THEMIS: Holistic Estimation of Economic and Human Impact for COVID-19

An analytics framework to understand the economic and human impact of COVID-19 governmental measures, and identify optimal policies for future pandemics.

Prof. Dimitris Bertsimas, Boeing Professor of Operations Research, Massachusetts Institute of Technology

Our fundamental goal is to answer the question: “How can governments implement better policies to combat future pandemics, based on the learnings from COVID-19?”. Since December 2019, governments around the world implemented many non-pharmaceutical interventions to combat COVID-19, ranging from social distancing to lockdowns. Although it is widely accepted that most measures did mitigate the virus spread, they carried significant economic costs from unemployment to reduced production, as well as increased depression, anxiety and PTSD.

In the US alone, when lockdowns were first implemented, GDP plunged over 30%, over 20,000,000 people lost jobs, and prevalence of clinical depression increased by 10%. Hence, there is a clear trade-off between reducing deaths and economic/social cost in a pandemic and there is significant debate, both among the public and the scientific community, on whether the measures were “worth” the cost. However, quantifying costs from multiple social and economic dimensions is notoriously difficult. The lack of such direct quantification has fueled non-compliance in groups highly affected by the policies, and increased public distrust in authorities if they felt the policies were not necessary. Our goal is to provide a holistic perspective on the impact of different policies to inform governments on handling future pandemics.

The direct target audience are governments worldwide. It is clear, through discussions with various authorities, that many governments were lacking knowledge about the effectiveness of available tools to combat a pandemic at its different stages. There are policies (e.g. contact tracing) that were effective in controlling early-stage pandemics (e.g. in South Korea) but did not stop more advanced ones. Furthermore, governments were also less than adequately prepared to deal with the fallout from the implemented measures (e.g., protests). Indeed, repeated application of stringent measures reduces compliance and effectiveness, while huge social and economic costs remain. 

THEMIS is designed to help governments prepare better for the next pandemic by showing them the trade-offs of implementing different measures at certain points in the pandemic. The holistic view provided by THEMIS would allow governments to identify effective policies that are appropriate for a particular region and time. Indirectly, this would benefit everyone when more effective and less costly policies. It would also serve to rebuild trust in institutions.

We plan to publicize our results in top journals to further develop the credibility of the solution, and continue discussions with governments to incorporate findings into their future pandemic response plan. 

We plan to release at least two public goods to the society: First, we plan to make the entire codebase public. We have already released DELPHI on our website www.covidanalytics.io, which has had half a million hits since last summer, and we plan to release the THEMIS framework on the same website for easy access by the public. This would enable anyone around the world to view and replicate the conclusions and findings for each country, and allow more people to understand the complexities behind the government decisions and thus help build trust for authorities worldwide. 

The second, long term public good that our work aims to provide is that by being easily accessible by anyone and adaptable to any country or region around the world, our models and insights can be used by governments in their future pandemic response playbooks. For that goal, we also plan to write academic papers to summarize the conclusions and findings using THEMIS, and aim to publish them in top journals.

Our work on DELPHI already created significant tangible impact at many organizations (https://www.covidanalytics.io/users), including Janssen, which used DELPHI to select Phase III trial locations for the world’s first single-dose COVID-19 vaccine. Our model is also consistently included, since April 2020, in the official CDC core ensemble prediction that many agencies utilize for planning purposes.

For THEMIS, our direct targets are the governments worldwide. During the pandemic, over 150 countries/regions implemented non-pharmaceutical measures to combat COVID-19. However, it was clear that many governments did not have a good understanding of the multi-dimensional impact of the applied policies. We expect our analysis to highlight this multi-dimensional impact, from economic loss and unemployment to mental health and depression. Given THEMIS, we can also easily extend our analysis to identify minority groups disproportionately hit by the policies. With our existing channels (e.g. CDC, ECDC), we are hopeful our research can affect how countries implement future pandemic policies. 

