A binary-based approach for detecting irregularly shaped clusters.

Area of Science:

• Spatial statistics
• Geographic information systems
• Epidemiology

Background:

• Spatial cluster detection is crucial for identifying disease hotspots and environmental risks.
• Existing methods often struggle with irregular cluster shapes or demand extensive computation.
• There is a need for efficient algorithms capable of detecting diverse cluster geometries.

Purpose of the Study:

• To introduce a novel spatial detection algorithm for lattice data.
• To evaluate the proposed method's performance against established scan statistics.
• To assess the algorithm's efficacy in detecting both circular and non-circular clusters.

Main Methods:

• A two-stage approach: Choynowski's test for individual cell significance, followed by a binomial approximation for cluster identification.
• Computer simulations were employed to compare the proposed method with scan statistics.
• Real-world application using Taiwan Cancer data from 2000.

Main Results:

• The proposed method is effective for large populations and irregular regions.
• Scan statistics generally exhibit higher power, particularly with smaller populations.
• Scan statistics identified more clusters, often circular/elliptical, while the proposed method found fewer, including non-circular ones.

Conclusions:

• The proposed method accurately detects circular and non-circular clusters when initial cell significance is correctly identified.
• The binomial-based approach effectively addresses multiple testing issues and reduces computational time.
• Scan statistics offer strong cluster detection power but may over-identify clusters and show lower accuracy for non-circular shapes.