ESTIMATING SURVIVAL GAIN FOR COST-EFFECTIVENESS ANALYSIS- THE EXAMPLE OF EARLY HORMONAL TREATMENT IN LOCALLY ADVANCED PROSTATE CANCER

OBJECTIVES: To examine the impact of various methods of estimating survival gain from censored survival data on the results of a cost-effectiveness analysis. METHODS: A cost-effectiveness analysis of the addition to radiotherapy of early hormonal therapy with a LHRH analogue (goserelin) in patients with locally advanced prostate cancer has been carried out. The clinical results derive from a randomized controlled trial (EORTC 22863, reported by Bolla et al. [1]), and resource utilization data were collected retrospectively in one French hospital recruiting 90/415 patients in the trial. Costs are calculated from the viewpoint of the French health insurance system. To estimate average survival, various methods are available and will be compared, notably a restricted means analysis truncated at a median follow-up of 45 months and extrapolations based on fitting a parametric model (e.g., Weibull or exponential) to the survival curves. RESULTS: The clinical trial documented an overall survival rate at 5 years of 79% in the group receiving radiotherapy plus hormonal therapy compared to 61% (p = 0.001) in those receiving only radiotherapy. The average costs in the two groups were 81,700 FF and 69,900 FF respectively. The survival improvement is thus associated with an increase of 11,800 FF in average cost, but to determine the incremental cost-effectiveness ratio (ICER) the exact gain in average survival must be known. The results of using the various estimation methods will be presented and discussed at the meeting. CONCLUSIONS: Survival data from clinical trials are normally not analyzed with the purpose of determining average survival, but such estimates are necessary for economic evaluation. Due to censoring, mean survival can not be estimated without extrapolation or without truncation in a restricted means analysis. The results in terms of ICER will vary according to the estimation method used. These variations should be examined in sensitivity analyses