Twelve-Month Outcomes for Remitters Following Electroconvulsive Therapy for Depression.

To the Editor: We thank Dr Andrade1 for his interest in our work.2 What he describes as an observational study was in fact an analysis of 12-month relapse from a randomized trial of unilateral vs bilateral electroconvulsive therapy (ECT) for depression, ie, a planned secondary outcome of an interventional study. The hypothesis underlying the trial was noninferiority of right unilateral versus bitemporal ECT.3 Randomization and blinding were preserved throughout the follow-up period. In this secondary analysis, we examined long-term outcomes because it could be argued that although acute outcomes were similar in the two groups, remission may be more transient in the unilateral group and the advantage of bitemporal ECT would only become apparent later. As such, electrode placement (ie, treatment group) was the main covariate of interest and cannot be dropped from the regression model simply because it turned out not to be “significant.” Dr Andrade seems to suggest that including a priori known prognostic factors in a regression model is overfitting. On the contrary, this is standard practice. Inclusion of these covariates reduces the amount of residual variance in the model. Choice of covariates is never an easy task, but it is particularly challenging in a situation like this one, in which thousands of clinical and biological datapoints were recorded. Knowing that we would ultimately be faced with a large number of candidate covariates and the danger of observing many (possibly spurious) associations, we limited ourselves to a handful of known prognostic factors found in systematic reviews of the ECT literature and/or large cohort studies of recurrence in treatment-resistant depression in order to avoid overfitting. There is over half a century of prospective research on post-ECT relapse. Several immutable patient and illness characteristics have been shown to predict both acute and long-term ECT outcomes, while ECT technical parameters or the adequacy of post-ECT prophylactic treatment do not moderate long-term outcomes to any clinically meaningful degree, with two known exceptions: lithium4 and continuation ECT.5 Maximizing the chances of a good long-term outcome, therefore, is largely predicated on careful patient selection, ensuring that ECT is delivered only to those who are suitable candidates for it. For these reasons, second-generation antipsychotics were not analyzed since there is, to our knowledge, no evidence demonstrating their usefulness in mitigating post-ECT relapse. Recent large observational studies from Scandinavia have shown that antipsychotics are associated with worse long-term outcomes after ECT in unipolar6 and bipolar depression.7 While causality cannot be inferred from these studies, the preliminary evidence is not encouraging. At any rate, the proportion of our remitters maintained on second-generation antipsychotics (mean dose of 8 mg olanzapine equivalents) who relapsed was similar to those treated with other medication classes (Fisher exact P = .599), though this is clearly not a randomized comparison. All of us working in the ECT field agree that very large multicenter studies are needed. In the meantime, clinical decisions must be made daily based on imperfect data. We therefore should not rule out most of the available evidence on post-ECT relapse.