Survival Curve Extrapolation in Oncology Drug Evaluations: A Review Of Brazilian Health Technology Assessment Reports

This study described modeling approaches for survival curve extrapolation beyond time periods studied in oncology adopted by the Brazilian National Committee for Health Technology Incorporation (Conitec).

A retrospective analysis of Conitec's recommendation reports for oncology drugs from 2019 to 2024 was conducted. Independent reviewers used a predefined form to extract data on model types, health states, time horizons, discount rates, and survival extrapolation methods. The reporting quality of the health economic models was assessed qualitatively using the Consolidated Health Economic Evaluation Reporting Standards 2022 checklist.

Forty-four reports (43 drugs) addressing 20 types of cancer were identified, representing 69 individual assessments. Partitioned Survival Analysis (PartSA) was the most frequent model (67%), followed by Markov model (26%). Survival curve extrapolation was performed in 90% of the models. In PartSA, Weibull distribution was used in 44% of overall survival extrapolations, whereas Log-normal and Generalized gamma distributions (17% and 9%, respectively) were common for progression-free survival. Furthermore, methodological gaps were prominent: 8% of the models provided no justification for the choice of parametric distributions, and 42% failed to report validation methods for survival curve extrapolations. Regarding reporting quality, significant omissions were found for Consolidated Health Economic Evaluation Reporting Standards 2022 checklist items related to economic analysis plan (68%) and the reporting of results, limitations, and generalizability (41%).

PartSA is commonly adopted in Conitec's oncology drug evaluations, aligning with international Health Technology Assessment practices. However, significant gaps in methodological rigor and transparency were identified, highlighting that the application of these models still faces uncertainty and reporting challenges.