survextrap: a package for flexible and transparent survival extrapolation

Area of Science:

• Biostatistics
• Health Economics
• Survival Analysis

Background:

• Health policy decisions often rely on long-term survival estimates derived from limited short-term data.
• Existing methods for incorporating longer-term survival information lack a comprehensive and accessible implementation tool.

Purpose of the Study:

• To introduce a novel Bayesian parametric survival model and an associated R software package (survextrap).
• To provide a tool for estimating long-term survival by combining diverse data sources, including short-term trial data and longer-term registry or elicited data.

Main Methods:

• Developed a Bayesian parametric survival model using M-splines to flexibly model the hazard function.
• Integrated individual-level, right-censored data with summary survival data from one or more time periods.
• Implemented the model in the R package 'survextrap', allowing standard R survival modeling syntax.

Main Results:

• The model automatically adapts to available data, acknowledging uncertainty where data are weak, ensuring confident long-term estimates only with strong long-term data.
• Accommodates various survival mechanisms, including cure models, additive hazards, and waning treatment effects.
• The 'survextrap' package offers flexible modeling of proportional and non-proportional hazards.

Conclusions:

• The 'survextrap' package provides a comprehensive and user-friendly tool for advanced survival extrapolation.
• Facilitates principled and robust long-term survival estimation crucial for health technology assessments and policy decisions.