I would like to preface this post by saying that nobody really knows how and why antidepressants work (poorly) and if somebody tells you they do they are either lying to you or lying to themselves.

Off the top of my head, there’s the monoamine theory, the serotonin theory, the macrophage theory, the HPA axis theory, the neurogenesis theory, the REM sleep theory, the response-styles theory, other cognitive theories, interpersonal theories… etc. The list will continue to grow as long as people continue to get more recognition for publishing/supporting new theories than disproving old ones.

Each one of these theories has evidence that antidepressants work by doing something that the theory predicts. So who is right? They’re probably all right – and all wrong – to some degree. Science is like a bunch of people standing in the middle of an indoor football stadium with the lights turned off. They all get a flashlight to shine in a different direction, so one guy sees grass, one guy sees chairs, another guy sees goal posts, and someone else sees the jumbotron. Unless they knew it was an football stadium to begin with, and what a a football stadium is, they’d have no fucking clue what they were looking at. And if they were academics then they’d also argue about it like children.

With that in mind…

Here’s a study I re-read last week while writing my last post and I cannot believe I missed this the first time I read this article years ago.

They wanted to know if SSRI’s were more effective for “anxious depression” (depression with a 6-item HAM-D anxiety-somatization score > 6, not necessarily a full-on anxiety disorder) compared to Wellbutrin. Why? Because most psychiatrists believes that “low serotonin” (which SSRIs increase) is “involved” in anxiety whereas dopamine and norepinephrine (which Wellbutrin increases) are “involved” in things like mood, motivation, and energy, but don’t help with, and possible worsen, anxiety.

So what did they do? They did a pooled-analysis, which means they took data from 10 old trials and analyzed it as if it were one big trial. Why? So that a very small effect would be statistically significant. If N goes up, p goes down. It isn’t more complicated than that. Does generalizability increase? Nope. All the studies use the same narrow population. In fact, they went out of their way to exclude two trials that had “special populations.”

And what did these masters of psychopharmacology find? [Brackets include comments added by blogger].

“Among patients with high levels of anxiety (anxious depression) (N = 1275), response rates were greater following treatment with an SSRI than with bupropion according to the HAM-D-17 [depression] (65.4% vs. 59.4%, p = .03) and the HAM-A [anxiety] (61.5% vs. 54.5%, p = .03). [6% different in depression response rate? Most drugs rarely do 10-15% better than placebo. So what does a 6% difference mean? Nothing. You would need to treat 17 people with anxious depression with an SSRI over Wellbutrin to get one additional response…] SSRI treatment also favored bupropion in producing a greater reduction in HAM-D-17 mean ± SD scores (–14.1 ± 7.6 vs. –13.2 ± 7.9, p = .03) [Hold up, that’s a fucking 1 point change difference on a scale that ranges from 0-52, where < 7 is not depressed. One point doesn’t mean shit! Each item gets you up to 2 or 4 points, so 1 point better might not even mean the absence of a single symptom, let alone the bulk of them. But p was < .05 so I guess it must be important. By the way, this minuscule difference would not have been statistically significant if they hadn’t pooled data from 10 trials. Even if they used data from 7 trials it might not have been. Certainly not had any of the individual studies tried to do this analysis] and a trend toward statistical significance for a greater reduction in HAM-A mean ± SD scores (–10.5 ± 7.4 vs. –9.6 ± 7.6, p = .05). [So the thing that the serotonin/anxiety theory predicted would get better wasn’t actually better? Great, I can’t wait to hear how you revise the theory based on this new finding…] There was no statistically significant difference in remission rates between SSRI- and bupropion-treated patients with high levels of anxiety as defined using the HAM-D-17 (50.0% vs. 46.8%, p = .2) or HAM-A (50.4% vs. 46.9%, p = .2). [Remission means people actually getting better, not just having an arbitrary score on a scale decrease by 1 or 3 or 9 or whatever points that might not mean anything in the real world.]

“There was no statistically significant difference in any of these 6 outcome measures between bupropion and the SSRIs among patients with moderate/low levels of anxiety (see Figures 1 and 2). The mean ± SD change in HAM-D-17 scores among patients with moderate/low levels of anxiety was –11.7 ± 7.2 versus –11.2 ± 7.1 for bupropion and the SSRIs, respectively (p = .2). The mean ± SD change in HAM-A scores among patients with moderate/low levels of anxiety was –7.5 ± 6.4 versus –7.4 ± 6.1 for bupropion and the SSRIs, respectively (p = .7). HAM-D-17–based remission rates among patients with moderate/low levels of anxiety treated with either bupropion or an SSRI were 55.7% and 53.2%, respectively (p = .4). HAM-A–based remission rates among  patients with moderate/low levels of anxiety treated with either bupropion or an SSRI were 54.6% and 52.1%, respectively (p = .5). HAM-D-17–based response rates among patients with moderate/low levels of anxiety treated with either bupropion or an SSRI were 65.2% and 61.5%, respectively (p = .2). Finally, HAM-A–based response rates among patients with moderate/low levels of anxiety treated with either bupropion or an SSRI were 59.3% and 60.6%, respectively (p = .7).”

So what did they find? Nothing. SSRIs and Wellbutrin are equally good in nonanxious depression. We already knew that. Are SSRIs better in anxious depression? Not really. The findings were minuscule. There’s no role for clinical application if you expect findings that small. So what did the authors conclude?

“Pooling data from 10 double-blind, randomized clinical trials revealed a greater resolution of depressive as well as anxiety symptoms following the treatment of anxious MDD with the SSRIs than with bupropion.”

Based on what, your manufactured p value?

“Although the difference favoring the SSRIs was statistically significant, it was also quite small (6%), of uncertain clinical significance”

Actually it’s of no clinical significance.

“The present work is in contrast to a large body of literature that suggests no difference in efficacy among the major antidepressant classes when treating anxious depression…In light of the magnitude of the difference in response rates estimated by our work, a mere 6%, the discrepancy between our findings and previous studies may be due to the limited statistical power of previous studies to detect such a treatment difference.”

You think? A difference of 1% would be statistically significant with 10,000 participants. Does that make it useful? No. But the theory predicts a difference, so instead of throwing away the theory in light of all the negative trials they just increased the number of participants so even a trivial difference would be statistically significant. Great work, Doctors!

“In conclusion, there appears to be a small advantage for the SSRIs compared to bupropion in the treatment of anxious depression (6% difference in response rates). Using the NNT statistic as one indicator of clinical significance, we found that nearly 17 patients would need to be treated with an SSRI in order to obtain 1 additional responder. This difference falls well above the limit of NNT = 10, which was suggested by the United Kingdom’s National Institute of Clinical Excellence. Nevertheless, the present work is of theoretical interest because it provides preliminary evidence suggesting a central role for serotonin in the regulation of symptoms of negative affect such as anxiety.”

There’s no such thing as a true theory, but there are useful theories. This theory isn’t useful, but they kept it anyway. Either I’m missing something here or Kuhn was right. The data was ignored, the theory was retained, and 9 years later the paradigm hasn’t changed.