[MO7] Evaluation of a Bayesian comprehensive decision-analytical modelling framework in chronic hepatitis c

[MO7] Evaluation of a Bayesian comprehensive decision-analytical modelling framework in chronic hepatitis c

2009 Value in health

Cawston, H. | Aballea, S. | Cure, S. | Volume: 12, Issue: 7, Pages: A231, economic evaluation, model, Meta-analysis, methods, Disease, Disease Progression,

OBJECTIVES: A standard approach to economic evaluation is to conduct a metaanalysis
to subsequently inform a decision-analytical model. An innovative approach
is to implement the meta-analysis and model evaluation simultaneously, in a Bayesian
framework. These approaches were compared in two Markov models for chronic
hepatitis C (CHC), where transition probabilities were not directly observable.
METHODS: We updated a published meta-analysis on progression rates in CHC.
Included studies provided distributions of patients by disease severity level, for different
durations of infection. Two structures were considered to model disease progression:
one with three states (“mild CHC”, “moderate CHC”, cirrhosis), another more
complex with five states. The following methodological approaches were compared:
1) the “standard” approach involving successively maximum-likelihood estimation of
progression rates for each study, random-effect meta-analysis of resulting estimates,
Markov model evaluation and probabilistic sensitivity analysis, and 2) an integrated
approach with estimation of progression rates and model evaluation in one Monte
Carlo Markov Chain simulation procedure. The impact of methodological approach
and model structure on predicted numbers of cirrhotic patients and total costs
from NHS perspective at 20 years was analysed. RESULTS: The five-state model predicted
the following proportions of cirrhotic patients: 39.6% (standard deviation:
0.075) and 50.2% (0.019) with approaches (1) and (2). The 3-state structure provided
lower estimates: 19.7% (0.0911) and 22.3% (0.011) with approaches (1) and (2). For
the 5-state model, predicted costs per patient were £7,120 (934) and £8,000 (856)
with each approach respectively. CONCLUSIONS: The comprehensive modelling
approach lead to higher mean estimates with lower variability than the standard
approach. The simpler three-state structure seems to underestimate the burden of
disease. The latter finding may have relevance for modelling progressive diseases more
generally.