Response to Letters

To the Editor: We thank Dr. Smith and Dr. Gould et al. for their interest in our study. The importance of conducting a power calculation based on suicide attempts instead of suicidal ideation with plans is a point well taken. Because the power calculator originally used for the article appears to have been removed from the Johns Hopkins web site, we identified a different power calculator and double-checked it with our in-house power calculation script. With N=94 and 50% dropout, and an attempt rate of 13% for lithium, the minimum hazard ratio for valproate detectable with 80% power is around 3.2. Based on these same assumptions for suicide events, the hazard ratio would be 2.2. In other words, based on these new calculations, it appears that the study was better powered than originally stated in the article. Note that these calculations 1) are based on the proportional hazards regression analysis and 2) assume exponential times to event or attempt, neither of which applies to this data set. We are currently working on a power calculator for the log-rank test based on resampling.

We concur with Dr. Smith that even a 20% effect size would be of great clinical utility. This would be especially true in the context of a randomized controlled trial, in which one can obviate problems such as confounding by indication (doctors shying away from giving lithium to those patients at risk for overdose), sample bias (many lithium clinic data come from samples with a mean age over 40, possibly excluding the high-risk patients who may have already died from suicide), and key clinical variables (routine monitoring of blood levels maximizes both patient adherence to treatment and the likelihood of therapeutic levels of medication). It is our opinion that subdividing the hazard curves into smaller intervals would be a stretch of the data, especially given that the curves cross each other more than once, casting doubt that observed variations in the position of the curves with regard to each other are caused by the pharmacologic properties of the drugs.