Opposing Duration of Treatment Effect Assumptions Used in the Health Technology Assessment of a Treatment for Rituximab-Refractory Follicular Lymphoma
Author(s)
Dave K1, Strunz-McKendry T1, Felizzi F2, Launonen A2, Krivasi T2
1Roche Products Ltd, Welwyn Garden City, UK, 2F. Hoffmann-La Roche Ltd, Basel, Switzerland
OBJECTIVES: Models were submitted to NICE for obinutuzumab in combination with bendamustine (G-benda+G) for the treatment of rituximab-refractory follicular lymphoma using data from the GADOLIN trial (NCT01059630). NICE accepted the partitioned survival model (PSM) that followed the Kaplan-Meier until the last death event after which hazards for G-benda+G matched that of bendamustine using a fitted Weibull function. This research aims to compare modelling approaches and assumptions related to the duration of treatment effect. METHODS: Extrapolations based on subsequent data cuts from the GADOLIN study were compared using different modelling preferences. The study read out at four clinical cut-off dates; primary (1-Sep-2014 [median follow-up: 22 months]), update 1 (1-May-2015), update 2 (1-Apr-2016), and the final data cut 30-Nov-2018. For update 1, a semi-Markov model was used since median survival for both arms was not reached. By 2016, further data allowed survival to be extrapolated by fitted parametric curves. RESULTS: A semi-Markov model (data: update 1) predicted a survival gain of 1.8 years. The Evidence Review Group’s preference used a PSM and limited treatment effect to the last observed death returning a more conservative estimate of 1.3 years. In the most recent review at the final data cut, this same approach generated an average survival gain of 2.8 years agreeing with a different approach (random change-point Weibull models) that did not assume a ceasing of treatment effect. CONCLUSIONS: Whilst the original semi-Markov model underestimated average survival gain, the PSM approach, assuming treatment effect ceases, underestimated this even further. The final data cut, with almost 3 years more follow up than the primary analysis, did not indicate a finite duration of treatment effect on survival. For indolent diseases, where immature survival data may be expected, a semi-Markov approach can provide reasonable survival predictions even without imposing limited duration of treatment effect assumptions.
Conference/Value in Health Info
2020-11, ISPOR Europe 2020, Milan, Italy
Value in Health, Volume 23, Issue S2 (December 2020)
Code
PCN245
Topic
Health Technology Assessment, Methodological & Statistical Research
Topic Subcategory
Decision & Deliberative Processes
Disease
Biologics and Biosimilars, Drugs, Oncology