Cost-Utility of Cancer Therapies - the 'Cost' of Different Utility Generation Strategies.

Abstract

Objectives: To explore the impact of different utility measurement strategies on the results of a cost-effectiveness analysis, funding decisions, decision uncertainty and value of information. Methods: Data from a UK trial of two cancer therapies (active versus standard care) were analysed using NICE reference case methods. Within-trial, cost-utility analyses were conducted with utility based on a number of strategies: A) Observed EQ-5D; cancer-specific utility based on the EORTC QLQ-C30 B) the EORTC-8D and C) the QLQ_U; Mapping from QLQ-C30 to EQ-5D using an algorithm generated in D) the same cancer patient group and E) a different cancer group. Incremental cost-effectiveness ratios (ICERS) were calculated. Bootstrapped net benefit estimates allowed generation of cost-effectiveness acceptability curves (CEACs) and population expected value of perfect information (EVPI) was calculated using incremental cost scenarios. Results were compared across utility strategies. Results: There were small but important differences observed in the incremental QALYs which ranged from 0.067 (EQ-5D) to 0.036 (EORTC-8D). Large differences were observed in the ICERs generated; for strategies A to E these were: £57,513; £106,264; £102,785; £90,049; £78,885. Using an incremental cost scenario of £3,000 only strategy A yielded an ICER <£30,000. At a QALY willingness to pay threshold (WTPT) of £20,000 there was little decision uncertainty. However, assuming WTPT=£50,000, the probability the active treatment was cost-effective ranged 0.34 (EQ-5D) to 0.025 (EORTC-8D). Using this threshold, the population EVPI for the strategies were: £3,597,844; £120,621; £155,858; £354,094; £805,847. Conclusions: Different utility sources can lead to very different estimates of cost-effectiveness and value of further research and change funding decisions. Estimates of cost-effectiveness based on mapping (even when the algorithm appears to perform well) can differ substantively from those based on observed scores. The lowest ICERs were obtained with the EQ-5D but this may not capture side-effects picked up by the cancer-specific utility measures.