Rationale: There is a shortage of studies analyzing the time course of recurrent episodes and comparing effectiveness of long-term treatments in bipolar disorder. 'Number needed to treat' (NNT) analyses have been proven to be useful for clinically meaningful comparisons, but results vary considerably among studies. The survival curves of different trials also show a great variability preventing reliable conclusions on the time course of maintenance therapies. The variance of survival analyses on long-term medication management can be reduced with increasing the statistical power by combining the life-tables of individual studies.
Methods: In this study the survival tables of 28 studies on maintenance treatment of bipolar disorder were reconstructed from the published diagrams, and the numbers of relapsed patients in the original studies were estimated for plotting composite survival curves of an inactive, mono- and combination therapy arm. The review was finally based on 5231 subjects.
Results: The resulting composite diagrams indicate that within the first year 48% of patients on monotherapy, and 35% on combination therapy experienced recurrence of any affective episode ('early relapsers'). The rest of the patient population was affected by recurrences in a smaller rate over a more extended period of time ('late relapsers'). For a favorable outcome at 40 months of episode prevention in bipolar disorder the NNT was 6 for mono- and 3 for combination therapy. Log-rank analyses of the composite data supported the effectiveness of both medication protocols over placebo, and the superiority of drug combination over monotherapy; though there were some indications of decreased efficacy in the two treatment arms after extended maintenance.
Conclusions: Composite analysis offers increased statistical power for studying the time course of survival data. Mood episodes in bipolar disorder are likely to recur early on and relapses in "real-life" can be more frequent than the rates published here. Our results favor combination therapy for the long-term management of bipolar disorder. Concerns are expressed that NNT analyses have significant limitations when applied to recurring events with cumulative deterioration instead of cases where cumulative improvement is expected over time.