The strange story of the Tamiflu data
In a recent post about chapter 1 of Ben Goldacre's “Bad Pharma”, I mentioned that the chapter included a strange story about how the Cochrane investigators tried to get access to data on Roche's anti-influenza medication, Tamiflu, which would require a whole blogpost by itself. When I started to look into the Tamiflu story, I realised I was wrong about that. It's actually going to require 2 blogposts to tell the story. This is the first of those posts. In this post, I shall tell the story of the utterly surreal interactions between the Cochrane investigators and Roche. I shall save the question of what the evidence actually tells us about the efficacy of Tamiflu for another day.
The Tamiflu story started a while back, and indeed I blogged about it in 2009. Briefly, the problem arose when the Cochrane Collaboration (a normally splendid organisation that does a great job of looking carefully at the evidence for medical interventions to find out whether they really work) were preparing a systematic review to determine whether Tamiflu was an effective treatment for flu. It turned out that some important clinical trials of Tamiflu had not been published, and the story concerns the Cochrane investigators' attempts to obtain the data from Roche.
As I wrote in my previous blog, it seems to me that both the Cochrane investigators and Roche behaved very badly. That situation has continued, and I have to say it fills me with despair that both sides have taken the thoroughly unreasonable positions that both of them have. In fact it probably fills me with even more despair that I even need to write the phrase “both sides”: surely scientists should all be on the same side? Science is not a football match.
At one stage, Roche offered to make the reports available if the Cochrane group would sign a confidentiality agreement. They refused. I don't see the need for a confidentiality agreement, so I think Roche behaved badly in demanding one, but I equally don't see the justification for refusing. If the Cochrane investigators had been interested in scientific truth, rather than scoring points over the other side, then I can't see why they couldn't have just analysed the data under conditions of confidentiality.
Some commentators, such as Goldacre, have defended the Cochrane reviewers' refusal to sign the confidentiality agreement. On page 83 of Bad Pharma, Goldacre writes of the confidentiality agreement: “This was an impossibility for any serious scientist”. I agree it's not desirable, but impossible? Surely not. In fact the Cochrane handbook acknowledges that there are times when it will be necessary to sign confidentiality agreements before receiving data, so it seems bizarre for Cochrane reviewers to claim that it's impossible.
Anyway, in the end, I'm not sure what ever happened about the confidentiality agreement, but Roche did put the clinical study reports on a website so that they would be available to the Cochrane reviewers. If you're already familiar with the Tamiflu story, this might surprise you. Much of the popular narrative gives out the story that the reports have been kept a closely guarded secret. For example, on page 88 of Bad Pharma, Goldacre says “Roche had promised: 'full study reports will also be made available on a password-protected site...' . This never happened.”
The British Medical Journal have also perpetuated that myth: their website states “In Dec 2009, Roche publicly promised independent scientists access to "full study reports" for selected Tamiflu trials, but to date the company has not made even one full report available.”
This is rather puzzling, because as far as I can tell, Roche have indeed made the reports available. I can think of two possible explanations for the discrepancy. One is that Roche have made it so difficult to access the reports that they might as well have not made them available. If you visit the website where Roche have posted the reports, you are first greeted with a security warning telling you that the site's security certificate has expired. This does not exactly inspire confidence.
If you continue past the security warning, you find that the reports are available, but they are only available if you have a password. You have to agree to various onerous conditions to apply for a password, one of which is that you don't work for a commercial organisation. Even then, that doesn't seem to be enough. One of the people I follow on Twitter, Dr Dave Briggs from the University of Manchester, which I'm fairly sure was not a commercial organisation last time I checked, attempted to get a password to download the reports, but was unable to do so. Roche didn't refuse as such, but the mechanism to get a password simply didn't work, and Roche made no attempt to help him, despite repeated requests.
So frankly, any kudos Roche might have deserved from making the reports available is comprehensively nullified by the thoroughly unhelpful way in which they have done it.
But I don't think the difficult access to the reports is the reason why the claim that the reports are not available is so widely made, because if you read the 2012 Cochrane review carefully, it appears that they did have access to the study reports.
I think the reason why the popular narrative seems discordant with what has actually happened is that those claiming that Roche did not make the full reports available are not being very clear about what they mean by “full reports”. A few words of explanation are probably necessary here for those who are not familiar with how clinical study reports are usually written.
Data from clinical studies may be presented in various forms. They may be published as papers in peer reviewed journals, or they may be presented in clinical study reports. Clinical study reports have a synopsis, which tells you the main things you want to know about a study, but may not necessarily have enough detail to let you do a proper meta-analysis. The synopses for the Tamiflu reports are available online.
