FORENOON: Framework fOR Epidemic iNtervention OptimizatiON

FORENOON uses historical data to analyze the impact of different non-pharmacological interventions against a disease (e.g., COVID-19), enabling automatic generation of scientifically-sound intervention plans. This in turn allows policy-makers to inspect different trade-offs between intervention plans’ effectiveness and cost, and based on that make informed decisions. 

Dr. Mitja Lustrek, PhD in computer science, head of Ambient Intelligence Group at Jozef Stefan Institute, principal investigator in many health projects, two terms chair of Slovenian AI Society.

The COVID-19 pandemic revealed how unprepared most of the world really was for a crisis of this type. The rapid spread of the disease required the deployment of non-pharmacological interventions (e.g., school closing). However, very different intervention plans were applied in different countries -- many of them ineffective and not supported by evidence. Such inadequate intervention plans could then lead to either rampant infections and/or unnecessary social and economic damage as the result of lockdown-like policies. This problem was partly due to evidence not existing, and partly to policy-makers not having tools that could effectively take advantage of it.

The proposed solution will take maximum advantage of the available evidence on intervention effectiveness and costs to provide policy-makers with the ability to estimate the effect of different interventions, to choose which to apply, and to justify their choice. Its usefulness will be increased further by designing ideal intervention plans (semi-) automatically using AI methods. Our preliminary work shows that the intervention plans that were actually in effect can in virtually all cases be improved upon using such methods, which shows both that the problem exists and that it can be solved.

Our solution is first and foremost envisaged as a tool for policy-makers. By providing them with the tool developed by combining expert knowledge and a data-driven system, they can choose the appropriate non-pharmacological interventions and understand the impact they have on epidemic development, economy and society. It can be used for COVID-19 or another future epidemic, significantly enhancing their readiness to deal with a public health crisis.

We have been collaborating with the Slovenian government and the Slovenian expert group for the COVID-19 pandemic. In our team we have the head of the Center for Communicable Disease at the National Institute of Public Health (the institution that develops public health policies in Slovenia), who has been a member of this expert group. We therefore have a good understanding of the needs of policy-makers. By considering their inputs we make sure our solution is fit for its purpose. We would like to expand our impact to other countries, with a special focus on the EU and the countries of the Western Balkans.

Direct users/customers of our solution will be policy-makers, whom it will support in making  evidence-based decisions regarding non-pharmacological interventions, which will rely on historical data and AI methods. The outcome of different interventions will be presented with clear and intuitive visualizations, which will help users of diverse backgrounds understand the impact of different interventions.

The intended users of our solution make public health decisions impacting the general population, so any improvements in interventions will benefit them by decreasing infections and/or socio-economic costs of the interventions. This is a clear case of public good, as COVID-19 demonstrated that the cost of such a pandemic is huge and even a small improvement in interventions can be significant for the population.

While we will provide our solution at a cost (as our work needs funding), we plan to make this cost modest, and one deployment can serve a large population (such as a whole country), making the cost per person impacted very small.

We are also writing scientific publications about our solution, which will be publicly available and will increase knowledge about epidemic modelling and intervention planning.

Our solution will reduce infections and/or socio-economic costs for the population in general. It has already been used by the Slovenian government. The interventions were evaluated with our solution and by applying them we were able to slow down the third wave of the pandemic with a lower cost.

Our methodology is well suited to tailoring interventions to vulnerable populations (elderly, schoolchildren, small companies ...). Our predictor can model distinct groups and infections among them separately. The model of socio-economic costs can also be as detailed as policy-makers are willing to make it, and can incorporate costs for distinct groups, which we encourage.

By receiving the funding and the exposure from the Trinity Challenge, we would be able to develop our work further. Our research regarding COVID-19 is currently mainly based on volunteer work and is therefore limited. Our framework could benefit several people and could also help put more trust in decisions policy-makers are making, as it will rely on scientific research, expert knowledge and historic data. By gaining trust from the public, the interventions could have a better impact, as proper public informing has had a big role in how the public adhered to the interventions.

