Utilizing Satscan and the Space-Time Permutation Scan Statistic to Detect Spatial and Temporal Clustering Based on Disease Registry Information

BACKGROUND:  The space-time permutation scan statistic allows investigators to mathematically investigate spatial and temporal clustering of a disease when only case information is available. This tool can be used in tandem with public health surveillance systems, such as the Extensively Drug Resistant Organism (XDRO) registry implemented by the Illinois Department of Public Health, to further understand disease epidemiology. Using SaTScan software to generate scan statistics, we were able to identify potential disease clustering as well as generate output for graphic visualization using geographic information software (GIS).   

METHODS:  A file containing location identifiers and geographic coordinates for 545 Illinois healthcare facilities participating in the XDRO registry was inputted into SaTScan along with a file containing the location and date of 766 XDRO cases from November 1, 2013 to December 28, 2014. We programmed SaTScan software to aggregate time in 7-day increments to minimize computing time and adjust for day-of-week effects. The maximum spatial cluster was set at a 50 km radius and then adjusted incrementally. The minimum temporal cluster was 1 day with maxima of 30 and 90 days based on previous epidemiologic knowledge about these types of organisms. The SaTScan output generated a shape file for Arc-GIS, which, when overlaid with a map of Illinois marked with participating facilities, allowed for a visual representation of the potential disease clusters across the state.

RESULTS:   A statistically significant cluster with a radius of 10.13 km ranging over a 2-week period was detected using the 30-day temporal window (scan statistic = 10.15, p=0.019). Out of the 10 participating facilities encompassed by this region, the 10 total cases were detected at 3 unique locations exceeding the 1.57 expected cases over that span as calculated by SaTScan. A secondary cluster at a single location was detected over a 6-day interval (scan statistic = 9.217, p=0.057) in both the 30- and 90-day permutations with 5 cases reported exceeding the expected 0.31 cases.

CONCLUSIONS:   This case study demonstrates the potential of using SaTScan software in conjunction with GIS mapping to identify and display disease clusters, which in turn inform outbreak response and intervention planning. This process can be automated and run periodically to provide the most current information about the significance of possible disease clustering.