Analytical approach:
: A Markov model was developed to estimate the cost-effectiveness of etoricoxib versus nsNSAIDs over a period of 52 weeks. In an additional analysis, the health states and utility scores at 52 weeks were assumed to persist over the life expectancy. The authors stated that the perspectives of society and the UK National Health Service (NHS) were adopted.
Effectiveness data:
The effectiveness and clinical data were derived mainly from randomised controlled trials (RCTs) and meta-analyses. The main effectiveness parameter was the probability of discontinuing treatment due to lack of efficacy. This effectiveness parameter was derived from a clinical trial in ankylosing spondylitis (van der Heijde, et al. 2005, see 'Other Publications of Related Interest' below for bibliographic details).
Monetary benefit and utility valuations:
Quality-of-life weights were derived from published data on Bath AS Functional Index (BASFI) scores and Short Form (SF)-36 scores.
Measure of benefit:
The measures of benefit were quality-adjusted life-years (QALYs) saved, the number of upper gastrointestinal perforations, ulcers or bleeding, and the number of lower gastrointestinal bleeds.
Cost data:
The direct costs were those that related to treatment (medications and dispensing), primary care consultations, out-patient gastrointestinal consultations, investigations, in-patient days and surgery. The resource use data were derived from published studies and an expert panel. In order to obtain the episodes of sickness absence, a logistic regression was performed that described the probability of sickness absence for a given BASFI score. Costs that related to sickness absence were calculated by applying the human capital approach and using average daily earnings. Health care unit cost data were derived from the literature, the MediPlus UK 2004 database, and data from NHS trusts. All costs were reported in 2004 prices and expressed in UK pounds sterling (£).
Analysis of uncertainty:
A probabilistic sensitivity analysis was performed to quantify the uncertainty in the model outcomes. Each model parameter was fitted with a distribution that reflected its uncertainty. A random value was then sampled over 1,000 times from each distribution. A series of scenario analyses were performed by varying the cost of nsNSAIDs, the proportion of patients in employment, and the number of patients receiving concomitant proton pump inhibitor from the beginning.