Georgia is tightening the coronavirus testing net, looking to join international top tier in key metric.
In a recent piece, I looked at how the Covid-19 outbreak in Georgia has left the trajectory of exponential growth and might have already surpassed its peak. These considerations are all based on the publicly available figures of confirmed cases. What I did not discuss much, however, is the reliability of those figures.
It has been suspected for a while that the number of actual coronavirus infections in most countries is significantly higher than the number of confirmed cases. Studies from the US, Germany and Brasil point to the conclusion that these countries likely had on the order of 10 times as many infections as were officially reported.
This discrepancy is mainly caused by the fact that about half of all coronavirus infections are asymptomatic, while many more cause only mild symptoms. Patients with none or mild symptoms often do not seek medical care, do not get tested, and thus do not make it into the official statistics.
The extent of underreporting likely differs a lot from country to country. Evidence is mounting that underreporting is closely associated with insufficient testing. This is one reason why the World Health Organisation has been encouraging countries to test broadly.
In response to this, Georgia has recently ramped up its PCR testing capacities and healthcare officials have emphasised they are able to perform all the tests needed. In the following analysis, I look at how tight the Georgian testing net currently is, and how it compares with other countries.
Much has been written about total number of tests performed, as well as the number of tests per capita. However, those numbers are unlikely to adequately describe the stringency of a country‘s testing policy. Instead, the number of tests relative to the number of confirmed cases is a better measure: it gives us an estimate of how easy it is to get tested in a given country. If enough tests are available, the ratio of negative to positive tests will be higher than if tests are scarce.
The following chart displays this ratio on a weekly basis for a selection of countries (for the subtle details of what is shown, read the caption and the methodology notes at the end):
The timeline is shifted such that for each country week 0 corresponds to the last week before the 100th case is confirmed.
Clearly, some countries entered the crisis very well prepared in terms of testing: when they just had a handful of cases, Taiwan and South Korea were already testing more than 1000 times as many people as they had cases. Other countries were ill prepared, most prominently the US which lost valuable time having botched the development of its own test.
In the three weeks after the 100th case the ratio dropped for most countries: case numbers were exploding, and testing capabilities couldn‘t immediately keep up with the new demands. However, thanks to the slow spread within its borders, Georgia did not experience a very significant dip in testing. By Easter, it had consistently performed around 20 tests for every new confirmed case. Many countries, including the US and Sweden, have not caught up to this level until now.
In the weeks after the first shock, countries started to ramp up testing while case numbers have started to drop in some, such that the ratio of negatives to positives is now growing again in most countries. One exception is neighbouring Armenia. Last week, fewer than ten people were tested for every positive case, while daily new infections kept growing. These numbers should raise some concerns in light of the Armenian government‘s decision to lift the lockdown early.
In the three weeks since Easter, Georgia has increased the daily number of tests to well over thousand. Last week, this corresponded to a weekly tests to case ratio of about 70. This week, it is projected to rise above 150, matching the reported figures of neighbouring Azerbaijan, and far higher than many European countries such as Germany or Italy.
According to Amiran Gamkrelidze, the head of the National Center for Disease Control (NCDC), it is planned to further increase the daily number of tests to over 2000.
What does this all mean for the reliability of officially reported case numbers? Does Georgia potentially still have a large number of unreported infections like other countries? Stories of recovered patients who have no idea where they got the virus from seem to suggest so.
To be sure, the Georgian situation is somewhat different from countries with large outbreaks. Internal transmission likely started a month later than in Europe and America, and contrary to authorities in many of those countries the Georgian authorities have not given up on contact tracing and imposing strict quarantine measures. This would at least seem to have given the virus less opportunities to escape the net of disease control.
Moreover, as the above statistics show, the Georgian testing net has been rather tighter than in those countries with suspected high levels of underreporting (US, Brazil, Germany).
Nevertheless, given the high level of asymptomatic cases it seems unavoidable that at least some proportion of cases have escaped the net of the NCDC. What the chart below shows, however, is that the tests to cases ratio has grown significantly in Georgia at the very same time that the relative daily increase in confirmed cases has dropped:
With this in mind, we can be rather confident that the recent Georgian trendline away from exponential growth is real, even if the absolute numbers may not fully account for all infections.
Somewhat paradoxically, the comparably small size of the Georgian outbreak makes it more likely that the true extent of undercounting will never be revealed. I will come back to this statistical artefact in another piece.
- The number of tests per case reported in the second figure above is a 7-day average calculated as follows: #tests within 7 days / #new cases within 7 days. This differs from the 7-day average of daily tests to cases ratio, calculated as 7-day-average(#daily tests / #daily cases), which is more volatile.
- The data in the first figure is based on the same formula, usually using 7-day intervals (Monday to Monday, or Sunday to Sunday, and for India Friday to Friday). The exception is Brazil, where, due to the lack of regular data, I have chosen to use slightly more arbitrary intervals.
- Moreover, because most countries started reporting test data only after a while, the first data point in most of the time series is (#tests until this week / #cases until this week), which tends to be slightly higher than what would be obtained using a 7-day interval. The exceptions are Georgia, Japan, Germany, Latvia and Switzerland, whose first datapoint is based on a 7-day interval as well.
My file with the data and sources can be found here: https://github.com/lo-hfk/covid-in-georgia