Predictors of relapse of psychotic depression: Findings from the STOP-PD II randomized clinical trial.

Psychotic depression has a high rate of relapse. The study aims were to identify a prediction model of risk of relapse of psychotic depression and examine whether predictors moderated the effect of treatment on relapse. One hundred and twenty-six men and women aged 18–85 years, who experienced sustained remission or near-remission of psychotic depression with sertraline plus olanzapine, participated in a 36-week randomized controlled trial that compared sertraline plus olanzapine with sertraline plus placebo in preventing relapse (NCT01427608). Cox regression analyses were performed to identify significant predictors of relapse and to model the combined role of significant predictors. Concordance statistic was calculated to determine the accuracy of the best fit multivariable models in predicting relapse. Finally, interaction terms were tested for each significant predictor to examine whether they moderated the effect of treatment on risk of relapse. Lifetime number of depressive episodes, severity of residual depressive symptoms at the time of randomization, and psychomotor disturbance both at acute enrollment when participants were depressed and at the time of randomization predicted risk of relapse. Multivariable models had 69–70% accuracy in predicting relapse. Psychomotor disturbance was associated with increased risk of relapse in the sertraline plus olanzapine group compared with sertraline plus placebo, whereas the other predictors did not moderate the effect of treatment on relapse. Future research is needed to determine whether a combination of clinical and biological variables can further increase the accuracy of prediction of relapse of psychotic depression. • Psychotic depression has a high rate of relapse. • Identify predictors and moderators of risk of relapse of remitted psychotic depression. • Lifetime number of depressive episodes, residual depressive symptoms, and psychomotor disturbance predicted relapse. • Multivariable models had 69–70% accuracy in predicting relapse. • Psychomotor disturbance moderated the effect of randomized treatment on relapse. [ABSTRACT FROM AUTHOR]