Clustering of rare events

Area of Science:

• Epidemiology
• Biostatistics
• Spatial Analysis

Background:

• Rare disease clustering requires robust statistical methods.
• Understanding spatial disease patterns is crucial for public health.
• Poisson distribution is a common model for rare event counts.

Purpose of the Study:

• To develop statistical criteria for identifying homogeneous disease rate clusters.
• To model rare disease occurrence using a mixture of Poisson distributions.
• To evaluate methods for separating units into risk groups.

Main Methods:

• Modeling unit event counts with a mixture of Poisson distributions.
• Employing maximum likelihood and Bayes approaches for cluster identification.
• Utilizing a likelihood ratio test for mixture significance.
• Applying combinatorial methods to assess contiguous high-risk areas.

Main Results:

• The proposed mixture model effectively identifies homogeneous disease rate clusters.
• A likelihood ratio test can determine the significance of a two-component mixture.
• Results challenge the sole reliance on ordered rates for risk stratification.
• Combinatorial tests assess the significance of spatial disease clustering.

Conclusions:

• Statistical clustering methods provide a more nuanced approach to disease risk assessment than simple rate ordering.
• The developed methods are applicable to spatial epidemiology of rare diseases.
• Further investigation into spatial patterns of sudden infant deaths (SIDs) is warranted.