Thread by @reichlab

In both the near term (the upcoming weeks) and long term (months), models are going to continue to play an important role in setting policy as we persevere in the fight against #COVID19. 1/13

https://twitter.com/cmyeaton/status/1248003777325600770

Ensuring that modelers work together to deliver the best data to #COVID19 decision-makers is critical. With this in mind, my group has started to replicate a collaborative framework that has been successful in predicting seasonal flu outbreaks. 2/
reichlab.io/flusightnetwork

To better coordinate #COVID19 forecasting efforts we have:

(i) built a centralized data repo that standardizes forecast data from many models into a common format,
github.com/reichlab/covid…
(ii) created an ensemble model,
(iii) built an interactive viz
reichlab.io/covid19-foreca…

3/

For now the #COVID19 forecast hub displays 4-week ahead forecasts of cumulative deaths in the US and by state. We have 9 models from 5 teams, with more on the way. We are hoping to add additional predictions on cases and hospitalizations in coming days and weeks. 4/

One of the key motivations here is to emphasize that different prediction models will say different things. For example, look at how two models are saying different things about cumulative #COVID19 deaths in Illinois in the next 4 weeks:
reichlab.io/covid19-foreca…
5/

Some preliminary state-level predictions from the ensemble model (currently only combining 2 models) of cumulative #COVID19 deaths by May 9th:

New York: 19,515 (95% PI: 11,524 - 42,097)
New Jersey: 6,012 (95%PI: 3,289 - 17,459)
Massachusetts: 5,254 (95% PI: 1,738 - 19,042)

6/

These models serve a reminder that there is so much that we don't know right now about the future. And that none of these models on their own are adequate to drive the important #COVID19 policy decisions that we will face in upcoming weeks and months.
7/

https://twitter.com/cmyeaton/status/1251139253293273088

Normally, we would take many more weeks to release all of this to the public, but we don't have the luxury of time right now. We are working hard every day to improve this, add data, make better visualizations. We will likely make mistakes as we go.
8/
xkcd.com/2289/

This work underscores for me the importance of centralized coordination of these efforts, and I've appreciated the guidance and advice from hard-working public health officials as we've spun this up. Time for a national disease forecasting center!
9/

https://twitter.com/B_resnick/status/1250797144002494466

This collaboration builds on the hard work and positive, open-science attitudes of many research groups. It takes bravery to make public predictions right now. Not risk-your-family's-health courage that healthcare workers show everyday (🙏🏽❤️), but bravery nonetheless.
10/

We are going to try our best to stay out of the fray of critiquing specific models, and let the data do the talking from here on out. We are working on building smart adaptive statistical ensemble approaches that will let the data decide how to best combine models.
11/

Our initial "release" has data from teams at @LosAlamosNatLab @Northeastern @MRC_Outbreak @IHME_UW and @Columbia. More teams are on the way! Many thanks to Dave Osthus @sdelvall @alexvespi @sangeeta0312 Chris Murray, Jeff Shaman and their teams for making data available.
12/

Finally, this sprint in the last ~10 days to build all of this would not have been possible without major contributions from Katie House, Nutcha Wattanachit, Andrea Brennen, and @johannesbracher. Thank you all for your hard work!
13/13