Antidepressants Are Still Neck and Neck

It is a perennial question: Are some antidepressants better than others? Many clinicians say yes, some academics agree; many academics say no, some clinicians agree. Peter Kramer gives us a thoughtful reading of a new hot-off-the press Lancet meta-analysis in support, which would seem to have lexapro and zoloft winning the race of efficacy and tolerability, as many clinicians believe. But, as Kramer also says, it is hard to make such judgments cleanly. Here is another reason to be cautious, based on an inherent problem in this kind of research.

I still see everything as neck-and-neck, based on two considerations:

1. Of 117 studies included in the meta-analysis, only 15 were unpublished studies obtained from the pharmaceutical industry. Another study found that about half of the antidepressant clinical trials are negative but unpublished, while about 90% of the published literature is positive. Massive unpublished negative study bias goes on; extrapolating from that study, we would expect that this study, which only consists of about 10% unpublished (and presumably mostly negative) data, overestimates benefits with study drugs versus other drugs by some percentage; perhaps there are up to 30-50% more studies out there not included in this review.

2. The benefits seen were small in effect, with about 30 to 50% relative benefits of one drug over another, where those benefits were seen. This is generally seen as a mild clinical effect; for comparison, smoking causes lung cancer by a relative risk of 1000%; that is considered a very large clinical effect. A doubling of benefit would be a 100% relative difference, tripling would be 200% (no difference is 0%). So a 30-50% difference, if real, is small.

If observation 2 is corrected for observation 1 above, that is, if we correct a small clinical benefit for an at least small probable overestimation of benefit (due to noninclusion of negative unpublished studies), then things are back to neck and neck again.

Meta-analysis done on an industrial scale often obscures instead of clarifying. The methods are platinum, hence a publication in Lancet, but the results are dross. The process has been likened to statistical alchemy by some researchers who have warned against this, which is why it is generally accepted that a single randomized study that is large is much more valid than adding up multiple small studies in a meta-analysis. This is where the utility of another study comes in: In fact, a large randomized study has been conducted comparing different antidepressants - the NIMH sponsored STAR-D study, and the main finding of that huge study (with over 3000 patients at its start), is that antidepressants seem quite similar in efficacy, including the ones picked out in the meta-analysis. Further, and perhaps most humblingly, even if antidepressants worked in the short term (2 months, which is also what the meta-analysis assessed), one-half of patients who stayed on them relapsed into depression within one year. At the one year outcome, only about 25% of patients actually had remained well on and tolerated an antidepressant, much below the levels most clinicians seem to feel occurs in their clinical experience. The best horcerace study on antidepressants has been done, and it is unlikely to be repeated: we should meditate on its sobering results.

We could all wish that clinicians' beliefs about antidepressants were true, or even half true. And perhaps they are the latter, for these agents surely have some uses in some settings; they are just not the dream drugs they seemed to be. In the end, perhaps Edgar Allan Poe best captured the clinician's dilemma today: He can believe none of what he hears, and only half of what he sees.