The Extrapolation Performance of Survival Models for Data With a Cure Fraction: A Simulation Study

Area of Science:

• Biostatistics
• Survival Analysis
• Epidemiology

Background:

• Curative treatments can lead to complex hazard functions, challenging standard survival models.
• Poor extrapolation performance is a risk with standard models when cure fractions are present.

Purpose of the Study:

• To compare the extrapolation performance of various survival models with and without cure fractions.
• To assess model performance when fit to data exhibiting a cure fraction.

Main Methods:

• Simulated data from a Weibull cure model across 9 scenarios (follow-up length, sample size).
• Evaluated cure and non-cure versions of parametric, Royston-Parmar, and dynamic survival models.
• Assessed mean-squared error and bias in hazard function estimates.

Main Results:

• Cure models showed improved extrapolation with longer follow-up; flexible cure models performed well.
• Models without a cure fraction yielded significantly worse extrapolations.
• Accurate cure fraction estimation was not essential for precise hazard estimates.

Conclusions:

• Failure to model cure fractions severely impacts extrapolation accuracy for curative treatments.
• Cure models enhance extrapolation, but model choice requires clinical judgment with limited data.
• Dynamic cure models demonstrated robustness to misspecification, unlike standard parametric cure models.