Analytical approach:
The analysis was based on a probabilistic Markov model with a lifetime horizon. The authors stated that the perspective of the National Health Service (NHS) was adopted.
Effectiveness data:
A systematic review of the literature was undertaken. Several medical and economic electronic databases were searched. Details of search methods and inclusion/exclusion criteria were reported. Only randomised controlled trials (RCTs) with a follow-up longer than 12 months and systematic reviews of RCTs were included in the analysis. The key clinical input was the probability of undergoing total knee replacement, which was considered as a definition of treatment efficacy and was derived from a published pooled analysis of two RCTs with long follow-up periods (three and five years).
Monetary benefit and utility valuations:
Utility values were derived from a published RCT that reported Western Ontario and McMaster Universities (WOMAC) Osteoarthritis Index data, which could be converted into a preference-based utility scale, namely the Health Utilities Index (HUI3). This conversion was based on the results of a Canadian study that included 255 patients with symptomatic knee osteoarthritis.
Measure of benefit:
Quality-adjusted life-years (QALYs) were used as the summary benefit measure and were discounted at an annual rate of 3.5%.
Cost data:
The economic analysis included costs associated with supplements, management of knee osteoarthritis (general practitioner visits, medications, out-patient visits, in-patient care, professions allied to medicine consultations, complementary therapists and X-ray procedures) and total knee replacement surgery. Resource use data were obtained from an RCT conducted in the UK and from NHS reference costs. Glucosamine sulphate had no UK licence, thus the price of glucosamine hydrochloride was used. Costs were in UK pounds sterling (£) and referred to 2007 to 2008 prices. A 3.5% annual discount rate was used.
Analysis of uncertainty:
The issue of uncertainty was investigated with two approaches. First, both stochastic and non-stochastic parameters were varied individually in a one-way sensitivity analysis that used lower and upper bounds derived from published confidence intervals. Second, a probabilistic sensitivity analysis was undertaken that used specific distribution probability for model inputs. The expected value of perfect information (EVPI) was determined in order to investigate whether it would be worthwhile commissioning further research on the cost-effectiveness of supplements.