B1526
Title: Coupled Markov switching count models for the detection and forecasting of COVID-19 outbreaks in Quebec hospitals
Authors: Alexandra Schmidt - McGill University (Canada) [presenting]
Abstract: COVID-19 has greatly strained the hospital system in Quebec since the first cases emerged in February 2020. We develop a novel Bayesian Markov switching model to understand better the emergence and persistence of COVID-19 outbreaks in the 30 largest Quebec hospitals and to detect/forecast the outbreaks within each hospital. We assume each hospital switches between outbreak and non-outbreak periods through a series of coupled nonhomogeneous hidden Markov chains. We allow the probability of an outbreak emerging, or persisting, in a hospital to depend on space-time covariates, such as lagged COVID-19 test positivity rates and lagged mobility data. We also allow the probability of an outbreak emerging to depend on the outbreak status of other hospitals previously, which allows the outbreaks to spread between hospitals. We assume the effects of outbreak spread can change over space and time to account for differing levels of connectivity in the hospital network. We assume incidence in the endemic period is stable and predictable overtime, following a log-linear negative binomial model with simple seasonal and time trends. During the epidemic period, we assume incidence follows a log-linear autoregressive negative binomial model to allow the cases to rise or fall rapidly.