Synthesis of Survival Outcomes in Economic Evaluation: Does the Network Meta-Analysis Model Matter?
Author(s)
Tebbs H1, Dietz J1, Downing B2, Welton N2, Claxton L1
1National Institute for Health and Care Excelllence, London, LON, UK, 2University of Bristol, Bristol, UK
Presentation Documents
OBJECTIVES: Network meta-analysis (NMA) of time-to-event data has been developed for a range of different parametric and non-parametric survival models. These models can lead to differences in survival estimates, which, in the context of HTA submissions that often rely on the proportional hazards assumption, could translate into different cost-effectiveness results and thereby affect the decision-making process. This research aims to assess the impact of different NMA methods on cost-effectiveness results in a recent NICE guideline.
METHODS: Using a network of 9 treatments for advanced melanoma, we fit survival models for progression-free (PFS) and overall survival (OS) for: Cox proportional hazards, generalised gamma with one and two treatment effects, fractional polynomial (FP), and piecewise exponential with different cut points. NMA model estimates were incorporated in the economic model and extrapolated over the patient lifetime to compare the impact on total costs, QALYs, and net monetary benefit (NMB).
RESULTS: The type of NMA model led to different survival and cost-effectiveness results, and the magnitude of the difference varied by treatment. The Cox model was a poor fit and overestimated survival relative to all other models: for example, pembrolizumab had 26% fewer QALYs with the FP model. All models except the generalized gamma with one treatment effect indicated the same treatment as the most cost-effective and all predicted the same treatment as the least cost-effective.
CONCLUSIONS: Despite different survival and NMB estimates across the different NMA models, the overall conclusions of the cost-effectiveness analysis were consistent in this case study. However, this may not be the case in other economic evaluations where differences in OS between treatments are smaller, or where the economic model is more sensitive to OS. As this is a single case study, further work is required to understand situations where decision-making is sensitive to the choice of NMA models.
Conference/Value in Health Info
Value in Health, Volume 25, Issue 12S (December 2022)
Code
MSR79
Topic
Clinical Outcomes, Methodological & Statistical Research, Study Approaches
Topic Subcategory
Comparative Effectiveness or Efficacy, Decision Modeling & Simulation, Meta-Analysis & Indirect Comparisons
Disease
STA: Drugs