Hierarchical modeling and other spatial analyses in prostate cancer incidence data

Area of Science:

• Epidemiology
• Biostatistics
• Geographic Information Systems (GIS)

Background:

• State cancer registries face challenges in spatial analysis due to evolving methods and technology.
• A methodological approach is described to assist registries with county-level spatial analyses.
• This study addresses the need for accessible spatial analysis techniques in cancer surveillance.

Purpose of the Study:

• To describe a general methodological approach for county-level cancer registry spatial analyses.
• To explore spatio-temporal patterns of prostate cancer incidence in Louisiana.
• To identify geographic clusters of elevated and lower prostate cancer rates by race.

Main Methods:

• Utilized prostate cancer data from the Louisiana Tumor Registry (1988-1999).
• Analyzed data across four 3-year time periods, calculating race-specific incidence rates and standardized incidence ratios (SIRs).
• Employed Bayesian smoothing, spatial autocorrelation tests (Moran's I), hierarchical generalized linear models (HGLM), and spatial scan statistics (SaTScan).

Main Results:

• Temporal trends in SIRs suggest an impact of prostate-specific antigen (PSA) testing, with a potential lag in Black males.
• Identified significant clusters of lower-than-expected prostate cancer rates in central and coastal areas for both white and Black males.
• Clusters of lower rates were more pronounced in Black males in central, southwestern, and southeastern coastal parishes.

Conclusions:

• Mapping disease occurrence over time effectively reveals spatio-temporal patterns.
• Hierarchical generalized linear models (HGLM) can control for covariates and spatial variability.
• The described methods provide a framework for registries to conduct spatial analyses and understand disease distribution.