Intelligent solutions for dealing with the pandemic - this has been required for a long time and is often surprisingly difficult to implement in practice.

This was discussed intensively about a year ago, when recreational facilities and restaurants were closed indiscriminately in a “lockdown light”, although relatively little information was available about the actual risk of infection transmission at these locations.

This lockdown variant did not actually lead to a decrease in new infections.

It was just one of the examples in the course of the pandemic that showed how much a lack of empirical data can run counter to differentiated and effective measures - and how complex at the same time the important discussion about the protection of personal data in Europe is.

Sibylle Anderl

Editor in the features section.

  • Follow I follow

It initially seems like an obvious idea that in view of the constantly and very dynamically changing infection situation, artificial intelligence (AI) should be used in order to be able to react as quickly as possible to current developments in the pandemic. What this could look like in concrete terms was to be followed last summer 2020 on the borders of Greece. At the time, the country was in a dilemma: On the one hand, it was heavily dependent on the entry of tourists, on the other hand, it had to try to keep the number of SARS-CoV-2 infections introduced as low as possible. While tests can be carried out extensively today, test capacities were still scarce back then. There were only PCR tests for just under 20 percent of those entering the country.In this situation, all other countries relied on a classification of the entry countries on the basis of national epidemiological data: incidences, number of deaths, test positive rates. Depending on how the respective home country performed in terms of these key figures, travelers could either enter unhindered, had to show a negative test, be in quarantine or were prevented from entering.

In Greece, however, people shy away from subjecting travelers to such rough classifications only on the basis of data that are not always easily comparable internationally, especially since it was feared that the requirement of a negative PCR test would discourage some travelers from visiting the country.

American and Greek scientists now describe in the journal Nature how alternative attempts were made at 40 border crossing points to identify as many infected travelers as possible upon entry through the efficient use of the existing test capacities.

Accordingly, travelers were asked to fill out a form at least 24 hours before their arrival in which the country and region of origin, age and gender were asked.

Transparency of the empirical motives

This information, together with the results of tests currently carried out at the borders using the artificial intelligence Eva, was initially used to divide the travelers into groups that had the most uniform risk of infection. For example, different regions were differentiated within a country with a very heterogeneous infection rate. Within these groups, the AI ​​then determined the current need to test the corresponding travelers. At the same time, Eva also identified groups whose prevalence appeared uncertain due to a lack of data and within which additional tests had to be carried out in order to reduce this uncertainty.

The result compared to model calculations: A random test of travelers would have discovered only 54.1 percent of those infected with Eva in the summer of 2020; if tested on the basis of national key figures, it would have been between 70 and 80 percent. The scientists explain the lower success of the use of official figures on the one hand by the fact that the group of travelers is not a representative subgroup of the respective total population. For example, German tourists were significantly younger than the average German citizen. On the other hand, they mention the well-known problem of the time lag in official key figures. An average of nine days were delayed compared to the results of the limit tests.

The scientists particularly emphasize that they coordinated their algorithm with lawyers, epidemiologists and politicians from the start. Great emphasis was placed on working with as little personal data as possible. For example, the respective occupation would have provided meaningful information, which, however, was deliberately omitted against the background of data protection concerns. It was also important to ensure full transparency in all decisions. Particularly in the tests to reduce uncertainties, it was important to present the empirical motives in an easily accessible manner.

Now, more than a year later, the problem of limited availability of tests is obsolete.

The need to be able to react quickly to the dynamics of the infection process, however, still exists.

The Greek study can perhaps serve as an inspiration to think in new directions while maintaining data protection standards.

This could open up scope for the challenge of living with the virus permanently.