Spatial cluster detection for ordinal outcome data.

Area of Science:

• Epidemiology
• Biostatistics
• Geographical Information Systems (GIS)

Background:

• Spatial scan statistics are crucial for identifying disease clusters in geographical surveillance.
• Existing methods for ordinal data, like cancer stage, use strict likelihood ratio ordering, potentially missing clusters with worse outcomes.
• This limitation can hinder accurate identification of high-risk areas.

Purpose of the Study:

• To introduce and evaluate a novel spatial scan statistic using stochastic ordering for ordinal disease data.
• To compare the performance of the stochastic ordering approach against the traditional likelihood ratio ordering.
• To assess the effectiveness in detecting spatial clusters with high rates of more severe disease stages.

Main Methods:

• Developed a spatial scan statistic incorporating stochastic ordering for ordinal data.
• Conducted simulation studies to compare the new method with the likelihood ratio ordering approach.
• Applied both methods to a real-world disease data example for illustration.

Main Results:

• The stochastic ordering-based approach demonstrated higher statistical power compared to the likelihood ratio ordering.
• The new method showed improved sensitivity and positive predictive value across various simulation scenarios.
• Simulation results indicate a superior ability to detect spatial clusters with adverse disease outcomes.

Conclusions:

• Stochastic ordering offers a more flexible and powerful approach for spatial scan statistics with ordinal disease data.
• This enhanced method can improve the accuracy of geographical disease surveillance, particularly for identifying areas with severe disease stages.
• The findings suggest broader applicability in public health for targeted interventions.