Thread by @joshuasweitz

Revisiting scenarios from last week, I am sharing a few thoughts on #COVID19 case reports via @GaDPH.

The image tells part of the story.

But wanted to explain why, irrespective of the next 8-day trends, @GovKemp should extend the shelter-in-place order due to expire on 4/13.

First, there are many uncertainties with respect to the spread of #COVID19, particularly from a quantitative perspective.

That means, it is possible to fit a local trend with different underlying rates of transmission, age-stratified risk, and levels of ascertainment.

2/15

Second, because of the time between exposure, hospitalization, and death, any change in transmission (e.g., as a result of a shelter-in-place order) will take a while to appear as outcomes.

This is both important for predictions and messaging.

3/15

In developing these scenarios, I used an assumption that social distancing began to have a statewide effect on 3/21, in the other, I assume it had an effect on 3/28.

The actual uncertainty extends beyond these scenarios.

4/15

Either way, the near-term predictions are picking up evidence from the observed exponential growth rate of deaths (and on hospitalizations). As such, the near-term models coincide, but diverge as the accumulated impact of social distancing take effect.

5/15

But the labels also matter.

The model includes “mild/asymptomatic” and “symptomatic” cases. Hence, although the deaths are directly compared, the plots re-scale both hospitalization and reported cases to coincide with the last date of the fitting period (3/22-3/28).

6/15

The re-scaling also tells a story about the unobserved state of the epidemic.

Model fits suggest that there are 33% fewer hospitalizations as expected (and this could reflect many things, including decisions not to go to a hospital, lack of options, model error, etc.).

7/15

The model suggests ascertainment levels of ~4%, i.e., with 5000 lab-confirmed cases, there might >100,000 cumulative cases in GA (many recovered, perhaps unknowingly).

This level of ascertainment depends on model assumptions (e.g., mild/asymptomatic fraction).

8/15

Irrespective of assumptions, low levels of ascertainment suggest that population #serology surveys are critical to understand the state of the epidemic, whether in Metro Atlanta, Dougherty County, or the state as a whole.

9/15

There are other latent predictions of the model, e.g., >90% of fatalities should be amongst those 60 and older (reality is ~80%). These differences may reveal issues of latent comorbidities, mixing patterns, and may help adapt the @neil_ferguson model to US states.

10/15

Indeed, we are in the process of expanding our efforts to try and ‘learn’ more about local impacts, while keeping in mind that long-term quantitative predictions can rapidly diverge from reality (another side effect of modeling exponential processes).

11/15

Finally, it may turn out that social distancing, particularly decisions by @KeishaBottoms to implement shelter-at-place orders earlier than the State, may reduce lab-confirmed cases, hospitalizations, and deaths this week.

That would be a small bit of positive news.

12/15

But, if that happens, then the prudent decision at both city and state levels should be to continue to impose social distancing measures even as testing protocols are expanded to the scale of the realized epidemic – beyond the April 13 state-wide deadline.

13/15

A failure to continue a statewide shelter in place beyond April 13 by @GovKemp would almost certainly lead to a rebound in cases, hospitalizations, and fatalities, that would unfold in the weeks after relaxing of social distancing guidelines.

14/15

This is yet one more reason to recognize that we have hard challenges ahead.

We must sustain rapid testing (PCR + serology) along with scalable contact tracing if we are to move forward towards containing localized outbreaks, rather than population-wide threats.

15/15

@threadreaderapp unroll