The effectiveness and resource use data were derived from the UK General Practitioner Research Database (UK-GPRD), which collected medical information from a representative sample of general practitioners (GPs). The time horizon of the analysis was four years.
The effectiveness data were derived from the UK-GPRD which, according to the authors, was the world’s largest computerised database of anonymous longitudinal records in primary care medicine. The treatment groups were compared at baseline on demography, general co-morbidity, and eye co-morbidity. Adjustments were made to control for bias from unbalanced confounding factors of treatment efficacy using propensity scores. Propensity scores were defined as the conditional probability of being treated with a specific treatment according to its observed covariates. A total of 54,513 glaucoma patients were identified in the database and 639 of these received first-line treatment with travoprost and 387 with dorzolamide and timolol. The mean duration of follow-up was significantly longer for the dorzolamide and timolol group (929 days) than for the travoprost group (515 days). The main clinical effectiveness estimate was the treatment failure (see 'Measure of Benefit' below for the definition).
Monetary benefit and utility valuations:
Measure of benefit:
The measure of benefit was treatment failure, which was defined as a prescription change, replacement or discontinuation of the designated medication, addition of another treatment, laser therapy, or surgery for glaucoma.
Only the glaucoma-specific direct costs were included in the analysis. These included surgery and laser therapy, hospitalisations, medications, GP and specialist visits, and prescription renewals by telephone. The resource use data were derived from the UK-GPRD and the unit costs of medication were taken from the British National Formulary. The unit costs of medical procedures were derived from a published study. Specialist and GP visit data were derived from a compendium of UK unit health care costs. As the duration of follow-up was significantly longer for dorzolamide and timolol, the estimates were adjusted using a regression model. All prices were updated to 2005 using the health service inflation rate. The costs were reported in UK pounds sterling (£) and discounting was not reported.
Analysis of uncertainty:
All the cost, resource use and outcome differences between the two groups were tested using statistical analyses. The continuous variables were compared using a z-test if they were normally distributed and the non-parametric Wilcoxon test if they were not normally distributed. Discontinuous variables were compared using a chi-square test. The significance was set at 5% and times to treatment failure were compared by means of an adjusted Cox model using the propensity score method.