Wednesday, February 27, 2008
The Huffington Post has several articles on why the press is getting it wrong about the results of the PLOS study discussing placebos and anti-depressants. One author, Gary Marcus writes about the terrible media coverage (oops!) and states,
Dashing by Google News before bed, I noticed that the usually careful Washington Post had ran a headline saying, "Only Severely Depressed Benefit From Antidepressants: Study", which I, can only imagine, will lead hundreds, if not thousands of people to get off their anti-depressants -- and probably for the wrong reasons. Time Magazine, meanwhile, has reported that "Antidepressants hardly help."
Both stories report on a study that was just published in the well-respected journal PLoS Medicine. But neither shows anything like what the headlines suggest.
Which -- and this is the part that gets me -- would be obvious not only to anyone who read the original study (which is available online for free), but also to anyone who even bothered to carefully read the news stories. Does the Washington Post article actually show that only the severely depressed benefit from antidepressants? Noooo. Does the Time Magazine article actually show that antidepressants hardly help? Noooo. Does the PLos Medicine study show either of these things? Noooo.
Let's consider first The Washington Post's headline -- "only the severely depressed benefit from antidepressants". Um, hello? Does the study really show that? Four paragraphs into the Time study, we find the truth (from a reporter who was far more careful than the person who crafted her headline), "The researchers behind this new paper did find that SSRI drugs [like Prozac and Zoloft - GFM] have a statistically significant impact for most groups of patients."
That's right, most groups of patients.
The effect is, to be sure small, for many people scarcely more than a placebo. But most people were in fact helped, at least a little. As a much more thoughtful blog on the Wall Street Journal put it, "for many patients, placebos work pretty well indeed."
The Time headline -- "antidepressants hardly help" -- is even more misleading than the Post's. It might be argued that antidepressants are only of mild help for people who are only mildly depressed (with less room for improvement). But the current study, because of its sheer scope. provides some of the strongest evidence to date that on average (individual mileage may vary), antidepressants are a great deal of help to those most severaly depressed --- and somewhere in between for those with intermediate levels of depression. All of which is pretty clear if you pause for a minute and look at this graph of the results.
The real story here is not that the antidepressants are ineffective, but that the magnitude of their effect is (roughly) proportional to the magnitude of the depression; if you're happy, there's no point in taking them, but the more depressed someone is, the more the medications may help. Isn't that true of most medicines? . . . .
Another commentator , Maia Szalavitz remarks,
A new meta-analysis of research on modern antidepressants -- some of it unpublished by the drug companies -- suggests that the drugs have little advantage over placebos.
Why then do so many people consider drugs like Prozac to be miracle drugs for depression -- many putting up with serious sexual side effects in order to take them? Are they simply being duped by a placebo effect or avoiding withdrawal symptoms? And how could drugs which are little different from placebo also produce suicidal or even homicidal thoughts in some patients?
The answer reveals a key flaw of randomized clinical trials and meta-analyses: when you are looking at aggregated data, huge individual differences can be washed out. For example, let's imagine a drug that causes people with one genetic variation to have a profound positive effect -- but causes those with another to get dramatically worse and has little effect on everyone else.
A clinical trial could easily find that this drug has no advantage over placebo, depending on the proportion of people with each gene in the study. Another study of the same drug might find it to be a blockbuster -- while another found it dangerous. Same drug, different populations. . . . .
This doesn't mean that we don't need randomized clinical trials or meta-analyses: RCT's are the gold standard of evidence-based medicine with good reason. But it does mean that until we can better understand how genetic variation affects drug response, we will continue to have these boring debates about Prozac: Angel or Devil.
Especially in a context where the media often refuse to explain how ideological biases affect people's positions on these medications and conveys the story as a clash between two conflicting views of the world. Both sides are right -- but only about the response of particular people to particular drugs. This is why it can be true both that 80% of depressed people can find a medication that works and that clinical trials don't find these drugs much better than placebo when looking at the general population of depressed people.