Bayesian estimation and prediction for network meta-analysis with contrast-based approach.

Area of Science:

• Clinical Epidemiology
• Health Technology Assessment
• Biostatistics

Background:

• Network meta-analysis (NMA) is crucial for assessing comparative treatment effectiveness.
• Bayesian methods, particularly reference analyses with noninformative priors, are standard for arm-based NMA.
• Existing methods often rely on iterative computations like Markov chain Monte Carlo (MCMC).

Purpose of the Study:

• To introduce generic Bayesian analysis methods for the contrast-based approach in NMA.
• To enable direct posterior and posterior predictive distribution sampling without MCMC.
• To provide a flexible framework accommodating both proper and improper prior distributions.

Main Methods:

• Developed generic Bayesian methods for contrast-based NMA.
• Incorporated direct sampling techniques, bypassing iterative computations.
• Included representative noninformative priors, such as the Jeffreys prior.
• Created an R package (BANMA) for practical implementation.

Main Results:

• The proposed methods allow direct sampling from posterior distributions, eliminating the need for convergence checks.
• The framework handles various prior distributions, including noninformative types.
• The BANMA package offers a user-friendly interface for applying these methods.

Conclusions:

• The novel Bayesian methods offer a computationally efficient and accessible approach to contrast-based NMA.
• These advancements facilitate broader application of Bayesian techniques in health technology assessments.
• The BANMA package simplifies the implementation of advanced Bayesian NMA.