Contact: Eric Marty ([email protected])

Contributors:
John M. Drake ([email protected])
Éric Marty ([email protected]) - visualization, nowcast architecture
Rachel Mercaldo ([email protected]) - forecasting
Austin M. Smith ([email protected])

• analysis
  Andrew M. Kramer ([email protected])
  Tim Wildauer ([email protected])
• deconvolution

• Code for functions is in the R/ folder.
• High level scripts or Rmd documents are found at the top level.
• Temporary data products are in the /data folder.
• Nowcast outputs are in the /data/output folder.
• Source data for US nowcasts are maintained at https://raw.githubusercontent.com/CEIDatUGA/COVID-19-DATA/master/US/US_wikipedia_cases_fatalities/

Estimate the current size of the epidemic.

A key problem in making management decisions is estimating the size of the epidemic. This project aims to estimate the size of the unknown epidemic. Case notifications are a poor indicator of epidemic size for several reasons.

• Case notifications are incomplete (there is under-reporting).
• Case notifications are primarily associated with patient isolation and therefore are not contributing greatly to transmission.
• SARS-CoV-2 has a significant incubation period. These presymptomatic cases are also part of the epidemic.

This project will use a non-parametric approach to deconvolving the case notification record to construct the actual size of the epidemic as it existed at past times (backcasting) and then use the backcasted estimates to feed a forecasting model that “predicts” the present time (nowcasting). Our nonparametric approach derives from Tim’s REU project. Backcasting will proceed in two steps using individual-level observations for the wait time distributions: (1) construct estimated curve of patients with symptom onset; (2) construct estimated curve of patients with active infection. From these curves we will use a statistical model (perhaps time-varying autoregressive model) to predict the epidemic size at the current time.