Power evaluation of disease clustering tests

Area of Science:

• Spatial statistics
• Epidemiology
• Biostatistics

Background:

• Numerous statistical tests exist for detecting spatial clustering.
• These methods are crucial in various applications, including disease surveillance.
• A comprehensive comparison of their statistical power under diverse clustering models is needed.

Purpose of the Study:

• To evaluate and compare the statistical power of eight different disease clustering tests.
• To identify which tests are most effective for detecting various types of spatial clustering.

Main Methods:

• Utilized a large dataset of 1,220,000 simulated benchmark data points.
• Data were generated under 51 distinct spatial clustering models.
• Compared the power of Besag-Newell's R, Cuzick-Edwards' k-Nearest Neighbors (k-NN), spatial scan statistic, Tango's Maximized Excess Events Test (MEET), Swartz' entropy test, Whittemore's test, and Moran's I (including a modification).

Main Results:

• Most tests demonstrated good power, with exceptions for Moran's I and Whittemore's test.
• The spatial scan statistic excelled at detecting localized clusters.
• Tango's MEET proved effective for identifying global clustering patterns.
• Besag-Newell's R and Cuzick-Edwards' k-NN showed good performance with optimized parameters.

Conclusions:

• Statistical power significantly differs among tests and clustering models.
• The choice of test statistic is critical and depends on the expected clustering pattern.
• Careful consideration of test power is essential for accurate spatial clustering analysis.