A few months after the Covid-19 outbreak, operations researcher Kimon Drakopoulos, who works in data science at the University of Southern California, offered to help the Greek government by developing a system that uses machine learning in order to determine which travelers had the most risk of being infected and thus should get tested. Greece was asked by the European Union to allow non-essential travel again, but of course the option of testing all travelers was not available. Consequently, they chose to implement a more efficient way to test incoming travelers than the usual practices of randomized sample testing or testing based on the visitor’s country of origin, by launching this system called ‘Eva’ and deploying it across all Greek borders.
Drakopoulos and his colleagues discovered that machine learning proved to be more effective at identifying asymptomatic cases than the aforementioned methods, by a factor of two to four times during peak tourist season. This was accomplished because Eva used multiple sources of data, besides just travel history, to assess and estimate the infection risk of an individual. These sources include demographic data like the age and sex of the travelers, which was then paired with the obtained data from previously tested passengers, to calculate who had the highest risk out of a group and needed to be tested. This process was also used to provide information to the border policies about real-time estimates of the prevalence of COVID-19.
When the researchers compared the performance of this model against the methods that only use epidemiological metrics, such as random testing, it was clear that it performed better in all aspects. One main reason for this was the limited predictive value that these metrics possessed in relation to asymptomatic cases. Consequently, the paper raises concern on the effectiveness of internationally proposed border policies that employ such population-level metrics.
All in all, Eva is a successful example of how the use of reinforcement learning and artificial intelligence in combination with real-time data can provide very useful assistance both in crisis situations but also in the public health sector.
References
Bastani, H., Drakopoulos, K., Gupta, V. et al. Efficient and targeted COVID-19 border testing via reinforcement learning. Nature (2021). https://doi.org/10.1038/s41586-021-04014-z
Nature (2021) ‘Greece used AI to curb COVID: what other nations can learn’, 22 September. Available at: https://www.nature.com/articles/d41586-021-02554-y (Accessed: 29 September 2021).
Hi Stylianos,
Thank you for this interesting piece on how to use AI technology on predicting COVID-19 infections and inherently ensuring others do not get infected! I wonder why the Netherlands did not have such a system in place to limit the time in lockdown… After reading your piece, I was still wondering about the following:
1. How much better did the AI system worked compared to the epidemiological metrics?
2. Do you think this system should be permanently installed on airports to reduce the risk of spreading diseases (not including COVID-19?)
Thank you in advance!
Hi Florian,
First of all, thank you for your comment and your interest in my post! To answer your first question, against modelled counterfactual scenarios, Eva identified 1.85 times as many infected asymptomatic travelers as random surveillance testing, while this number jumped to up to 2-4 times as many during peak travel periods. Against testing policies that only utilize epidemiological metrics, it identified 1.25-1.45 times as many infected asymptomatic travelers.
As far as your second question goes, I would have to think that this system could not easily be used universally, as it is simpler to focus on one specific disease or virus that has certain characteristics and is tracked at the level that COVID-19 is, as opposed to creating a universal system that checks for all diseases. Moreover, COVID-19 is a global phenomenon that had and still has the power to disrupt the regular functionality of governments and countries, while almost all other diseases are just a part of everyday life.