Article type
Year
Abstract
Background: Bipolar I disorder is a severe and recurrent mood disorder associated with high morbidity and mortality. A number of pharmacological interventions have been evaluated for the prevention of relapse. Previous systematic reviews have focused on pairwise, direct comparisons of treatments, which makes it difficult to draw an overall conclusion about which treatment is optimal when several possible treatments are available.
Objectives: To apply Bayesian statistical methods to compare simultaneously multiple treatments, combining direct and indirect evidence from sixteen randomized controlled trials (RCTs).
Methods: We undertook simultaneous comparisons of eight treatments using a Bayesian mixed treatment comparison model (MTC). This provided a single statistical model providing simultaneous estimates for all the 28 possible pair-wise comparisons reported in the RCTs. Markov Chain Monte Carlo methods, adapted to apply to a network of treatment comparisons, were used to combine direct and indirect evidence for each possible comparison. Results were estimated separately for both manic and depressive relapses. When outcomes were not reported separately the model borrowed strength from the data available on all types of relapses.
Results: Direct data were available for 12 possible pair-wise comparisons. The results from the MTC model demonstrated broad agreement with these results. The use of MTC confers significant advantages over conventional approaches, enabling a simultaneous assessment in order to identify which treatment is the most effective option. For the prevention of manic episodes, olanzapine was clearly identified as the most effective (mean posterior probability =0.94), followed by valproate (0.002). For the prevention of depressive episodes, the combination lithium and imipramine was identified as the most effective treatment (mean posterior probability = 0.35), closely followed by valproate (0.32).
Conclusions: In the absence of direct head-to-head evidence for all relevant comparators, the combination of direct and indirect evidence in a single analysis enables decision makers to identify the most effective treatment.
Objectives: To apply Bayesian statistical methods to compare simultaneously multiple treatments, combining direct and indirect evidence from sixteen randomized controlled trials (RCTs).
Methods: We undertook simultaneous comparisons of eight treatments using a Bayesian mixed treatment comparison model (MTC). This provided a single statistical model providing simultaneous estimates for all the 28 possible pair-wise comparisons reported in the RCTs. Markov Chain Monte Carlo methods, adapted to apply to a network of treatment comparisons, were used to combine direct and indirect evidence for each possible comparison. Results were estimated separately for both manic and depressive relapses. When outcomes were not reported separately the model borrowed strength from the data available on all types of relapses.
Results: Direct data were available for 12 possible pair-wise comparisons. The results from the MTC model demonstrated broad agreement with these results. The use of MTC confers significant advantages over conventional approaches, enabling a simultaneous assessment in order to identify which treatment is the most effective option. For the prevention of manic episodes, olanzapine was clearly identified as the most effective (mean posterior probability =0.94), followed by valproate (0.002). For the prevention of depressive episodes, the combination lithium and imipramine was identified as the most effective treatment (mean posterior probability = 0.35), closely followed by valproate (0.32).
Conclusions: In the absence of direct head-to-head evidence for all relevant comparators, the combination of direct and indirect evidence in a single analysis enables decision makers to identify the most effective treatment.