Diseases like H1N1 can be prevented from becoming a wide spread epidemic through timely detection and containment measures. Similarity of H1N1 symptoms to any common flu and its alarming rate of spread through animals and humans complicate the deployment of such strategies. We use dynamic implementation of graph matching methods to overcome these challenges. Specifically, we formulate a mixed integer programming model (MIP) that analyzes patient symptom data available at hospitals to generate patient graph match scores. Successful matches are then used to update counters that generate alerts to the Public Health Department when the counters surpass the threshold values. Since multiple factors like age, health status, etc., influence vulnerability of exposed population and severity of those already infected, a heuristic that dynamically updates patient graph match scores based on the values of these factors is developed. To better understand the gravity of the situation at hand and achieve timely containment, the rate of infection and size of infected population in a specific region needs to be estimated. To this effect, we propose an algorithm that clusters the hospitals in a region based on the population they serve. Hospitals grouped together affect counters that are local to the population they serve. Analysis of graph match scores and counter values specific to the cluster helps identify the region that needs containment attention and determine the size and severity of infection in that region. We demonstrate the application of our models via a case study on emergency department patients arriving at hospitals in Buffalo, NY.
Available at: http://works.bepress.com/jomonaliyas_paul/19/