Stable survival extrapolation using mortality projections.

Area of Science:

• Biostatistics
• Health Economics
• Epidemiology

Background:

• Mean survival estimation is vital for health economic evaluations, requiring data extrapolation beyond observed survival curves.
• Current extrapolation methods can lack stability, necessitating the integration of long-term evidence from registries and demographic data.

Purpose of the Study:

• To develop and validate a flexible, interpretable, and robust Bayesian approach for survival extrapolation.
• To apply the proposed methods to estimate mean survival in breast cancer, advanced melanoma, and cardiac arrhythmia.

Main Methods:

• Utilized a Bayesian mortality model to project baseline population data, anchoring the survival model.
• Employed flexible parametric poly-hazard models for extrapolation, accommodating diverse survival curve shapes and non-proportional hazards.
• Applied the approach in a competing risks context for cardiac arrhythmia to assess cause-specific hazard stability.

Main Results:

• Successfully estimated mean survival and related metrics for triple-negative breast cancer, melanoma treated with immunotherapy and mRNA therapeutics, and cardiac arrhythmia.
• Demonstrated the model's ability to handle complex scenarios, including crossing survival curves and competing risks.
• The cause-specific hazard approach in competing risks minimized instability in cardiac arrhythmia analysis.

Conclusions:

• The proposed Bayesian and flexible parametric poly-hazard modeling approach provides a robust and interpretable solution for survival extrapolation.
• This method enhances the reliability of health economic evaluations by improving the accuracy of mean survival estimates.
• The approach is versatile, applicable to various diseases and clinical scenarios requiring long-term survival predictions.