Data-driven, Dynamic Optimization of Greece’s COVID-19 Travel Protocols


At the 2021 INFORMS Annual Meeting, researchers Hamsa Bastani, Kimon Drakopoulos, and Vishal Gupta presented the results of their efforts to optimize Greece’s pandemic travel protocols in mid-to- late 2020. At the time, countries around the world were under travel restrictions designed to slow the spread of COVID-19. Importantly, these restrictions were not without consequence. In Europe, estimated losses from tourism alone totaled $1 trillion, leaving approximately 19 million people unemployed. Consequently, as the first wave slowed, many countries sought to ease travel restrictions while still safeguarding public health – exactly the type of complex challenge the decision sciences exist to address.

Unfortunately, as countries looked to partially reopen, COVID-19 tests became scarce. Shortages made testing all visitors untenable, which meant that a best-case testing scenario (screening 100% of travelers and quarantining all of those with positive tests) would not be possible. The Greek government partnered with Bastani, Drakopoulos, and Gupta to design an optimal testing system under these constraints.

The partnership developed the Targeted Testing System (TSS), a machine learning algorithm that “dynamically adapts to real-time information to most effectively allocate Greece’s limited testing resources.” Designing the TSS required the marriage of several key components. The first was the design of a supply chain to collect and process thousands of testing samples per day. The second element involved an empirical Bayes estimation strategy for estimating COVID-19 prevalence among different passenger types at different times and with limited data. Finally, a multi-armed bandit algorithm was designed to dynamically allocate tests to arriving visitors.

Once the team had developed and assembled the necessary elements for the TSS, it began the implementation and testing phases. The team found that the most significant system impact was the number of infected visitors it identified at the border. During its operation from Aug. 6 to Nov. 1, 2020, the system prevented “approximately twice as many visitors with COVID-19 from entering Greece compared to traditional random surveillance testing.” Additionally, the real-time accuracy of the system allowed the Greek government to adapt their border policies as conditions changed, specifically prompting travel limits on 10 countries that exhibited particularly high case rates at varying instances during the observed time frame. Hindsight analysis confirmed that these policy changes significantly reduced the number of potential positive cases in Greece, doubtlessly improving public health and safety.

Finally, applying lessons learned from the Targeted Testing System also helped shape other policies in Greece and throughout the European Union (EU) more broadly. It is impossible to identify all incoming infections without testing every person who enters a country and some infections inevitably slip through. The Greek government used the system in conjunction with visitor itineraries to dynamically predict likely hotspots inside the country and accordingly mobilize testing and contact-tracing resources. TTS estimates became a regular feature in briefing the EU Subcommittee on Travel Protocols – the governing body that helped shape EU travel policy throughout the summer of 2020. The team’s work is a poignant demonstration of the lifesaving power of operations research and analytics.