Reliable event rates for disease mapping

Area of Science:

• Spatial statistics
• Biostatistics
• Geographic Information Systems (GIS)

Background:

• Defining reliable estimates for spatially referenced event data is challenging, especially in small area analysis.
• Existing spatial models can lead to oversmoothing, diminishing the data's informational content.
• Crude estimates often lack sufficient reliability for small area settings.

Purpose of the Study:

• To define a unified criterion for "reliable" event rate estimates applicable to both crude and model-based approaches.
• To develop a spatial Bayesian framework that enhances estimate reliability while mitigating oversmoothing.
• To provide a method for focusing small area studies on areas with sufficient data for robust inference.

Main Methods:

• Developed a novel definition for "reliable" statistical estimates, allowing for discrete and continuous reliability statements.
• Constructed a spatial Bayesian framework incorporating prior information to enhance reliability.
• Applied the framework to county-level birth data in Pennsylvania to demonstrate its efficacy.

Main Results:

• The proposed definition of reliability is applicable to crude and model-based estimates.
• The spatial Bayesian framework effectively improves estimate reliability and guards against oversmoothing.
• Analysis of Pennsylvania birth data illustrated the impact of oversmoothing and the benefits of the new approach.

Conclusions:

• The developed definition and framework offer a robust approach to small area statistical analysis.
• This methodology allows researchers to better identify areas where data supports reliable inferential decisions.
• The definition of reliability can inform the design of future small area studies.