This algorithm has filtered travelers in Greece for COVID-19 • The Register


Faced with limited resources in a pandemic, Greece turned to machine learning software to decide what types of travelers to test for COVID-19 upon arrival in the country.

The system in question used reinforcement learning, in particular multi-armed bandit algorithms, to identify which potentially infected asymptomatic passengers deserved to be tested and quarantined if necessary. It has also been able to produce up-to-date infection statistics for authorities to analyze, such as the first signs of emerging COVID-19 hotspots overseas, we are told.

Nicknamed Eva, the software was used at Greece’s 40 entry points from August 6 to November 1 of last year. Incoming travelers were asked to complete a questionnaire detailing the country and region they came from as well as their age and gender. Based on these characteristics, Eva chose whether they should be tested for COVID-19 upon arrival. At her peak, Eva was apparently processing between 30,000 and 55,000 forms per day, with each form representing a household, and around 10-20% of households were tested.

Importantly, the software would take the test results as feedback and learn from them to improve its accuracy, and make sure that it didn’t just identify a few types of passengers – it would test the types of travelers. that he rarely encountered to make sure he had a large dataset. In other words, there was more than logic like: if the starting location equals London, then test.

The software was designed to target high-risk travelers without relying on test figures provided by individual countries, who could underreport infections, suffer from bias, or lag behind the actual spread of the disease. virus. Instead, Eva would use her own fresh, real-time data on people arriving in Greece and try to keep infected people away from the general population to help mitigate the pandemic.

There is a very interesting pattern that we observed and reported in our study which shows that the increases in prevalence that we measure through our system are followed by a resumption of cases a few weeks later in the corresponding countries.

“First, given the current information, Eva is looking to maximize the number of asymptomatic infected travelers identified,” explained the US-Greece academic team behind the software in an article on their work published this month in Nature. “Second, Eva strategically assigns certain tests to the types of travelers for whom she currently does not have precise estimates in order to better understand their prevalence.”

The code identified 1.85 times more asymptomatic and infected travelers than random testing methods, “with up to 2-4 times more during peak travel,” according to the team. “To achieve the same effectiveness as Eva, the randomized tests would have required 85% more testing at each entry point,” their article said, regarding that first figure.

Eva also identified “1.25 to 1.45 times more asymptomatic and infected travelers than testing policies that only use epidemiological measures”, and she was also able to reveal which countries were about to experience a slight increase in cases, it is also claimed.

This last part is important because, according to the newspaper, Eva’s recommendations led to Greece being graylisted 10 countries, which meant people from those countries still had to be tested, reducing non-essential travel. from these places.

“There is a very interesting pattern that we observed and reported in our study which shows that the increases in prevalence that we measure through our system are followed by an increase in cases a few weeks later in the corresponding countries”, Kimon Drakopoulos and Vishal Gupta, assistant professors in the Department of Data and Operations Sciences at the Marshall School of Business at the University of Southern California, who co-authored the article with colleagues, said The register by email.

Ultimately, Eva was designed to help Greek authorities perform virus testing effectively, ensuring that groups of people who may have the virus have been tested and those who probably don’t. not have not been. One way to reduce the spread of COVID-19 in the country would be to test everyone at the border before entry and block or quarantine people with coronavirus. Greece’s resources were limited, but it could not completely close its doors; tourism is the country’s biggest industry, and so it depended on it.

“We had enough resources to test about 10% of arrivals in high tourist season and 20% in low tourist season when arrivals were lower,” said the university duo.

Eva was fired after November. “At the end of the tourist season, the number of arriving international passengers became very low, and therefore there was very little benefit to allowing non-essential travel into the country,” Drakopoulos said. The Reg.

“Therefore, Greece has decided to close the borders to non-essential travel and to reallocate all medical staff and resources from Eva to internal measures in the event of a pandemic – testing of the local population, vaccinations, reopening of schools, monitoring of local closures and social distancing measures. “

The researchers declined to say how many people in total were tested after being singled out by Eva, citing privacy concerns. The register asked for the percentage of passengers selected by the system for testing who were confirmed infected with COVID-19. The accuracy rate for catching asymptomatic bionasty carriers was not very high, although accuracy was difficult to assess as a measure of performance for Eva.

“In our context, the prevalence of COVID-19 is generally low (eg, around 2 in 1,000) and arrival rates differ considerably from country to country. Combined, these characteristics make our test data both unbalanced (few positive cases among those tested) and sparse (few arrivals from some countries), ”as the paper puts it.

Gupta said El Reg this accuracy and Eva’s false positive and negative rates were irrelevant in the study. “The objectives of the project were not to predict or guess if someone is [infectious] or not. Rather, the goals were largely prescriptive: to recommend how many and what types of passengers to test to both identify asymptomatic infections and maintain good estimates of the prevalence of COVID-19 for all types of passengers. “

To improve Eva’s performance in catching people asymptomatically infected with COVID-19, academics would need more information from passengers. This is not trivial, however, given the rules regarding privacy and medical data.

Obviously, having access to more data would improve performance but compromise people’s privacy.

“Obviously, having access to more data would improve performance but compromise people’s privacy,” Drakopoulos and Gupta told us.

The team plans to improve their open source code so that other countries as well as businesses, college campuses and schools can deploy Eva during the pandemic.

“One area where we would like to see these ideas implemented is in the now ‘routine’ testing that we currently see in American schools, large office buildings, etc.,” they said.

“By leveraging a reinforcement learning approach similar to Eva’s, it might be possible to make these systems more efficient and focus testing on high-risk individuals. The models should be slightly different, but this could be an interesting avenue for future work, especially in areas of the country where vaccination rates are low. “®

Leave A Reply

Your email address will not be published.