A semiparametric cluster detection method--a comprehensive power comparison with Kulldorff's method.

Area of Science:

• Biostatistics
• Spatial Analysis
• Statistical Computing

Background:

• Cluster detection methods often require prior knowledge of data distributions or case counts.
• Moving window techniques are common but can be limited by fixed sizes and assumptions.
• Overlapping scanning windows introduce multiple testing problems, necessitating robust statistical controls.

Purpose of the Study:

• To introduce a semiparametric density ratio method for cluster detection that borrows strength across samples.
• To integrate this method with Storey's q-value for controlling the false discovery rate (FDR).
• To evaluate the performance of the proposed method against existing techniques, particularly Kulldorff's method.

Main Methods:

• A semiparametric density ratio approach applied to variable-sized moving windows for cluster detection.
• Incorporation of Storey's q-value to manage multiple testing issues arising from overlapping scan windows.
• Simulation studies using benchmark datasets (e.g., Kulldorff's Northeastern data) to assess performance.

Main Results:

• The semiparametric method demonstrates comparable performance to Kulldorff's method for binary data.
• For non-binary data, the semiparametric method maintains power without requiring specific probability models.
• Kulldorff's method's power is contingent on selecting the correct probability model, risking power loss otherwise.

Conclusions:

• The proposed semiparametric method achieves robust power for localized cluster detection.
• It avoids stringent distributional assumptions, relying only on a tilt function.
• The methodology is adaptable to various scan schemes, including elliptic and flexible shape scans.