A Bayesian Network Meta-Analysis With Random Inconsistency Effects for Multi-Arm Studies

Palash Kumar Malo , National Institute of Mental Health and Neuro Sciences (NIMHANS)
B. Binukumar, National Institute of Mental Health and Neurosciences (NIMHANS)
Muralidharan K., National Institute of Mental Health and Neurosciences (NIMHANS)
K. Thennarasu, National Institute of Mental Health and Neurosciences (NIMHANS)

The design-by-treatment interaction model for network meta-analysis with random inconsistency effects can also be applied whenever arm-level binary outcome data are available. This study aims to investigate the ranking of treatments for efficacy of 11 pharmacological drugs in 45 randomized controlled trials for acute bipolar mania in adults using arm-based analysis within a Bayesian framework. For binary data, a binomial distribution has been adopted for the number of events and the logit scale to model the probability of event occurrence. Two sensitivity analyses were conducted: (1) for the fixed values; and (2) for the prior distributions of inconsistency variance. The impact of including inconsistency in the model is 14.05%. The probability of carbamazepine being the best decreases from 0.41 to 0.27 as the level of inconsistency increases from 0 to 0.3; whereas it varies from 0.38 to 0.40 for the selected prior distributions. Carbamazepine ranked as the most efficacious drug.

See extended abstract

 Presented in Session 11. Health & Mortality 2