Updated: May 22, 2020
Collaborators at Children's Hospital of Philadelphia (CHOP) have uploaded updated county-level projections of incidence and Rt rate around the USA. Check it out at this link
The principal investigator in this work is David Rubin, an epidemiologist at CHOP, and the primary statistical modeler is Jing Huang, an Assistant Professor of Biostatistics in the Division of Biostatistics that I lead here at University of Pennsylvania. I have been participating in the project in a supportive role as a consultant/collaborator providing some ideas and feedback.
Models are very important in understanding the spread of SARS-CoV-2, but each modeling approach has its weakness. The classic SIR or SEIR infectious disease models are based on the mathematical mechanisms underlying infectious disease but assume constant spread rates and homogeneity across society, while the disease spread depends on many factors including population density, social distancing, temperature and humidity, socioeconomic status, and the measured incidence upon which these models are based depend on testing and recording patterns. Pure statistical models are well suited to model these covariates, and handle correlated time series, but many statistical models ignore the mathematics of infectious disease spread.
I really like this modeling approach that seems to be a hybrid of SEIR models and statistical models. It focuses on modeling the realized reproduction rate, Rt, as it varies over time using these important covariate factors. It also models at the county level, which allows it to customize predictions to specific characteristics of each county, which aids making county-level policies and recommendations. It uses autoregressive time series models and time lags to capture the structure in the data, and can be used to project future changes in Rt and incidence.
This model is gaining a high level of visibility -- it has been presented 3 weeks in a row to Deborah Birx, and has been presented at the White House meeting including Deborah Birx, Anthony Fauci, Mike Pence, and a group of governors, and Deborah Birx has been promoting the model to others in the media, and it is being discussed in articles in the New York Times, Washington Post, and numerous other settings, plus David is speaking on CNN on Thursday evenings. This tool is also informing PA governor Wolf and his Health Secretary Levine in helping them make county-level decisions to declare counties as red, yellow or green, various stages of re-opening.
The website contains 3 key tabs: one lists existing incidence and death data, a second shows 4-week projections for R (>1 means incidence is growing, <1 means declining) for >200 counties in the USA, and the third tab shows changes in social distancing over time.
I encourage you to look through the plots -- there's lots of interesting results there. A few things to look for:
Look at Pennsylvania data: do you think some more counties should have lockdowns lifted?
Look at Georgia: how is incidence since opening in late April, and how much has social distancing gone back from the 70% levels during lockdowns, and how are their projections? Check out also Colorado and Missouri who had early lockdowns that did not get as much political attention as Georgia.
Look at the key hot spots: southeastern Florida, Northern Virginia, Baton Rouge Louisiana, and Dallas and Houston areas in Texas.
Texas ... yikes! Be careful friends (the model considers temperature and humidity, but hasn't seen 90's and higher yet -- hopefully for Texas' sake the temperature effect ends up being stronger for those higher temperatures)
Most counties across the country have low incidence and low R -- even some with very low levels of social distancing are OK because of their low incidence and lifestyle that has lots of natural social distancing built in. These places can likely be opened safely.
How does California look? Does it make sense to keep near-total lockdowns in place throughout the summer? They seem to have beaten the incidence back quite well and are probably able to roll back a little.
NY and NJ have had a big surge, but have experienced high enough infection rates that future outbreaks will have trouble taking root here -- i.e. they are further towards herd immunity than any other place and they may be better off in the fall than many other places in USA.
Ongoing model improvements are being added using some incredible intricate data that is available from municipalities and mobile phone carriers and other technology companies. We are living in the big data era and there is a lot of good data to inform models like these!