Let's Keep an Eye on Georgia

Given their controversial lifting of lockdowns that even president Trump thought was too extreme, Georgia will be important to watch in the coming weeks as a case study of such a strategy -- how much will incidence, hospitalizations and deaths increase after this policy has time to take effect, and will there be evidence of a "second breakout" in the state from this policy as expected?


Georgia has pretty extensive tracking data on its website, and with some nice features. Here is some of the current data.

Daily incidence data:

Incidence data is difficult to interpret over time for various reasons:

1. It is dependent on the testing rate, which as testing increases can create erroneous impression of growing case rate even if case rate not growing.

2. An incidence it typically recorded on the test date, when in fact a person testing positive likely was infected 1-3 weeks before. This lag time makes it difficult to interpret the effect of policy on case rate.


Georgia is doing some nice things to combat these problems. The date they record a COVID-19 case is determined based on a combination of dates, trying to adjust for these effects:

1. Date of reported symptom onset

2. If date is invalid or missing, first positive collection date is used

3. If both of these are invalid or missing, the data the case is first reported is used.

Note that even with this procedure, the dates are likely biased late for when the case actually happened. This seems like it will improve accuracy over time, but note that it will also tend to suppress recent counts since the older dates will have incidence "fixed" by this procedure, while newer dates will become "fixed" in coming weeks but not yet.


Note that the incidence is sharply declining in the past couple of weeks in Georgia. Of course, this data does not yet reveal what is happening since the lockdown was lifted, because of the factors mentioned above. We'll have to keep watching this over time and any effects of the new policy will become evidence in these data.


Given these problems with accurately measuring new cases, it is important to look at other factors as well. Here is the daily COVID-related death rate in the state, which of course does not depend as directly on testing:


It is encouraging to see them also decreasing sharply over time. This date is the actual death rate, so is lagged 2-3 weeks or so after the date of infection, but this graph does not have the potential bias problem in more recent days as the incidence graph above.


Given the lag time, this plot won't be meaningful to assess effect of lifting of lockdowns for another week or two, but bears watching.


Another important indicator to follow that is less lagged than deaths, and also less influenced by the testing rate, is number of hospitalizations. Georgia measures these on a daily basis -- they don't have a graph on their website, but it is in http://covidtracking.com so can be seen on that site. Our web app for plotting data from that site can be used to look at raw numbers (BTW it seems to be down right now -- hopefully we will have it back online soon).


Another positive aspect of Georgia's reporting is the reporting of incidence by race. Given the controversy about details of the state's lockdown lifting strategy, which some say may be racially motivated, it will be important to track the incidence, hospitalizations, and deaths by race to accurately assess whether the policy has any disparate effects across racial groups. We should keep an eye on this as well.


The next couple of weeks should start revealing whether the new state policies are leading to dramatic uptick in cases and hospitalizations here in Georgia as well as other states (Texas and Florida are big states with dense urban centers that will also be good cases to watch).


These should give us empirical evidence to inform policy makers, but as everything with this crisis, the data are nuanced and need to be evaluated carefully.

 

©2020 by Covid Data Science. Proudly created with Wix.com