Read on Twitter
1. On Friday night I posted a rapid post-publication peer-review of the @IHME_UW model and associated white paper. Here's the IHME website and projections: https://covid19.healthdata.org/projections
2. In my thread, below, I expressed concern about whether noisy, possibly under-reported data on deaths this early-on in time are adequate to predict the timing and height of the peaks. I remain concerned but uncertain of how big of a problem this is. https://twitter.com/CT_Bergstrom/status/1243819232950751233
3. A few days later, my biggest concern remains around the framing of the model. This is a model of successful suppression of the epidemic. Moreover if I understand correctly, it assumes strong measures by all states with efficacy (if not severity) comparable to Wuhan.
4. The assumption underlying this model is that social distancing will be implemented and *will work*.
That makes this a best-case scenario model, in contrast with e.g. the first Imperial College model wherein suppression fails and 1.1 to 2.2 million Americans die.
That makes this a best-case scenario model, in contrast with e.g. the first Imperial College model wherein suppression fails and 1.1 to 2.2 million Americans die.
5. My grave concern is that this distinction has is not being made in the news reports about the model, and that it is lost on or ignored by politicians eager to for a more optimistic forecast than Imperial College provided.
6. The shaded regions are being interpreted almost everywhere as spanning the best and worst case scenarios.
I believe they are instead something like uncertainty ranges on the best-case scenario of successful suppression.
I believe they are instead something like uncertainty ranges on the best-case scenario of successful suppression.
7. If I am correct, this is an absolutely critical message to get right. Government leaders and public health authorities cannot afford to plan for nothing worse than the upper bound of the best case, if the worst case is still a realistic possiblity—and sadly I believe it to be.
8. I am calling on the @IHME_UW team to address my questions about this issue. I want to be certain I'm correctly representing the model when speaking to the media about this, and I would welcome their clarifications to the broader community using the model for planning purposes.
9. My DMs are open, or please feel free to email me at my @UW email address.
i. Addendum: One way to think about the IHME model versus e.g. the initial Imperial College model is that they represent very different two very different storylines for how things unfold in coming weeks.
ii. In the IHME model, we're on the good timeline. Social distancing works. The pandemic is brought under control with ~3% of the US infected. Deaths are still high—five to low six figures. But it could have been a lot, lot worse.
iii. In the first Imperial College model, we're on the bad timeline. At best, social distancing helps slow the pandemic, but doesn't halt it. Over a million Americans die.
iv. It's very important to model both scenarios. No complaints there.
I hope and pray we end up on the IHME storyline.
I just want to make sure that we are prepared for the dreadful possibility that don't.
I hope and pray we end up on the IHME storyline.
I just want to make sure that we are prepared for the dreadful possibility that don't.
