Real-Time Estimation and Forecasting of COVID-19 Cases and Hospitalizations in Wisconsin HERC Regions for Public Health Decision Making Processes

PURPOSE: The spread of the COVID-19 and surging number of cases have resulted in overtaxed healthcare systems. However, limited availability and questionable reliability of the data make outbreak prediction and resource planning difficult. Moreover, any estimates or forecasts are subject to high uncertainty and low accuracy when measuring such components. The aim of the study is to apply, automate, and assess a workflow for the real-time estimation and forecasting of COVID-19 cases and hospitalizations in Wisconsin HERC regions. METHODS & MATERIALS: The reported cases are corrected for under-reporting and adjusted for test positivity by date of report. The corrected cases are estimated by date of infection, forecasted into the future, and transformed to date of report by region over time using a Bayesian latent variable model. The cases are corrected for hospitalization delay using log-normal distribution, and hospitalizations are estimated by county over time using Bayesian regression model. Models will be automated for real-time estimation and forecasting via RStudio Connect and made available as an R package. RESULTS: Both models were fitted weekly and forecasted over a 1-day or 3-day period during the peak of the epidemic from September 20, 2020 to December 6, 2020. For cases, both scenarios outperformed the credible level of the forecast where the 3-day period (20% CrI: 0.468, 50% CrI: 0.810, 90% CrI: 1.000) performed slightly better than the 1-day period (20% CrI: 0.462, 50% CrI: 0.785, 90% CrI: 1.000). Similarly, for hospitalizations, both periods outperformed the credible level of the forecast where the 3-day period (20% CrI: 0.368, 50% CrI: 0.667, 90% CrI: 0.987) performed slightly better than the 1-day period (20% CrI: 0.358, 50% CrI: 0.653, 90% CrI: 0.987). CONCLUSION: We present an approach to estimate and forecast cases and hospitalizations and the corresponding uncertainty using publicly available data. The models were able to infer short-term trends consistent with reported values at the HERC region level. Models were able to accurately forecast and estimate the uncertainty of the measurements. This study can help to elucidate which regions are most affected and which regions will encounter outbreaks as well as support decision making processes.