Parameterization of treatment effects for meta-analysis in multi-state Markov models

Area of Science:

• Biostatistics
• Health Economics
• Clinical Epidemiology

Background:

• Standard Markov models for randomized controlled trials (RCTs) analysis limit generalizability of treatment effects and evidence synthesis across studies.
• Event history data from RCTs on disease progression present challenges for traditional meta-analysis.

Purpose of the Study:

• To demonstrate the application of pair-wise and mixed treatment comparison meta-analysis to event history data from RCTs.
• To develop a structured modeling approach for synthesizing evidence and characterizing summary treatment effects.
• To apply these methods to asthma treatment data for ranking clinical effectiveness.

Main Methods:

• Developed a multi-state continuous-time Markov model, relating transition rates to a multinomial likelihood via Kolmogorov's forward equations.
• Applied pair-wise and mixed treatment comparison meta-analysis to aggregated discrete time transition data from five RCTs comparing eight asthma treatments.
• Utilized Bayesian inferential techniques and Markov Chain Monte Carlo (MCMC) simulation in WinBUGS for parameter estimation.

Main Results:

• A flexible characterization of summary treatment effects was achieved through a rate-based formulation.
• Models were developed where relative treatment effects influenced forward, backward, or both transition types, with model comparison using the Deviance Information Criterion (DIC).
• An intuitive mechanism of action was identified, with a single parameter affecting all backward transitions, enabling pooled treatment effect estimation.

Conclusions:

• The developed Markov modeling approach effectively synthesizes evidence from RCTs, overcoming limitations of standard methods.
• This methodology allows for flexible characterization and comparison of treatment effects, facilitating ranking of interventions.
• The findings enable the ranking of different asthma treatment options within evidence networks to identify the most clinically effective interventions.