

If anything, I think our field makes it easier to repeat work by accident. The same finding can sit in two literatures under two different names, the way marine heatwaves were often written up as warm anomalies before the phrase existed. My own run-in with this was tiny next to these stories: a dissertation topic that had more or less been done already.

That is what I'm building unkwn42 for. Fergusson's group asked researchers to search for the earlier evidence before they start. I want that search to be quick enough that nobody skips it, and good enough to catch the papers that use different words for the same thing. If you have an idea you're about to spend a year on, describe it to unkwn42 first.

Sources

Cite this post

Pawar, R. (2026). Aprotinin: answered in 1992, tested 52 more times. unkwn42 blog. https://www.unkwn42.com/blog/aprotinin-answered-in-1992

This post as Markdown

unkwn42 is the missed-literature check you run before you submit: whether your main claim is already published, which earlier work a referee will expect you to cite, and whether your reference list is clean. Your first check is free.

Run a free check