A Comparative Analysis Between Three Spatio-Temporal Scan Statistics for Outbreak Detection and Antimicrobial Resistance

Abstract

Antimicrobial resistance is a growing problem throughout the world. Government agencies and other global health organizations have created software for tracking emerging microbial threats on a facility, state-wide and international scope. Epidemiological software uses statistical algorithms to discern between aberrations of resistance in a given location and time, and whether these “clusters” can be attributed to a specific cause versus chance alone. The retrospective spatiotemporal scan statistic software, SaTScan, has diverse applications. In this project we demonstrate the use of SaTScan version 9.8 within the free epidemiological software WHONET. Microbiology data collected from 2000-2006 at Brigham and Women’s Hospital was used for retrospective analysis. Three algorithms: the space-time continuous uniform, the space-time discrete Poisson and the space-time permutation were run on the same data set with limited parameter adjustments per model. Three variables were considered within each model: maximum spatial size within the population at risk, baseline days, maximum temporal size and number of Monte Carlo simulations. As demonstrated by this study, we found that there was a significant difference in cluster detection between the three probability models. Furthermore, we demonstrate the effects of varying values when defining the spatiotemporal scanning window. The findings of this study allow epidemiologists and computational biologists alike in better understanding the parameters of these algorithms, and create a more precise narrative of outbreaks in antimicrobial resistant clusters. Further prospective validation of these results would give a better understanding of the accuracy and clinical importance for the clusters found. This could be utilized world-wide for other WHONET users, and can guide public health policies in the event of an outbreak or epidemic.