BACKGROUND:Delayed onset of efficacy of antidepressants and a high proportion of depressed patients being poor or non-responders to antidepressants are well known clinical challenges. Therefore, it seems to be necessary to identify predictors for response and - even more important - for remission. It has been suggested that reduction of depressive symptoms at an early stage of antidepressant treatment may predict treatment outcome. Our objective was to test, if this hypothesis derived from randomized controlled studies (RCTs) in outpatients, would be confirmed in a large naturalistic study in a cohort of inpatients with major depression. Patients were treated with various antidepressants and co-medication according to the protocol based on evidence-based clinical guidelines. METHODS:This was a large naturalistic prospective study. All patients (N=795) were hospitalized and met DSM-IV criteria for major depression according to a structured clinical interview (SCID). Assessments were conducted biweekly. Several definitions of early improvement (20%, 25% and 30% reduction in HAMD-21 baseline total scores) at two different visits were tested. Sensitivity, specificity and predictive values were calculated for the different definitions of early improvement. ROC-analyses as well as logistic regression models have been performed. Response was defined as 50% improvement of the total baseline HAMD-21 score and remission as a score of