Objectives: To develop methodology for the analysis of individual patient level data from multicentre/multinational randomized controlled trials with the aim of estimating location-specific parameters to populate decision models for location-specific decision making. Methods: Multilevel or hierarchical modelling is the analytical framework used to handle hierarchical cost-effectiveness data. Hierarchical modelling was developed in a Bayesian framework and Bayesian shrinkage estimation procedures were used to obtain location-specific cost-effectiveness estimates. Results: Using data from a recently conducted economic analysis of the RITA 3 trial, locationspecific cost-effectiveness measures were obtained and compared to the trial-wide results. For the analysed centres, the centre-specific cost-effectiveness planes showed higher variability in mean differential cost and mean differential QALY estimates compared to the trial wide results, with the latter having longer left tail estimate distribution. The majority of the location-specific incremental cost-effectiveness ratio results show higher cost per QALY for the intervention strategy compared to the trial wide results (approx. £41,400/QALY). With respect to centre-specific cost-effectiveness acceptability curves, the curves for the selected centres display great variability across centres in cost-effectiveness for given values of the threshold, λ. If the decision maker is willing to pay £50,000 for an additional QALY, the probability that the intervention strategy is cost-effective is, for instance, 0.34 for centre 37, compared to the 0.65 for the trial wide results. Conclusions: This thesis shows how Bayesian hierarchical modelling can be used to estimate more appropriate cluster-specific parameters for use in decision analytic models where individual patient level data from a multi-location trial are available. Bayesian hierarchical modelling estimates can be used to explore correctly the variability between centres/countries of the cost-effectiveness results allowing the correct quantification of uncertainty by adjusting the standard errors to reflect the estimates variability both within and between locations.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)