Studies that evaluated patients 18 years of age or older who were diagnosed with a major depressive episode, defined by a standard, accepted scale (eg, Diagnostic and Statistical Manual of Mental Disorders or International Classification of Diseases-9) as either endogenous or unipolar depression, were eligible. Patients must have scored 15 or greater on the Hamilton Depression (HAMD) scale (any version), or 18 or greater on the Montgomery-Asberg (MADRS) scale . Patients could not have been taking any concomitant antidepressants, lithium or thyroid, nor have any concomitant diseases (especially metabolic, endocrine or psychiatric). They could, however, have been taking other drugs such as tranquilizers or hypnotics.
Efficacy was defined as a 50% reduction in HAMD or MADRS score. Tolerability was defined in terms of dropouts due to adverse drug reactions (ADRs) and lack of effect. Clinical rates for in-patients and out-patients were determined separately.
Two reviewers independently identified studies to be included in the analysis. Inter-rater agreement of both study selection and data extraction was evaluated by calculating kappa. Disagreements were resolved through consensus. The rationale for decisions was discussed until reviewers agreed on the final decision.
Data were combined using a random effects model, modified for use in combining data from individual arms of trials or cross-sectional studies . Results were summarized across studies or arms of studies to arrive at a single estimate for success rates, along with a 95% CI. Both within-study variance and between-study variance are incorporated into the standard error, which reduces the impact of heterogeneity among studies. The CI produced by this approach allows for evaluation of the sensitivity of parameters to changes in estimates and can serve as a rough statistical test for differences in rates.