Cure fraction estimation from the mixture cure models for grouped survival data

Area of Science:

• Biostatistics
• Survival Analysis
• Epidemiology

Background:

• Mixture cure models analyze failure time data with long-term survivors, estimating cure proportions and latency distributions.
• Identifiability of cure fraction and kernel distribution parameters is a critical challenge in these models.
• Cure fraction estimates are sensitive to latency distribution choices and follow-up duration.

Purpose of the Study:

• To investigate the sensitivity of parameter estimates in mixture cure models under various parametric and semi-parametric settings.
• To evaluate the robustness of cure fraction estimation using different latency distributions, including lognormal, loglogistic, Weibull, and generalized Gamma.
• To assess the impact of follow-up time and latency distribution specification on cure fraction estimation through simulation.

Main Methods:

• Exploration of parameter estimate sensitivity for semi-parametric and parametric mixture cure models (lognormal, loglogistic, Weibull, generalized Gamma).
• Conducting a simulation study to examine the effects of follow-up time and latency distribution choice on cure fraction estimation.
• Application of mixture cure models with generalized Gamma latency distributions to population-based cancer survival data from the SEER Program.

Main Results:

• The generalized Gamma distribution demonstrated robust cure fraction estimates compared to other tested distributions.
• Simulation results highlighted the significant impact of follow-up time and latency distribution specification on cure fraction estimation.
• Analysis of SEER data using the generalized Gamma distribution provided insights into cancer survival patterns.

Conclusions:

• The generalized Gamma distribution is recommended for modeling latency in mixture cure models due to its robustness.
• Careful consideration of follow-up time and latency distribution is essential for accurate cure fraction estimation.
• The study advises caution in the general application of mixture cure models, emphasizing the importance of model selection and validation.