We started modeling COVID-19 in March to predict the surge in hospital patients. We've tried to expand our model to answer many questions. 
We produce a reference forecast; what we think is most likely to happen. We use the modeling framework to address a long list of alternative scenarios.
We've gone through 3 revisions of our model in the past couple months.
How does our new RKCS-SEIR model work?
If you just look at cases, you see explosive or steady growth as we test more people. This random knot combination spline algorithm reduces testing bias.
Deaths are sensitive to testing data. We have found in examining places with cause of death reporting that there is substantial undercounting in Peru and Ecuador.
We're estimating results for 400 geographic units. For each of them we estimate 1000 models.
How are we predicting beta(t)?
We recently got access to two surveys that @Facebook uses on mask usage; 1 million people in the US and 1 million people outside the US. It is an extremely powerful sample size.
The seasonality pattern of pneumonia is the most powerful predictor of transmission thus far. We should see reduced transmission through August, and then the reverse afterward.
Testing capacity is highly variable throughout the world.
We're using cellphone mobility data from four providers to help us track the spread of the disease.
There is a global rise in mobility as social distancing mandates lift worldwide.
Mandates were implemented largely at the same time worldwide.
Here is our forecast of cumulative COVID-19 deaths for every region in the world. We anticipate seeing large expansions into the millions of deaths by October.
We are seeing substantial upticks starting in mid-July for North Africa and the Middle East.
This is a tool that can help health providers plan for healthcare surge.
What's coming next?
Here is a comparison of national sero-prevalence studies against our model predictions. This data is not currently used in our model.
An external group has been evaluating the performance of various COVID-19 models over time.
What's happening around the world? There's been low disease transmission in South Asia and Africa, initially due to worldwide mandates, but we anticipate transmission to pickup in these places as mandates lift.
Thank you for following along and to @ghi_aub for hosting this webinar!
For answers to frequently asked questions, check out our #COVID19 Model's FAQ Blog: healthdata.org/covid/faqs
And finally, for details on #COVID19 model updates, we recommend bookmarking our Updates blog: healthdata.org/covid/updates
Correction: These are preliminary results. Our team is hard at work on these estimates and will publish final results when they are ready.
