Investigating The Impact of Structural Changes In A Nice Single Technology Appraisal Cost-Effectiveness Model.

One of the major critiques with submitted manufacturer’s cost-effectiveness models is surrounding the structural uncertainty. However, methods dealing with structural uncertainties are not well-developed, even though these might have a significant impact on model results. This study investigates the impact of structural changes in a National Institute for Health and Care Excellence (NICE) single technology appraisal cost-effectiveness model of Erlotinib versus Best Supportive Care as a maintenance therapy for patients with non-small cell lung cancer. The manufacturer’s model submission was criticised for having a “Markov” model not governed by transition probabilities. It considered an independent projective survival functions for progression-free survival and overall survival, which allowed a negative post-progression survival (PPS) estimate to appear in later cycle. Using published summary survival data, this study adopted three approaches, covering both fixed- and time-varying, to estimate health state transition probabilities that are used in a restructured Markov model. Unlike for placebo, the parametric approach estimates post-progression probabilities and probabilities of death for Erlotinib differently than fixed-transition approaches. The best fitting curves are achieved for both PPS and probability of death across the time for which data were available, but the curves start diverging towards the end of this period. The alternative (Markov) model which extrapolates the curves forward in time suggests that this difference between a time-varying and fixed-transition becomes even greater. The alternative models produce an Incremental Cost-Effectiveness Ratio (ICER) of £54k -£66k per quality adjusted life year (QALY) gain, which is comparable to an ICER presented in the MS (£55k/QALY gain). The results from restructured alternative models do not suggest different cost-effectiveness results to those reported in the manufacturer submission; however, in terms of magnitude they vary. This variation in cost-effectiveness results produced by restructured models might be crucial for interventions falling near a threshold value.