A Bayesian piecewise survival cure rate model for spatially clustered data.

Area of Science:

• Biostatistics
• Spatial Epidemiology
• Survival Analysis

Background:

• Time-to-event data often exhibits spatial clustering, impacting survival and cure rate estimations.
• Traditional survival models may not adequately capture complex spatial dependencies in disease data.
• Cure rate models are essential for understanding long-term outcomes in diseases with potential for complete recovery.

Purpose of the Study:

• To develop a Bayesian hierarchical cure rate survival model for spatially clustered time-to-event data.
• To incorporate spatial correlation structures using Multivariate Conditionally Autoregressive (MCAR) frailties.
• To apply the methodology to Hodgkin lymphoma cancer survival data in Connecticut.

Main Methods:

• A mixture cure rate model with covariates and a flexible baseline survival distribution for uncured individuals.
• Introduction of spatial correlation via regional frailties following an MCAR distribution on a map.
• Estimation of posterior parameters and smoothed regional maps of spatial frailties and cure rates.

Main Results:

• Simulation studies show that models with spatially correlated frailties yield smaller relative biases and Mean Squared Error (MSE) compared to simple frailty models.
• The methodology effectively smooths estimates of spatial frailties and cure rates at a regional level.
• Application to Hodgkin lymphoma data provides insights into survival patterns influenced by spatial factors.

Conclusions:

• The proposed Bayesian hierarchical cure rate model effectively accounts for spatial clustering in survival data.
• Incorporating spatial correlation through MCAR frailties enhances the precision of parameter estimates.
• This approach offers a valuable tool for analyzing cancer survival data with spatial dependencies, as demonstrated with Hodgkin lymphoma.