Following the Wakefield fraud story, I’ve seen several blogs and blog commenters suggesting that The Lancet was at fault for failing to catch the fraud in the peer review process. I (and I think most practicing scientists) don’t agree with that condemnation, which I think shows that many people outside the field don’t really understand what peer review is supposed to do. I’m just going to together a couple of comments I’ve made on other blogs …
At Hoyden About Town, I said (in part):
There’s been lots of criticism of The Lancet for publishing this study (they’ve since caused it to be retracted) — but I don’t agree with most of that criticism. Scientific journals should try to catch fraud, but this kind of fraud, wholesale fakery, is ironically hard to catch than the simpler forms where, say, a single part of one figure is changed. It’s not possible for a journal to go back and re-scrutinize all the primary data in all the papers it publishes, for example. It’s necessary to rely on the scientific process as a self-correcting mechanism. Of course, that’s pretty much what happened in this case — Wakefield’s work was rapidly and thoroughly refuted in the scientific literature — but the mainstream press has lagged far behind the scientific consensus. If it wasn’t for ambulance chasers, scandal-seeking newspapers, ignorant and naive reporters, and greedy lawyers, this would have diappeared within a year of Wakefield’s first article, as happens with almost all the mistaken, careless, and misinterpreted scientific papers that are published by the dozen every day.
Here I’ve bolded a sentence that’s worth re-emphasizing: This was fraud, but it didn’t directly set science back very far, because good science refuted it quickly and rapidly. It was the non-science world that set back research on autism, by accepting what scientists already knew was wrong.
Effect Measure made a similar point about peer review:
Wakefield provided the case summaries (which we now know were doctored) and a reviewer would not have had access to or had the time to look at the original medical records. The same is true for the journal. Accurate representation of raw data is taken on trust. I don’t think The Lancet can be taken to task for not catching this. This kind of scientific misconduct is only found after the fact.
In the comments on the Effect Measure article, Sam C made an interesting suggestion:
Review is fine for good science and for research where errors will not have far-reaching impact (so the normal process of correction and extension by future workers is appropriate).
But cases like this need audit. An audit can not always detect deliberate fraud (just as in financial auditing), but it might pick up errors of protocol (like the O’Leary lab’s inadequate controls in their DNA/RNA work) or substandard or imperfect work (inappropriate statistical techniques, equipment whose limitations are not understood, results transferred incorrectly, etc.).
Engineering organisations use ISO standard QA systems, but these only work if applied correctly and conscientiously.
Perhaps any grant award should require that some percentage of the award be allocated to an independent audit of techniques, results and conclusions?
I replied to his comment as follows:
Sam, this is an interesting idea I haven’t heard floated before. I think it’s not doable as a portion of every grant, but I wonder if there could be a separate fund set aside specifically for audits. I’m not sure how it could work, and it would be a real problem to get it balanced properly, though — if there were a standing committee or organization, I could see it getting bogged down in bureaucracy, dinging every paper they come across for trivial procedural errors (“Patient #214 signed the form but failed to initial the 17th page”) so that genuine problems would be hidden anyway.
I don’t know much about formal audit procedures. Is there a precedent for a useful type of audit that would focus on fraud detection and completely broken research, without getting bogged down in trivia?
I want to make a mention of the Journal of Experimental Medicine and the Journal of Cell Biology, which seem to me to be taking a much more proactive attitude toward fraud than most other scientific journals:
“The issue of data integrity should not be left to chance and probability. This is scholarly publishing, not blackjack.”
–M. Rossner (2008). A false sense of security The Journal of Cell Biology, 183 (4), 573-574 DOI: 10.1083/jcb.200810172
I like the idea of an audit, I’m just not sure how the implementation would occur. How do you propose the audit take place? Which papers get flagged for audit (or do they all)? What would get audited?
The most immediate analogue i can think of is the Bernie Madoff scandal. His auditor noted that there is very little you can do if fake books (or in this case fake data) are being used. Granted, that’s the auditor’s job, to crunch through fake books, but I suppose I’m curious what would fall under the audit.
To the point about faulty controls — at what point is that “scientific editorializing” (what constitutes as a sufficient control, or a sufficiently good study) and at what point is the audit not going far enough?
Would like to get your thoughts, as I have no clear answers in my mind.
First, I agree with you in that it is ridiculous to assume that any kind of peer review or editorial action would be able to detect serious fraud.
On the other hand, I don”t think I would like the idea of scientific auditing. Most likely, the most immediate effect would be to make scientific research much more bureaucratic.
I think that science can live well with the way things are currently organized – as long as people don’t put too much trust in results that have been found only by a single group. Nothing beats independent validation by a competing group.
