Fully semiparametric Bayesian approach for modeling survival data with cure fraction

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Cure rate models are essential for analyzing long-term survival data.
• Traditional models may not adequately capture the complexities of time-to-event data with a cured fraction.
• The piecewise exponential model (PEM) offers flexibility but requires careful handling of its time grid.

Purpose of the Study:

• To develop a comprehensive semiparametric Bayesian cure rate model.
• To incorporate a random time grid within the PEM framework.
• To investigate the impact of this random time grid on cure fraction estimation.

Main Methods:

• A novel Bayesian cure rate model integrating a piecewise exponential model (PEM) with a random time grid.
• Hierarchical modeling approach for prior specification of failure rates.
• Development of an efficient collapsed Gibbs sampler for posterior computation.
• Bayesian goodness-of-fit and model comparison diagnostics.

Main Results:

• The proposed model effectively handles the randomness of the time grid in PEM.
• The random time grid significantly impacts the estimation of the cure fraction.
• The developed Gibbs sampler provides efficient posterior computation.
• The methodology demonstrates utility in analyzing real-world clinical trial data.

Conclusions:

• The novel Bayesian cure rate model with a random time grid PEM offers a powerful tool for survival data analysis.
• This approach provides improved estimation of cure fractions and robust model assessment.
• The methodology is validated through its application to a melanoma clinical trial.