Analytical approach:
A Markov model was constructed to assess the costs and outcomes of the two interventions under study. The time horizon of the study was the lifetime of the patient. The perspective adopted in the economic analysis was that of the Swiss healthcare system.
Effectiveness data:
The clinical and effectiveness data were derived from a single phase III placebo-controlled randomised study by Ciuleanu et al. (see Other Publications of Related Interest). The main measure of effectiveness used in the model was progression-free survival. Overall survival was considered as a secondary outcome. The authors reported that progression-free survival was only estimated from the 481 non-squamous cell lung cancer patients included in the trial. Hazards were assumed constant over time and were converted into transition probabilities using median time spent in each stage of disease.
Monetary benefit and utility valuations:
Utility estimates were derived from previously published studies.
Measure of benefit:
Quality-adjusted life-years (QALYs) gained. Benefits could be generated over the lifetime of the patient. The short life expectancy of patients meant that benefits were left undiscounted.
Cost data:
The direct costs included in the economic analysis were pemetrexed, chemotherapy, monitoring, treatment of side-effects and follow-up treatment. For the pemetrexed group, resource use was obtained from the same clinical trial as that used to obtain treatment effectiveness. Costs associated with best supportive care were derived from a Dutch study. The authors reported that all costs were adjusted for purchasing power differences between Netherlands and Switzerland. The price year was 2010. All costs were reported in Euros (€) at an exchange rate of 1 Swiss Franc = €0.72. Costs could be incurred over the lifetime of the patient but were left undiscounted due to the short life expectancy of patients.
Analysis of uncertainty:
The authors conducted a series of one-way and probabilistic sensitivity analyses to assess the influence of uncertainty on key model parameters. A probabilistic sensitivity analysis was conducted by fitting model parameters with corresponding probability distributions and based on 1,000 sets of randomly drawn input parameters.