Using the Best Spin Possible

Over at the excellent Last Psychiatrist blog, you can see an example of a graph being used to paint the rosiest possible picture of the data in a study (Sachs et al, 2002) investigating the potential benefit of adding risperidone to a mood stabilizer in the treatment of patients experiencing a manic episode. He points out that on the graph purports to show a large difference in “remission” rates between patients receiving a combination of a mood stabilizer and risperidone (Risperdal) and patients receiving a combination of mood stabilizer plus placebo. Indeed, the “remission” rates are 57% and 33%, respectively according to the graph. A patient reportedly had remission of manic symptoms if he/she scored at less than 12 on the Young Mania Rating Scale (YMRS). Please refer to the Last Psychiatrist’s site for questions about whether this is a valid rating -- is there really something magic to a score of less than 12 on the YMRS? It’s a good read.

There’s even more to this study that needs discussion…

There is a moderate effect size difference favoring the combination of risperidone and mood stabilizer over mood stabilizer plus placebo if you look at the endpoint data, which are based on taking participants’ last score and including it in the analysis. For many conditions, this analysis makes some sense. However, we know that manic episodes tend to remit with time. You don’t hear of many manic episodes lasting for several months at a time. In this study, 49% of patients dropped the study from the placebo plus mood stabilizer group versus 35% from the risperidone plus mood stabilizer. This means that the data from the placebo + ms group are coming from an earlier point in time (because the data are taken from the last observation of the patient) than the risperidone + ms group, so the risperidone plus ms group has a better chance of change, due to nothing more than being in the study longer, as manic symptoms tend to decline with time alone. So part of the difference in the endpoint analysis is due to time – we don’t know how much, but I’d bet it is substantial.

So one has to look at the differences at each week, and you’ll find a small effect size favoring risperidone plus ms at week one, a moderate effect size difference at week two, and a small difference at week three. All of these differences are smaller than the endpoint analysis.

Let’s also weigh the risks. In table 5, it can easily be seen that risperidone plus ms related to twice the incidence of somnolence, three times the frequency of an extrapyramidal disorder, and six times the incidence of dizziness compared to the mood stabilizer plus ms. If that is not enough fun for you, consider that the average weight gain for risperidone plus ms was 5.3 pounds versus 1.1 pounds for the placebo plus ms group. Oh, and the effect size for weight gain was higher than the effect size for treatment at any of the weeks. I’m willing to bet that the drug company program from which the infamous graph showing “remission” rates came from did not mention that the between group differences in weight were greater than the differences in manic symptoms.

Of course, the study discussed in this post is old news. There is only so much money to be made by using risperidone (or any other antipsychotic) for three weeks at a time. The real cash is to be made in maintaining patients on medication for as long as possible, which is of course why there is an increasing amount of attention being paid to studying the long-term use of atypical antipsychotics for mania. That, however, is a topic for another day.