VP82 Impact of Evidence Synthesis Methods on Outcome of Economic Evaluation. (2019)

• Record Type: Journal Article
• Authors: Gorry, Claire; Leahy, Joy; Lamrock, Felicity; Walsh, Cathal; White, Arthur; Barry, Michael; McCullagh, Laura
• Is Part Of: International journal of technology assessment in health care. Volume 35(2019)Supplement 1
• Page Start: 93
• Page End: 94
• Publication Date: 2019
• Subjects: Medical technology -- Periodicals; Technology assessment -- Periodicals; 610.28
• DOI: 10.1017/S0266462319003325
• Languages: English
• ISSNs: 0266-4623

Abstract

Abstract : Introduction: Evidence synthesis (ES) is often required for economic evaluation (EE) of pharmaceuticals. Commonly used methods are based on the assumption of proportional hazards in trial data, using the hazard ratio (HR). Alternative methods for ES are increasingly used in EE, in situations where the pattern of hazards in the trial data indicates that the proportional hazards assumption may be violated. The impact of these methodological choices on model outcomes is explored. Methods: A network of trials of BRAF-targeted treatments for advanced melanoma, derived using a systematic review of the literature, is chosen for the study. Guyot's method is used to create individual-patient Kaplan-Meier (K-M) data from published survival curves. Log-cumulative hazard plots and Schoenfeld residuals are derived to examine patterns in hazards within the trial data. All analyses are conducted in R version 3.5.0©. Three alternative methods for ES are tested: 1) Network meta-analysis (NMA) based on published HRs and the assumption of proportional hazards. 2) NMA using fractional polynomials (FP) based on digitised K-M data, allowing the relaxation of the proportional hazards assumption. 3) NMA using an accelerated failure time (AFT) model based on digitised K-M data, allowing the relaxation of the proportional hazards assumption. The derived estimates of relative efficacy from each method are applied in a partitioned survival cost-effectiveness model programmed in Microsoft Excel™. Results: The model outcomes predicted by each method (HR, FP and AFT) are presented and compared. Both deterministic and probabilistic results are presented, alongside a discussion around how the uncertainty in these structural assumptions may be captured in EE. Conclusions: Structural assumptions in ES may lead to differences in model outcomes. The impact of these differences may be important in situations where decision uncertainty is high. Methods should be chosen and justified based on patterns of hazard present in the trial data.