Predictors of 6- and 12-Month Relapse After Stopping Electroconvulsive Therapy: Critical Considerations, Including Overfitting in Regression and Confounding in Follow-up Studies.

Andrade, Chittaranjan

Patients with major mental illness are at high risk of relapse if they discontinue acute or continuation phase pharmacotherapy. This may explain the high rates of relapse after the termination of an effective course of electroconvulsive therapy (ECT) or after the discontinuation of effective maintenance ECT (M-ECT). Two moderately large studies prospectively examined predictors of relapse in the context of ECT. The first study, conducted in 61 depressed patients who had remitted with ECT and who were maintained on individualized pharmacotherapy, found that 39% of the patients relapsed within a year, with most of the relapses occurring during the first 6 months. Older age and the presence of psychotic symptoms before ECT predicted a lower risk of relapse, and a bipolar II diagnosis and a larger number of previous depressive episodes predicted a higher risk of relapse. Lithium appeared to protect against relapse. The second study, conducted in 81 patients with different diagnoses, found that 44% of patients relapsed within 6 months of the abrupt, unplanned discontinuation of M-ECT; the median time to relapse was 8 weeks. Predictors of relapse were psychosis, receipt of a larger number of previous courses of ECT, and need for more frequent M-ECT. The methods and results of these studies are critically examined. Special mention is made of overfitting and confounding in data analysis in follow-up studies such as these. Overfitting happens when investigators use more predictor variables in their statistical model than the sample size allows for; overfitting results in overly optimistic models. Confounding happens when the statistical model excludes important explanatory variables, including variables such as the appropriateness and adequacy of maintenance pharmacotherapy, adherence to maintenance pharmacotherapy, the stress-support dimension, and interactions between important explanatory variables.