We are probably wired to be outraged by fraud. Too much so in this case, if our outrage is tempting us to curse science with nosy bureaucrats.
I agree with the contention that the amount of untrue science will always be large and fraud will always be an insignificant contributor. For a radical perspective see Ioannidis’ much-read “Why Most Published Research Findings Are False” in PLoS Medicine.
I’m not saying we should not be outraged by fraud at all. We should continue to punish it harshly if it comes to light.
Ben, I don’t know how auditing would work. I do have a few ideas of what wouldn’t work. I agree with Kay that there would be a real risk of bureaucracy imposing restrictions on research. I also agree with Kay (and Eric) that the best fraud protection is ordinary scientific procedure. The scientific community knew, pretty quickly, that Wakefield’s claims were wrong, because the ordinary system of reproducibility quickly disproved his work.
The problem, as I said, has been more about the non-scientific people who are taken in by fraud, and who force scientists to repeatedly disprove what they already know is false; and I don’t think an audit procedure would help there.
Where audits might help is if they could catch the high-profile frauds before they reach the public consciousness and start to poison the well. But how audits could quickly focus on problematic research, without themselves poisoning innovation … that I don’t know.
The ideal situation would be for people in the street to have a better awareness of how science works — the simple fact that the vast majority of bold new hypotheses are wrong, that the lone voice refuting dogma is almost always mistaken, that most new findings disappear into the rubbish heap of irreproducibility — and then they might not be so eager to see conspiracies so quickly.
But I think a simple non-bureaucratic audit might be easier to implement than that.
I haven’t contemplated this deeply, but I would think audits would have to be pretty much random and cover only a fraction of scientific works.
It’s hard to imagine how this could be done more simply than just making people’s lab notebooks available, provided that that would in fact amount to a fairly effective deterrent. I would think it would be a pretty large pain to do phony stuff in a way that made it look like you were gradually and progressively “discovering” it, refining your experiments, etc.
Holding in mind real empirical human nature rather than ideals, even this simple scheme is a recipe for mire IMO. The overseers have to justify their existence, and also compete with one another for importance. They are going to end up always hassling a certain more or less set percentage of the overseen – if there’s no fraud to be found, they will hassle X percent of people about (relatively) poor transparency in their lab notes, more or less regardless of whether anyone is poorly transparent absoluter. Result = a huge pain. Transparency standards will be “by committee” and will always be at least slightly oblique to the goals of lab notes – recording stuff in so it will very be clear to you and other people in the lab, with the method of clear expression highly dependent on what the content is and what your way of thinking is. Bearing in mind also that, IME at least, notes are often worked on in a short piece of downtime while you monitor one or two time-sensitive processes going on on the bench, which you are simultaneously contemplating how to clean up super fast in time for some seminar. Biomed is no serene business.
The ideal situation would be for people in the street to have a better awareness of how science works — [...] that most new findings disappear into the rubbish heap of irreproducibility
That would be a really good thing to have in curricula. We learn some pretty strange things in high school. Why did I learn about conic sections, catenaries, and cardioids in the standard math sequence instead of statistics, which was an elective I didn’t take? Stats is as enlightening and mind-strengthening as finding volumes of rotated curves – plus extremely useful, whereas many of those other things are almost completely academic.
Likewise, why so many obscure integration techniques like integration by parts and VU integration. They didn’t do that much for the mind since we didn’t learn the proofs which connect them to the rest of the mathematical world – just used them to grind out problems. Much better would be a little real analysis, which is an epitome of a very navigable and repletely interconnected intellectual edifice – it’s much like geometry in that way, but it is a more potent body of thought.
The US already has an Office of Scientific Integrity. It says its mission is to “monitor institutional investigations of research misconduct and facilitate the responsible conduct of research (RCR) through educational, preventive, and regulatory activities”, which seems to imply a proactive and preventive role, but for what it’s worth the only times I hear about it is through its misconduct investigations. A quick shuffle through their web site doesn’t really turn up anything proactive or preventive (again, I don’t know what a good preventive that wouldn’t be disruptive would be, but apparently they don’t have any kind of it).
Again for the little its worth, as far as I’ve heard the OSI has been quite well run lately. At the time of the Baltimore/Imanishi-Kari case it seemed that the OSI was heavy-handed, highly politicized, and frankly incompetent; but since then I’ve heard very little criticism of it (and I believe, but can’t quickly find eviudence that, the OSI was reorganized after and becaase of the Baltimore case).
Well – good to know it’s running well. I guess it’s possible I’m anti-bureaucratic to a fault, or at least to a prejudice.