Thread by @joshuasweitz

How many people have been infected with #COVID19 in Georgia?

A: ~17,000, at least according to the GA DPH:
dph.georgia.gov/covid-19-daily…

But, it is likely far more.

I will use this thread to explain why I would not be surprised if actual GA numbers exceed 150,000+.

First, we have been working on GA-level scenarios since mid March, building upon the work of @neil_ferguson and others. Our models seed a COVID19 epidemic in GA until simulations appear to be “close” to the reports of cases, hospitalizations, and deaths.

But, getting reasonable models to get “close” to data was hard. The simulations could get close to hospitalizations and deaths OR cases BUT NOT all at once.

It could be that our model assumptions were wrong. Indeed, we know these are uncertain assumptions.

Yet, the larger issue also links to reporting. It’s almost certain (as has been discussed by @meyerslab, Sam Flaxman, @drake_lab and others) that case ascertainment rate is (very) low.

In other words there may be 5x or 10x (or even 25-50x) more cases for each reported case.

Hence, it was possible to seed an epidemic with reasonable disease transmission assumptions and reach a point where the deaths (and to some extent the hospitalizations), and the age structure of fatalities resembled reported data.

These fits – as seen below – could then be used as the basis to ask the question regarding how the epidemic would spread given alternative effects of social distancing.

But, there’s a caveat.

The fits of ‘reported cases’ don’t coincide with the model - not even close.

Instead, the number of cases the model thinks are present in GA, which we’ll call C(model) must be scaled-downwards to match the reported cases C(reported).

This factor is the ascertainment rate, which we’ll call “a”. The initial model fits suggested a~1/26, in other words only one in 26 cases were reported. Hence, according to that rate, the current value of 17000 cases implies ~440,000 cases (or ~4% prevalence in the state).

Although deaths/hospitalizations track a similar forecast, the cases seem to trending upwards. This could mean many things. But, even with a 2x increase in ascertainment (consistent with the case trend), that suggests 220,000 cumulative cases.

Another way to think about underreport: there have been ~200 new reported fatalities this past week in GA.

Given a lag of ~2-3 weeks between infection and death, and <1% infection fatality rate, there should have been ~20,000-30,000+ new cases between 2-3 weeks ago.

Instead, there were only around 3,000 new cases that occurred over a week period, 2-3 weeks ago. This suggests a lower scenario of about ~8x under-ascertainment.

Hence, I think it plausible that the 17,000 reported cases via @GaDPH (or 0.15% of the GA population) actually correspond to roughly between 1-4% of the population; even this range is an under-estimate of reasonable bounds of uncertainty.

Wrapping up: it would not surprise me if population scale serology reveals that GA has somewhere between 150,000 to 450,000 cases, rather than the 17,000 cases we see now. There are many implications for hospitalization, fatalities, and planning.

And to add to clarion calls from many – widespread testing is needed not only to understand the prevalence of COVID19 but to stem the tide of future infection.

#PCRNow #SerologyNow