The synopses, which are typically a few pages long, are just one small part of a full study report. This would typically extend to hundreds of pages, and would include very detailed descriptions of the methods and results of the study, with many data tables. Those reports would contain far more detailed information than you would ever hope to find in a paper published in a journal, and would almost certainly give you more than enough detail to do any reasonable meta-analysis.
In addition to the main body of the study report, there are also appendices. They contain supplementary information, such as a copy of the protocol, details of the ethics committee that approved the study, certificates of analysis for drug batches, and so on. On the whole, they don't add much to the understanding of the study: they contain information that regulators need so that they can check regulations have been followed. But they are not necessary to be able to interpret the results of the study. The main body of a study report is intended to be readable as a stand-alone document.
So as far as I can tell, the complaint that Roche have not made “full” study reports available arises because Roche have made available only the main body of the reports, and not the appendices. To my mind, claiming the full reports are not available is a little misleading, as the main body of the report really should tell you everything you need to know, unless you are are regulator.
In any case, given that a clinical study report contains vastly more data than a published paper, and that most systematic reviews are done perfectly happily using nothing more than published papers, I find it utterly baffling to find it suggested that the Cochrane reviewers could possibly be missing any of the information that they need.
One of the Cochrane reviewers, Tom Jefferson, did attempt to explain this on Ben Goldacre's blog in a discussion of Tamiflu data, saying “Missing from the site are the remaining modules including protocol, certificates of analyses, SAP and other documents vital for the understanding of each trial.” However, he did not explain why they were vital for the understanding of each trial, despite my asking him that specific question. I don't think it would be too hard to explain what specific piece of information was missing, if any important information was missing, would it? A protocol is of course an important document, but any half-way competently written clinical study report would reproduce all relevant information from the protocol, so it should not be necessary to read the protocol as well to understand how the trial was conducted.
In fact I asked the same question on a blog from another of the Cochrane reviewers, Carl Heneghan, but he also failed to answer.
So we are left knowing that the Cochrane investigators believe that important information is missing (which seems unlikely if they have access to the main body of the study reports), but they have never specified what information is missing. Do they not know the dosing schedule? Do they not know how the randomisation list was generated? Do they not know how many patients were recruited but then excluded from the analysis? All those things would typically be included in the main body of a clinical study report, but if some of them are missing, it should not be too hard to specify which.
So what does the latest Cochrane review itself tell us?
Sadly, it doesn't shed much light on this strange state of affairs. The authors describe (see page 11 of the pdf version of the review) a rather complex process for assessing the completeness of information. It's quite confusingly written, but as far as I can tell, the authors required not only that all study reports needed to have complete information on a number of items, but that it also had to be possible to verify the information in the clinical study report from an external source (eg a regulatory summary report). If that was not possible, then the authors simply excluded the study from their analysis.
To my knowledge, that is unique in any systematic review.
It's not clear where this method came from. I have not been able to find it described in the Cochrane handbook as a recommended method. Given that most Cochrane reviews are based purely on published papers, to require a complete regulatory report together with confirmation of the information from external sources before studies can be included in a review would probably mean that there would not be a single Cochrane review including any studies at all. It seems a ridiculously high bar to set. It would be interesting to know if the authors of this Cochrane review are of the opinion that every other Cochrane review that has ever been produced is untrustworthy. That would be the only logical conclusion of their method.
The authors of the Cochrane review do not report which part of this process led to exclusions of papers. They merely tell us that there was insufficient information to include the studies. So we are still none the wiser about what information is missing.
So let's have a brief recap of what's going on here.
Much fuss has been made, by folk such as the British Medical Journal, Ben Goldacre, and the Cochrane Collaboration, about lack of access to Tamiflu data. It is strange to single out Tamiflu. Lack of access to data is a problem in clinical research in general, as I have blogged about more than once before.
But the data on Tamiflu are far more available than on many drugs. While it's disappointing that many of the Tamiflu trials remain unpublished in peer-reviewed journals, that situation is certainly not unique to Tamiflu. It's important to note that all Roche's trials on Tamiflu have been made available on their website in summary form to anyone, and their clinical study reports have also been made available to the Cochrane investigators. There are probably not many drugs which have been disclosed to systematic reviewers to a greater extent than Tamiflu has.
So it's all very strange that the Cochrane investigators are claiming that they do not have sufficient access to Tamiflu data and that Tamiflu is being used as a poster child for lack of transparency from the pharmaceutical industry.
While any attempt by me to explain the reasons why they are doing that would be speculation, it is interesting to note one or two things in passing.
