This is from the always excellent CL Psych. As usual I have reproduced the whole of his article – it’s all here and a real eye opener.
If you need to catch up on Study 329, then read this. Otherwise just carry on:
A bombshell has just appeared [April 2008] in the International Journal of Risk & Safety in Medicine. The subject of the paper is Paxil study 329, which examined the effects of the antidepressant paroxetine in adolescents. The study findings were published in the Journal of the American Academy of Child and Adolescent Psychiatry in 2001. These new findings show that I was wrong about Paxil Study 329. You know, the one that I said overstated the efficacy of Paxil and understated its risks. The one that I claimed was ghostwritten. Turns out that due to legal action, several documents were made available that shed more light on the study. The authors (Jureidini, McHenry, and Mansfield) of the new investigation have a few enlightening points. Let’s look at the claims and you can then see how wrong I was, for which I sincerely apologize.
The story is actually worse than I had imagined.
Here’s what I said then:
Article [quote from the study publication]: Paroxetine is generally well-tolerated and effective for major depression in adolescents (p. 762).
Data on effectiveness: On the primary outcome variables (Hamilton Rating Scale for Depression [HAM-D] mean change and HAM-D final score < 8 and/or improved by 50% or more), paroxetine was not statistically superior to placebo. On four of eight measures, paroxetine was superior to placebo. Note, however, that its superiority was always by a small to moderate (at best) margin. On the whole, the most accurate take is that paroxetine was either no better or slightly better than a placebo.
I went on to bemoan how the authors took differences either based on arbitrary cutoff scores or from measures that assessed something other than depression to make illegitimate claims that paroxetine was effective. Based upon newly available data from the study, here’s what happened.
- The protocol for the study (i.e., the document laying out what was going to happen in the study) called for eight outcome measurements. To quote Jureidini et al: “There was no significant difference between the paroxetine and placebo groups on any of the eight pre-specified outcome measures.” So I was wrong. Paxil was not better on 4 of 8 measures — it was better on ZERO of eight measures. My sincerest apologies.
- Another quote from Jureidini and friends: “Overall four of the eight negative outcome measures specified in the protocol were replaced with four positive ones, many other negative measures having been tested and rejected along the way.“
Let’s break this thing down for a minute. The authors planned to look eight different ways for Paxil to beat placebo. They went zero for eight. So, rather than declaring defeat, the authors then went digging to find some way in which Paxil was better than a placebo. Devising various cutoff scores on various measures on which victory could be declared, as well as examining individual items from various measures rather than entire rating scales, the authors were able to grasp and pull out a couple of small victories. In the published version of the paper, there is no hint that such data dredging occurred. Change the endpoints until you find one that works out, then declare victory.
How About Safety?
I was incensed about the coverage of safety, particularly the magical writing that stated that a placebo can make you suicidal, but Paxil could not. I wrote:
It gets even more bizarre. Remember those 10 people who had serious adverse psychiatric events while taking paroxetine? Well, the researchers concluded that none of the adverse psychiatric events were caused by paroxetine. Interestingly, the one person who became “labile” [i.e., suicidal] on placebo – that event was attributed to placebo. In this magical study, a drug cannot make you suicidal but a placebo can. In a later document, Keller and colleagues said that “acute psychosocial stressors, medication noncompliance, and/or untreated comorbid disorders were judged by the investigators to account for the adverse effects in all 10 patients.” This sounds to me as if the investigators had concluded beforehand that paroxetine is incapable of making participants worse and they just had to drum up some other explanation as to why these serious events were occurring.
Turns out I missed a couple things. Based on looking at an internal document and doing some calculations, Jureidini et al. found that serious adverse events were significantly more likely to occur in patients taking paroxetine (12%) vs. placebo (2%). Likewise, adverse events requiring hospitalization were significantly disadvantageous to paroxetine (6.5% vs. 0%). Severe nervous system side effects — same story (18% vs. 4.6%). The authors of Study 329 did not conduct analyses to see whether the aforementioned side effects occurred more commonly on drug vs. placebo.
Funny how they had time to dredge through every conceivable efficacy outcome but couldn’t see whether the difference in severe adverse events was statistically significant.
One quote from the discussion section of the paper sums it all up:
There was no significant efficacy difference between paroxetine and placebo on the two primary outcomes or six secondary outcomes in the original protocol. At least 19 additional outcomes were tested. Study 329 was positive on 4 of 27 known outcomes (15%). There was a significantly higher rate of SAEs with paroxetine than with placebo. Consequently, study 329 was negative for efficacy and positive forharm.
But the authors concluded infamously that “Paroxetine is generally well-tolerated and effective for major depression in adolescents.”
Enter Ghostwriters. Documentary evidence as shown on indicated that the first draft of the study was ghostwritten. This leaves two roles for the so-called academic authors of this paper:
- They were willing co-conspirators who committed scientific fraud.
- They were dupes, who dishonestly represented that they had a major role in the analysis of data and writing of the study, when in fact GSK operatives were working behind the scenes to manufacture these dubious results.
Remember, this study was published in 2001, and there has still been no apology for the fictional portrayal of its results, wherein a drug that was ineffective and unsafe was portrayed as safe and effective. Physicians who saw the authorship line likely thought “Gee, this is a who’s who among academic child psychiatrists — I can trust that they provided some oversight to make sure GSK didn’t twist the results.” But they were wrong.
By the way, Martin Keller, the lead “independent academic” author of this tragedy of a study said, when asked about what it means to be a key opinion leader in psychiatry:
You’re respected for being an honorable person and therefore when you give an opinion about something, people tend to listen and say – These individuals gave their opinions; it’s worth considering.
So is completely misrepresenting the data from a study “honorable”? Is Keller’s opinion “worth considering?” As you know if you’ve read this blog for long, such behavior is, sadly, not a fluke occurrence. Many others who should be providing leadership are leading us on a race to the scientific and ethical bottom. What will Brown University, home of Keller, do? Universities don’t seem to care at all about scientific fraud, provided that the perpetrators of bad science are bringing home the bacon.
Not one of the “key opinion leaders” who signed on as an author to this study has said, “Yep, I screwed up. I didn’t see the data and I was a dupe.” Nobody. Sure, I don’t expect that every author of every publication can vouch for the data with 100% certainty. I understand that. But shouldn’t the lead author be taking some accountability?
This is a Fluke (?) Some may be saying: “But this is just a fluke occurrence.” Is it? I’ve seen much evidence that data are often selectively reported in a manner like this — looks like (sadly) it takes a lawsuit for anyone to get a whiff of the bastardization of $science that passes for research these days. If GSK had not been sued, nobody would have ever known that the published data from Study 329 were negative. A reasonably educated person could see that the writeup of the study was a real pimp job — lots of selling the product based on flimsy evidence, but nobody would have seen the extent of the fraud. Apparently lawyers need to police scientists because scientists are incapable of playing by some very basic rules of science.
See for Yourself. Documents upon which the latest Jureidini et al. paper are based can be found here. Happy digging.