OBJECTIVES:

Immuno-oncology (I-O) therapies are often associated with complex hazard functions, which standard parametric distributions may not adequately reflect. This research aimed to explore the use and acceptance of alternative approaches for modelling survival in cost-effectiveness analyses submitted as part of UK technology appraisals of I-O therapies.

METHODS:

The NICE website was searched in May 2018 for technology appraisals of seven I-O therapies (atezolizumab, avelumab, durvalumab, ipilimumab, nivolumab, pembrolizumab and tremelimumab). Information relating to the choice of modelling approach (manufacturer’s rationale and appraisal committee response) was extracted from appraisals that included either of the following approaches: mixture cure or response-based modelling.

RESULTS:

Alternative survival modelling approaches were included in the base case analysis of 4/22 appraisals: 3 mixture cure and 1 response-based. Manufacturers justified the use of alternative modelling approaches within these appraisals by claiming that standard parametric distributions could not accurately reflect the survival profile, for example, would fail to capture expected long-term survival benefits. However, alternative approaches were only accepted in 2/4 appraisals (both mixture cure) and in these appraisals, the cure fraction was set to 0% in the absence of long-term data. Instead, appraisal committees preferred flexible parametric or piecewise extrapolations. Key criticisms of these alternative approaches were that conventional modelling approaches were not shown to be inappropriate, and that evidence for a prolonged treatment effect was insufficient – for the trials used in each analysis, duration of follow-up was less than 5 years.

CONCLUSIONS:

Alternative approaches to modelling survival for I-O therapies have been included in cost-effectiveness analyses submitted as part of UK technology appraisals, but have not yet gained wide acceptance. Appraisal committees have acknowledged that these approaches are worth considering if properly evidenced, however, a lack of long-term trial data at the time of submission may limit their acceptance.