A comparison of spatial scan methods for cluster detection

Spatial scan methods are extremely popular for identifying disease clusters using disease count data. The original circular scan method proposed by Kulldorff [A spatial scan statistic. Comm Statist Theory Methods. 1997;26(6):1481-1496] is simple to implement, is computationally inexpensive to apply, and has high power for detecting circular clusters; however, it can struggle to identify non-circular clusters. Many extensions of the original method have been proposed to better detect irregularly-shaped clusters. We briefly describe several popular spatial scan method extensions (e.g. Upper Level Set, Flexibly-shaped, Dynamic Minimum Spanning Tree, Fast Subset, etc.). We then compare the performance of the various methods using power, sensitivity, positive predictive value, and overall accuracy by applying these methods to 126 publicly-available benchmark data sets based on 46 different cluster shapes. The comparisons go into more depth and include more methods than any previous studies of this topic; many of the methods have never been directly compared. The comprehensiveness of our study allows us to draw reliable conclusions and make concrete recommendations about the best performing methods. R packages and scripts are provided to make results reproducible.

Automated summary

This study describes several extensions of the spatial scan method and compares their performance using 126 benchmark datasets. The comprehensive nature of the study allows for reliable conclusions and concrete recommendations for detecting disease clusters.