The lead investigator of the Cochrane review, Tom Jefferson, has been found collaborating with anti-vaccine conspiracy theorists in a rather disturbing way. Is it possible that there is some desire to denigrate the pharmaceutical industry, rather than to engage in a scholarly examination of the evidence? I don't think an anti-industry bias within the Cochrane Collaboration can be ruled out, given that another recent Cochrane review drew some very strange anti-industry conclusions that seemed to have little relationship to the results they found.
And what of others who have picked up the story? Well, Ben Goldacre has a book to sell. Telling stories of how the evil pharmaceutical industry hide all their data is probably good for business. The BMJ also tend to take a rather anti-industry stance on many things (although they deny this). Is the Tamiflu story just a convenient peg on which to hang a few general rants about evil pharmaceutical companies?
I honestly don't know what's going on here. I do realise that the previous couple of paragraphs read a bit like some of the conspiracy theory rants you read from anti-vaccinationists and homeopaths, albeit in reverse. Maybe I'm way off base with all of this.
But still, you have to admit that the Tamiflu story is all rather odd, don't you?
Thanks for writing this post. This is the first counterpoint I've read suggesting that there is another, potentially even non-evil, side to the story. Is it possible I've been misled by the (normally eminently reliable and trustworthy) Cochrane group, Ben Goldacre, and the BMJ on this issue? I would like to hope not, but your post gives me food for thought.
I sincerely hope Ben Goldacre, Fiona Godlee, or Tom Jefferson takes the time to respond to this well-reasoned, honest, re-assessment of the oseltamivir issue.
Thanks for putting up this thought-provoking post Adam. If full avialability of all background documents is required to undertake a valid meta-analysis it would be interesting to know what proportion of Cochrane Collaboration(CC) meta-analyses would be valid. My guess is not more than 1%. Have you tried to contact the CC for a position?
Stephen
My declaration of interest is here:
http://www.senns.demon.co.uk/Declaration_Interest.htm
Thank you for the reminder, Stephen. I hadn't contacted the CC, but I have now. One good thing about the Cochrane reviews is that they provide a handy little feedback form with each review. I'll report back with developments in due course.
Happily, it does appear that they take such feedback seriously. I provided feedback in the same manner on another recent review, which I blogged about here. They have told me they will be incorporating the concerns I raised in my blog in their next update in April.
Thanks, Adam. I have a less rosy view of Cochrane. Some years ago I contacted them about three or four meta-analyses in which the same data had been counted twice. It took them an age to react and after two years I gave up checking to see whether these meta-analyses had been corrected. See http://www.biomedcentral.com/1471-2288/9/10
CC made a decision right at the outset to make it possible for persons with a very limited grasp of statistics to carry out meta-analyses. This had the positive aspect of providing a much bigger work- force but errors such as entering the control arm twice in a three-armed trial (I am not joking) are the price you pay. Basically I think they are a force for good but their pompous self-rightousness prevents them being even better.
Stephen
My declaration of interest is here http://www.senns.demon.co.uk/Declaration_Interest.htm
The Cochrane authors have now replied to this post. Rather than leave a comment here, they have emailed me a document containing a detailed reply, which they have asked me to post here.
You can read their response here. It's quite a long document, and I must confess I haven't had the time to read it yet, but I shall do so when I have a moment, after which I'll post a response if there's anything there that needs one.
OK, I've read the Cochrane authors' response. It's quite long, so I won't attempt to summarise all of it here. Much of it is taken up explaining what a study report is, which is something I thought I'd already done myself in my original post.
The response does, however, answer one of the questions I asked, namely what specific data are missing. The response discusses the difference between the ITT population (intent to treat, which includes more or less all randomised patients) and the ITTI population (ITT influenza-infected, which is the subset of patients who turned out to have laboratory confirmed influenza infection). The latest response states that data on complications only for the ITTI population was available in Roche's study reports, and not for the ITT population.
It's odd that this wasn't mentioned in the original Cochrane review.
It also doesn't really explain why they didn't go ahead and analyse the data based on the ITTI population. They don't seem to be able to explain what information was missing that would prevent them running that particular analysis.
But perhaps the most disappointing thing of all is that they have made no attempt to address the question of why they are expecting far higher standards of data disclosure than would normally be required in a Cochrane review. I still think that if these standards were applied consistently to all Cochrane reviews, then there would be very few Cochrane reviews indeed that analyse any data at all.
Adam, I have to disagree with you here. I think that the ITTI analysis is unimportant. It is ITT that matters.
Stephen
My declaration of interest is here
http://www.senns.demon.co.uk/Declaration_Interest.htm
Oh, I agree that the ITT analysis would be better. But the prespecified analysis plan said that they would analyse both, so it seems a little odd not to do as much of it as possible, no?
I suppose it is reasonable to perform the analysis if it was pre-specified. Was it the case that the FDA required Roche to have a significant results in both? However, that does not make the ITTI analysis sensible. The ITTI population is one, in practice, you cannot treat or, at least, you can't treat it without treating the larger ITT population.
