Statistical Analysis
We compared the two treatment groups on all demographic, illness history, and medication variables listed in Table 1. Site and pretreatment variables that were imbalanced or that predicted earlier study termination were included as covariates in the statistical models. Time to recovery was calculated as the number of weeks from randomization until weekly Psychiatric Status Rating Scale scores were #2 for at least 8 consecutive weeks for depression, mania/hypomania, or both. Time to recurrence was computed as weeks from the point of recovery to the point when the patient had Psychiatric Status Rating Scale scores $5 for at least 1 week for mania/hypomania or at least 2 weeks for depression. All randomized patients were included in the at-risk sample; those who did not recover and those who left the study early were censored at the point of their final research assessment. Survival curves for time to recovery or recurrence were compared across groups using the Kaplan-Meier product-limit equation (22). Cox proportional hazards models (23) were used to control for covariates (site, baseline symptom severity and polarity, and weeks of follow-up, as well as any variables identified in the attrition analyses). With a sample size of 145, a two-sided log-rank test had 90% power to detect a 25% difference in survival proportions, assuming 20% attrition (p,0.05).