PanSim: Prediction and optimal control of pandemics

PanSim realistically simulates population movements, interactions, and disease spread to design and test the effects of possible interventions to control pandemics   

Dr. István Zoltán Reguly - Associate Professor, Pázmány Péter Catholic University, Faculty of Information Technology and Bionics

Global problems require globally applicable solutions that consider the socioeconomic disparities. As the COVID pandemic has shown, disease knows no borders, and our response needs to be coordinated. Furthermore, our response needs to consider socioeconomic disparities. We are addressing the challenge of predicting the spread of infectious diseases in large populations, and how to design optimal control measures. Our solution considers the underlying structure and mobility patterns of populations, how diseases spread in them, and how the potential response affects both the spread of the infection and the impact on different parts of society. 

Our work addresses both the current pandemic which has already killed over 2.7 million people and prepares for the next one, by giving a tool that allows for the integration of the latest data on mutations, infectiousness, hospitalization rates, and many others into an open science simulation framework that then enables exploring the best courses of action in terms of slowing the spread of the disease and causing the least damage to already vulnerable populations. Utilizing the outstanding computational efficiency of the simulation, it is possible to design and test several optimization and control (e.g., mitigation, suppression) scenarios in real time.

Our solutions on the one hand target key decision makers at different levels of government, by providing them with predictions on the spread of a disease and helping them explore the best courses of action both in terms of limiting the spread of infection and socio-economic impact. The types of predictions we currently make and data extracted from them have been iteratively refined with the help of epidemiologists and government to focus on the most important factors that affect decisions. One of our goals going forward is to widen the array of metrics by which measures can be evaluated, thereby further supporting decision making. To better understand these requirements, we are going to engage scientific advisors at other countries, to help determine non-trivial constraints to such decisions. Our results will also inform non-governmental organizations about what populations are impacted most severely with respect to various criteria by both the disease and the response to it - which will help direct efforts to those most in need. Here, we are planning to engage NGOs, and with the help of Trinity Members, refine our analysis to provide the most relevant and actionable data for these organizations.

Our solution will provide an open-source simulation model available to use for anyone. Our work will also provide a dashboard of openly-available analysis of simulation results that can be directly utilized to determine the spread of infection, to determine the populations impacted either directly or indirectly (e.g. having to stay at home, risk of losing jobs), and to evaluate the effects of efforts to limit the spread.

Our solution creates a tangible impact for key government stakeholders at both national and local levels in determining the best courses of action during a pandemic that can balance between the socioeconomic and health impacts in different subgroups of society. We will also have an impact on NGOs looking to provide aid to vulnerable populations, by helping to identify them in advance. Having developed the methodology to estimate and predict factors not directly related to the spread of infection, our solution will be extensible to produce further types of indicators relevant to other industries - this will be subject to collaboration with specific companies or sectors.

Our current pilot project has already affected the lives of Hungarians (population 10 million) by providing actionable evidence to the government regarding the spread of COVID and the effects of potential restrictions on the spread of the disease. We plan to scale up our solution to provide data on the direct economic effects of the pandemic and the restrictions in the next year. Working with mentors, we will scale our models to cover multiple countries in Europe and the US. Depending on the availability of data, our moonshot goal is to have a simulation of the spread of a respiratory disease across the entire world - which we already deem to be possible in terms of computational power.

At the same time, we will extend our solution’s capability over the first two years to design optimal measures for individual countries based on a control theoretical approach, and scale this to cover multiple countries over the third year.

We are going to use the following indicators:

1. Population size for which we have adequate input data (currently ~180,000)

2. Number of countries for which the model makes good predictions based on past data (currently 1)

3. Number of metrics that can be used to evaluate the efficacy of measures (currently: number of people infected, hospitalized, dead, quarantined. Future: socioeconomic impacts - loss of working hours, loss of life expectancy, risk of losing jobs, etc.)

4. Number of constraints with respect to which optimal control measures can be designed

5. Public availability of our solution (website with results, analysis), and its reach (visits)

We currently operate in Hungary, and our solution targets the same country. Over the next 3 years, we wish to extend the scope of our solution to European countries and the US, however, we do not believe we need to establish a physical presence in those countries.

The key barrier is the availability of data on the population and “locations” - much of this is public data, and therefore the challenge lies with the automation of extracting the relevant statistics. Data on businesses and mobility patterns will likely involve cooperation with companies such as Google, Facebook, and McKinsey, who already have much of this data - and therefore legal agreements will have to be in place so that we can appropriately access, handle, and anonymize data. We do not foresee requiring any personal or business-critical data 

University of Szeged

There are two key barriers the Trinity Challenge will help us address through collaboration with its members; acquiring uniformized data on the populations to be simulated and helping reach key decision-makers. We believe collaboration with Trinity members will let us meet our goals and create tangible impact for public health, tackling socioeconomic disparities, and more.

Google: data on populations (spatial and age distribution), businesses, mobility patterns, and computational infrastructure

Facebook: data on populations and mobility patterns

McKinsey & Company: data mining and analysis - expertise and support

Clinton Health Access Initiative, Bill and Melinda Gates Foundation - engaging key decision makers at governments and NGOs.