Parallel tempering strategies for model-based landmark detection on shapes

Shape-restricted estimation and spatial clustering of COVID-19 infection rate curves

The study of regional COVID-19 daily reported cases is used to understand pattern of spread and disease progression over time. These data are challenging to model due to noise that is present, which arises from failures in reporting, false positive tests, etc., and the spatial dependence between regions. In this work, we extend a recently developed Bayesian modeling framework for inference of functional data to jointly estimate and cluster daily reported cases data from US states, while accounting for spatial dependence between US states. Shape-restriction allows us to directly infer the number of extrema of a smooth infection rate curve that underlies noisy data. Other parameters in the model account for the relative timing of extrema, and the magnitude and severity of infection rates. We incorporate mobility behavior of each US state's population into an informative prior model to account for the spatial dependence between US states. Our model corroborates past work that shows that different US states have indeed experienced COVID-19 differently, but that there are regional patterns within the US. The modeling results can be used to assess severity of infection in individual US states and trends of neighboring US states to aid pandemic planning. Retrospectively, this model can be used to see which factors (governmental, behavioral, etc.) are associated with the varying shapes of infection rate curves, which is left as future work.