My declaration of interest is here
http://www.senns.demon.co.uk/Declaration_Interest.htm
Adam, I don't find this response satisfactory. I also find the position of Dr Godlee (editor of the BMJ) who has recently been sending out messages encouraging readers to report difficulties in accessing data very strange. I did a quick pubmed search for meta-analyses of trials in the BMJ during 2012 and turned up about 70 plausible hits. (Probably some of these are commentaries and methodological pieces but that still will leave a fair number.) Can we be told how many of these meta-analyses are based on summaries and how many based on original data? I suspect that many will have been based on summaries. If only original data are good enough, then what is the BMJ doing publishing inadequate studies? If summaries are the norm then should the publicity surrounding the Tamiflu case not explain a) that exceptionally original data are being requested and b) why an exception is being made here?
As I explained in Senn S. Hans van Houwelingen and the Art of Summing up. Biometrical Journal 2010; 52: 1-10, meta-analysis based on sufficient statistics can, given suitable care, reproduce almost exactly what one would get from the original data. In fact in a section entitled 'Four misconceptions of meta-analysis' begins 'The first misconception of MA is that an analysis of original data is more precise than an analysis based on summary statistics.'
Of course it may be that meta-regression is what they wish to carry out, in which case one would need patient level data. Or there may be some other reason that is not clear but certainly if only original data are good enough then we can consign most Cochrane reviews, including many published over the years in the BMJ, to the dustbin.
One is left with the distinct impression that this is as much about politics as science.
Stephen
My declaration of interest is here
http://www.senns.demon.co.uk/Declaration_Interest.htm
I have now had time to look at the paper cited(1) in the Cochrane Tamiflu Team's reply to you . As a general statement of what one would like for the future of meta-analysis I have no quibble with this but it is a principle of fairness before the law that one does not selectively apply standards on the basis of personal prejudice. To give an analogy, even if every woman driver ticketed by a traffic warden were breaking the law this does not excuse his only ticketing women drivers as a matter of principle is illegal and as a matter of policy it is hardly likely to improve public parking.
In the wake of various recent scandals of published evidence, (the Duke University case is an example that springs to mind(2)) I cannot see the justification in exempting academics from a policy of requiring original data, protocols, programs etc.
I pulled out the following report(3) from the BMJ 2012 meta-analysis crop. A sample of one is not much to go on (even if randomly chosen) but it does show that what the BMJ is prepared to publish is light years away from the Doshi et al manifesto. I quote
"...independent reviewers extracted means (final scores or change score), standard deviations, and sample sizes from studies using a standardised data extraction form. When there was insufficient information in trial reports, we contacted authors or estimated data using methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions.17 Briefly, when the mean was not reported we used the median; when standard deviations could not be estimated, we adopted the standard deviation from the most similar study."(3)
See (4) for a discussion of the problems with data imputation.
I am not claiming that this(3) is a bad study, in fact I think it is above average, I am just pointing out that if the BMJ will publish this but join the Tamiflu crusade they are being very inconsistent.
Personally what I would like to see for the future is a publication plan to cover on-the-web archiving as a mandatory part of the regulatory submission for industry trials . However I also want to see a corresponding undertaking for independent trials, even though this will be more difficult to police.
At the moment what we seem to have is the following strange 'policy'. Because the FDA carries out a thorough check of regulatory trials in addition to the cursory one carried out by journals (should the trial be published) we must have the data that the FDA sees. Because original data for non-industry trials are neither checked by the FDA nor the journals that publish them, we don't need them.
My declaration of interest is here
http://www.senns.demon.co.uk/Declaration_Interest.htm
References
1. Doshi, P., M. Jones, et al. (2012). "Rethinking credible evidence synthesis." BMJ 344: d7898.
2. Baggerly, K. and K. R. Coombes (2009). "Deriving chemosensitivity from cell lines forensic bioinformatics and reproducible research in high-throughput biology." The Annals of Applied Statistics 3(4): 1309-1334.
3. Pinto, R. Z., C. G. Maher, et al. (2012). "Drugs for relief of pain in patients with sciatica: systematic review and meta-analysis." BMJ 344: e497.
4. Senn, S. J. (2009). "Overstating the evidence: double counting in meta-analysis and related problems." BMC Med Res Methodol 9: 10.
So, I looked at the website. 'The synopses for the Tamiflu reports are available online.'. I wanted PK/PD. Two studies found, neither of them available when you try the link
Oh, that's interesting. Which studies specifically were those?
[...] Tamiflu story is one I’ve written about more than once before. It’s a strange story. It’s as much about politics as about science. [...]