Read on Twitter
Can we forecast the near future of the pandemic with data we have now? That’s the goal of this work led by @MauSantillana. The idea is simple;
in pandemic activity will lead to changes in multiple data streams before increases in cases become apparent https://www.nytimes.com/2020/07/02/health/santillana-coronavirus-model-forecast.html 1/n
in pandemic activity will lead to changes in multiple data streams before increases in cases become apparent https://www.nytimes.com/2020/07/02/health/santillana-coronavirus-model-forecast.html 1/n
This is in part because there are a lot of mild cases, and it takes more time to get people into their primary care doc and tested than it does to ask whether more people are googling symptoms. I think of it as like a disturbance in the force. 2/n
This sort of work has a mixed history. Google flu trends tried to project the near future of influenza based on search patterns right now (eg how many people are googling 'fever'?). It looked great at first but then failed in epic fashion. https://www.wired.com/2015/10/can-learn-epic-failure-google-flu-trends/ 3/n
So why is this different? Well the crucial issue is that if you’re using data from the past to forecast the future, it is *really* hard when you know the future is likely to be different from the past. The guide offered by your training dataset is only so reliable 4/n
A way to handle this is to look at data from many different independent sources. That way if one is out of whack and needs to be recalibrated, it can be corrected. So this uses a *lot* of different data sources 5/n
The whole thing can be read here - note this is a preprint. Reporters beware and seek comment from other scientists. https://arxiv.org/abs/2007.00756 6/n
Based on this algorithm, States to watch based on recent signals include NE and NH. Both are starting from a low level of activity but current data suggest an increase is coming in next few weeks. Meanwhile places like RI and MD are expected to remain flat 7/n
This is going to be a long fight, and lives will be saved by being nimble and reacting to local fluctuations in disease in the coming months. I hope work like this can help. It is certainly the sort of information I would like to have when making decisions 8/n
And I am so tired that I forgot to end that thread properly.
This is now the end of the thread. Bye
9/end
This is now the end of the thread. Bye
9/end
