When first we practise to deceive!
Sir Walter Scott
In the last post, I was talking about the need for the raw data from clinical trials to be posted after discovering that the STUDY 329 was finally available. The format is tough, an image based pdf file. The only way to get the numbers out where they can be analyzed that I know is to copy them one by one into a spreadsheet or database program. Thus, the absence of posts here while I took a day+ to do that [only for the Paxil group so far]. This is a plot of the raw data for the HAM-D [primary outcome measure]. I expect I’ll do the Placebo and Imipramine groups sooner or later, but this is enough to take a stab at a point [and give my bifocals a rest]:
Each of those lines represents one of 91 depressed teenagers on Paxil – some dropouts, some failures, some successes. Obviously, it’s meaningless without the placebo group to compare it to, but just looking at the raw results, the variability is striking. And each nodal point on the graph is the sum of the responses to seventeen questions. There’s an ocean of variables in that graph.
There are a number of mathematical maneuvers we can make with these garbled lines once we see them. We can express them as change from the baseline or % change from the baseline like Keller et al did. We can reduce the whole thing down to a bland arithmetic mean,
[red line = arithmetic mean]
or the least squares mean. We can quantify the variability as a standard deviation, or a standard error of the mean. We can define ‘cut-offs’ like,
We can do the same things with the Placebo response, and then do all manner of statistical tests to compare Paxil to Placebo.
All of these mathematical gyrations are operations specifically designed to reduce confusion and put the spaghetti of raw data into a format that allows comparisons that ultimately address the efficacy of Paxil in teens. But these same machinations that are supposed to bring clarity to the question can [and have been] used to pervert and hide those very meanings, to disembody and depersonalize these teenagers, and to be used in the service of something other than figuring out if Paxil helped them – to hide rather than explicate the meaning. We know that’s what happened with Study 329. That tangle up there was turned into something using the tools of legitimate science, but that something was not the truth.
So the reason I spent more time than I wanted to transcribing the raw data into something that could be analyzed and displayed [and will likely waste a few more days doing the same thing with more of the data] is that I want to make the loud case that if this information had been available eleven years ago, all of this wouldn’t have happened. GSK, Sally Laden, and the guest authors couldn’t have published a study that said, "Paroxetine is generally well tolerated and effective for major depression in adolescents." The only reason they could say that is because this raw data wasn’t available to the curious or the critics. The authors could twist it and turn it into something it wasn’t, because it couldn’t be checked.