A multi-level spatial clustering algorithm for detection of disease outbreaks

In this paper, we proposed a Multi-level Spatial Clustering (MSC) algorithm for rapid detection of emerging disease outbreaks prospectively. We used the semi-synthetic data for algorithm evaluation. We applied BARD algorithm [1] to generate outbreak counts for simulation of aerosol release of Anthrax. We compared MSC with two spatial clustering algorithms: Kulldorff's spatial scan statistic [2] and Bayesian spatial scan statistic [3]. The evaluation results showed that the areas under ROC had no significant difference among the three algorithms, so did the areas under AMOC. MSC demonstrated significant computational efficiency (100 + times faster) and higher PPV. However, MSC showed 2-6 hours delay on average for outbreak detection when the false alarm rate was lower than 1 false alarm per 4 weeks. We concluded that the MSC algorithm is computationally efficient and it is able to provide more precise and compact clusters in a timely manner while keeping high detection accuracy (cluster sensitivity) and low false alarm rates.