Detecting cancer clusters in a regional population with local cluster tests and Bayesian smoothing methods: a...

Area of Science:

• Epidemiology
• Biostatistics
• Geographic Information Systems (GIS)

Background:

• Rising demand for prospective cancer cluster monitoring.
• Limited evidence on cluster detection test performance with small, heterogeneous sample sizes and varying spatial scales.
• Need for evaluating methods using real-life data from population-based cancer registries.

Purpose of the Study:

• Evaluate different cluster detection methods for identifying lung cancer clusters.
• Assess performance under conditions of small/heterogeneous sample sizes and varying spatial scales.
• Utilize real-life data from a German epidemiological cancer registry.

Main Methods:

• Constructed risk surfaces with relative risks (RR) of 2.0 and 4.0 for lung cancer.
• Sampled lung cancer cases using an inhomogeneous Poisson process.
• Analyzed data at small (census tracts) and large (communities) spatial scales using cluster detection methods and Bayesian smoothing models.

Main Results:

• Local cluster tests showed higher detection rates and lower false-positive rates at moderate risk (RR=2.0) across scales.
• At higher risk (RR=4.0), local tests excelled at small scales, while Bayesian methods had lower false-positive rates at large scales.
• Data aggregation at large scales diluted risk increases, impacting method performance.

Conclusions:

• High-resolution spatial scales are more suitable for cancer cluster testing and monitoring than aggregated scales.
• Suggests a two-stage approach: first-stage screening with high detection rate methods, second-stage with higher predictive ability methods.
• Emphasizes the importance of spatial scale in accurate cancer cluster detection.