Uncertainty Quantification in Zip Code Tabulation Area-Level Breast Cancer Screening: A Bayesian Geospatial Analysis...

Area of Science:

• Public Health
• Biostatistics
• Geographic Information Systems (GIS)

Background:

• Geographic variation (GV) in breast cancer screening exists in Florida Zip Code Tabulation Areas (ZCTAs).
• Traditional spatial analyses often rely on frequentist point estimates, lacking formal uncertainty representation.
• Small area public health surveillance benefits from robust uncertainty quantification.

Purpose of the Study:

• To analyze geographic variation in breast cancer screening patterns at the ZCTA level in Hillsborough County, Florida.
• To compare frequentist and Bayesian regression approaches for modeling screening data.
• To quantify geographic variation in screening activity probabilistically.

Main Methods:

• A three-stage approach: frequentist Poisson regression, global spatial autocorrelation (GSA) using Moran's I, and Bayesian Poisson/negative binomial regression.
• Bayesian models were estimated using the No-U-Turn Sampler in the brms/Stan framework.
• Spatial analyses were conducted using ArcGIS Pro 3.6.

Main Results:

• The number of breast cancer screenings correlated with the number of females in households across all subgroups.
• No independent correlation was found for median household income, insurance status, or age after adjustment.
• No significant global spatial autocorrelation was detected (Moran's I = 0.003, p = 0.745).
• The Bayesian Poisson model performed best (Bayesian R² = 0.91, RMSE = 5.40).

Conclusions:

• Bayesian uncertainty quantification is valuable for small area public health surveillance.
• A probabilistic framework can be used to quantify geographic variation in screening activity.
• Results reflect screening volume, not standardized participation rates.