Furthermore, the planned public THEMIS codebase and academic papers would allow anyone to replicate our results and understand the different dimensions governments need to consider when choosing which policy to implement. This would hopefully help to rebuild trust in institutions worldwide.

DELPHI has already been implemented in over 200 countries/regions worldwide. THEMIS, building on DELPHI, is currently implemented for Germany and the state of New York in the United States. Over the next year, we plan to extend THEMIS to other countries and regions to compare and contrast the results. In particular, beyond our current focus in western developed countries, we are interested in extending to other regions in the world, including large developing countries (India, Brazil) but also regions where the response has been largely effective (Japan, South Korea, Vietnam),  to provide more insights to the governments on the optimal policies for each region. 

Over the next three years, we plan to deepen our analysis by looking at how the policy effectiveness is affected by compliance and other social behavior. We would further engage with authorities worldwide in an attempt to affect the governments’ preparedness for the next pandemic, so that policies can be implemented with more effectiveness while inducing less strain on the greater society. A detailed timeline is included under the “Your Plan and Funding” section.

In the near term, the primary statistic that we are concerned with is the number and the distribution of regions that the THEMIS analysis scales to effectively.

We hope to add as many regions as possible over the next year, and analyze the differences in the optimal policy based on each region’s economic and social situation. Objectively, we would measure the performance based on how well THEMIS can replicate the actual cost of the pandemic under the actual policies implemented in the country over that period of time. 

More importantly, we are also focused on the geographical distribution of the regions that THEMIS applies to. Given the disparate measures and results of measures around the world, we would like to not only focus on developed countries in Europe and North America but all possible areas worldwide (Asia, Africa, Oceania, South America, etc). Therefore, we would closely monitor the geographical distribution as we expand. 

We also plan to complete at least one academic paper on the results.

Long-term, we would be measuring our success in our ability to affect policymakers around the world to adopt our conclusions into their pandemic planning for the future.

The primary difficulty regarding this project is scaling. Developing a realistic THEMIS simulation for each country requires extensive data collection and cleaning from tens or hundreds of sources, and is a resource-intensive task. Fortunately, the team can build upon the experience in scaling our epidemiological model, DELPHI, to over 200 countries worldwide on www.covidanalytics.io. Therefore, we believe that we would be able to expand the applicability of THEMIS to most regions worldwide. This could be further accelerated through utilizing any research funds from the Trinity Challenge to expand our team.  

The second difficulty is to influence policymakers worldwide to adopt the conclusions that we have generated into their future pandemic planning. To overcome such barriers, we would utilize our current channels with authorities (e.g. CDC, ECDC) based on our success with DELPHI. Moreover, we hope that by publicly releasing the codebase, this would stimulate future research and interest in this field, and more people would be aware of the disparate impact of the policies on the society, leading to action from the authorities. We also hope that participating in the visible Trinity Challenge would also further raise awareness about this topic.

Massachusetts Institute of Technology (Sloan School of Management, Operations Research Center)

We believe the Trinity Challenge could directly help us address the two primary challenges that we are facing related to scaling and impact. Funding from the Trinity Challenge would enable us to scale to more countries worldwide; we have also already seen, working with Trinity Challenge members Swiss Re and Palantir Technologies, on how resources within the Trinity Challenge would be extremely useful in publicizing our solution quickly. We further believe that the Trinity Challenge would help our solution be more visible and thus increase the chances that institutions worldwide would incorporate our findings in their future pandemic planning.

We are already collaborating with Swiss Re, Palantir Technologies, and members from the University of Cambridge as part of the previous application, and we plan to continue our work with them going into the future. Swiss Re and Palantir Technologies provide much needed expertise on scaling the solution and they also maintain a very large pool of data resources that will be extremely helpful to our research. Dr. Simone Schnall and Dr. Catherine O’Brien are extremely knowledgeable on compliance behavior and sentiment in the pandemic, which is an important next step for us to include in THEMIS.