Infectious Disease
A spatial graph neural network for predicting the 2022 polio and 2025measles outbreaks in the United States Dustin Hill* Dustin Hill Hill Hill Hill Hill Hill Hill Hill Hill Hill Hill Hill Hill Syracuse University
For vaccine-preventable illnesses, identifying where a pathogen may spread next is a key public health concern. We use the Advan dataset of generalized movement patterns to model and predict the spread of polio in New York State in 2022 and measles in the US in 2025. We incorporate social network theory, machine learning, and infectious disease modeling to build a network contagion model called SPARC (spatial prediction analysis of the risk of contagion) based upon the premise that we can predict the location of where an infectious disease might be found next from generalized human movement patterns. We use a machine learning approach built on graph network data over time and across space. Weekly movement patterns between locations provide layers of data for the network. We use previous weeks to predict the visitors from one location to another in the next week to show the next place we might see a disease given an observed initial outbreak. SPARC has over 99 percent specificity in classifying counties as free from poliovirus circulation based on wastewater detection. The results support current monitoring plans that include counties with risk of polio re-emergence while excluding counties of low risk to maximize resources. However, the model only has 45 percent sensitivity in identifying counties with polio detections in wastewater. To improve the prediction of positives, we are including additional parameters like community vulnerability and wastewater detection method sensitivity. For identifying counties with and without measles cases, SPARC has 95 percent specificity but only 28 percent sensitivity. Our next step is to add estimates of vaccine uptake to improve the model’s predictive power. Combining the theory and methods of social networks, machine learning, and epidemiology shows that this interdisciplinary approach is relevant for building tools that harness the power of public health surveillance to proactively prepare for disease outbreaks.
