PURPOSE: Bipolar I disorder (BPI) is a severe and recurrent mood disorder associated with high morbidity and mortality. Previous evaluations of pharmacological interventions have focused on direct (pair-wise) comparisons which makes it difficult to draw an overall conclusion about which treatment is optimal, particularly when several competing treatments are available. To facilitate a direct comparison of all relevant comparators, Bayesian evidence synthesis approaches were employed within a decision-analytic framework to estimate the relative cost-effectiveness of alternative pharmacological treatments for relapse prevention in BPI.

METHODS: In the absence of direct ‘head to head' evidence for all comparators, a network was employed in order to link direct and indirect evidence for the separate treatments. The network comprised evidence on relapse rates for seven strategies, including six monotherapy treatments (lithium, valproate, carbamazapine, imipramine, lamotrogine and olanzapine) and one combination treatment (lithium plus imipramine), reported in 16 separate RCTs. A Bayesian mixed treatment comparison (MTC) model, using Markov Chain Monte Carlo methods, was used to combine relapse rates from the RCTs. Data on different definitions of relapse were used to ‘borrow strength' in estimating the main outcomes of interest. This analysis was used as an input into a long-term Markov model used to evaluate the relative cost-effectiveness of the treatments during the maintenance phase of BPI. The model provided a framework to estimate lifetime QALYs and costs from a health service perspective. Heterogeneity was assessed according to presenting episode (manic or depressive), which was seen as an important predictor of both the risk and type of subsequent episodes.

RESULTS: The cost-effectiveness of the alternative treatments was demonstrated to be closely related to previous episode history. For patients with a recent depressive episode, valproate, lithium monotherapy and the combination of lithium and imipramine appear potentially cost-effective, depending upon the threshold cost per QALY. For patients with a recent manic episode, olanzapine and lithium monotherapy appear potentially cost-effective. These conclusions are sensitive to the assumption concerning the reduction of suicidal risk.

CONCLUSIONS: The use of an MTC model enables the simultaneous comparison of multiple treatments, combining direct head-to-head and indirect evidence, in a single coherent analysis. These methods allow fuller consideration of the existing evidence base, retaining the benefits of randomisation, and provide the most appropriate basis for informing cost-effectiveness analyses.