In the first year we plan on scaling to European countries, mainly by connecting with public institutions. We already collaborate with the Slovenian government, thanks to our team member Matjaž Gams being the National Councillor for Research and Development (National Council is one of two chambers of the Slovenian parliament). We are discussing how to take advantage of the Slovenian Presidency of the Council of the EU (in the second half of 2021) to promote our solution to other European countries. In this we collaborate with European Connected Health Alliance. We will also take advantage of available European research grants.

In the following two years, we will have a more polished solution that can be offered to governments and other institutions commercially. Our focus will continue to be Europe, but we plan to expand to other countries. Our success at the XPrize Pandemic Response Challenge gave us access to organizations such as International Telecommunication Union of the United Nations (ITU) and North American members of the Pandemic Response Alliance. Our collaboration with Taipei Medical University gives us access to Asia. We will use these connections to establish partnerships or commercial relations with national or regional policy makers.

The performance of our predictor and prescriptor will be measured with scientific KPIs. The prediction accuracy will be measured with the mean absolute error on historic data, and the optimization quality with improvements over actual interventions in terms of infections and estimated socio-economic costs.

The impact on the work of policy-makers will be measured through their satisfaction with our solution. This will be assessed during workshops we plan to organize to educate them about the solution, using questionnaires and qualitative methods (focus groups, semi-structured interviews).

The most important impact is on the general population, which should be less burdened by epidemics if policy-makers adopt our solution. The previously mentioned scientific KPIs can approximate this, and we can measure public opinion via online surveys, but true evaluation would require a controlled real-life policy experiment, which may not be feasible.

Additional KPIs will include the number of institutions using our solution after each year (goal: 2, 4, 10 institutions), and the time necessary for deployment after specifications are provided by a partner or customer (1 month, measurable in the second half of the project).

First barrier: finances. Currently we rely on the National Research Agency core funding. We won half of the main prize at the 500k$ Pandemic Response Challenge, by XPRIZE and Cognizant (i.e. 250 k$). Additional funding from the Ministry of Health is currently uncertain. The Trinity challenge can overcome this.

Second barrier: the governments (Slovenian and other) are reluctant to embrace solutions based on computer models, since the politicians have to balance a wide variety of interests, thus not always choosing “mathematically” optimal solutions. Our solution can incorporate politicians’ preferences, so we plan to overcome this barrier after the first year with a proactive approach, involvement of policy-makers in the development process, and workshops for the stakeholders. In addition, we expect that the success of computer models will increase the reputation of such approaches in the eyes of the general public, leading to an easier adoption.

Third barrier: we are from a research organization. The solution is to create a spin-off company that will take care of the commercialization in the second or third year. Our institute has support services for such cases, and we also hope for support from Trinity members.

University of Ljubljana, National Institute of Public Health, Ministry of Health of Slovenia, University Medical Centre Ljubljana, several regional hospitals in Slovenia,  XPrize, Cognizant

Our work began with a sense of moral duty of scientists that have the means and the knowledge to help fight the pandemic -- in our case, with the data analysis and forecasting models. We started as newcomers to the field of epidemiology, however, since the spring of 2020, we developed skills and learned a lot, resulting in, for example, being one of two grand winners of the XPrize Pandemic Response Challenge.

 However, our solution is still at a prototype level. We require further funding to develop practical solutions that will be directly available to the policy-makers and other stakeholders. We believe the Trinity Challenge is a great opportunity to pursue this goal, not only in the financial aspect but also in view of supporting services, collaborations, international exposure, and further networking opportunities.

To make this possible, we would like to collaborate with organizations from the following areas:

Research organizations, companies and individuals with expert knowledge in the economic and psychological impacts of non-pharmacological intervention. Such information plays a critical role in determining optimal intervention plans.

• The London School of Economics and Political Science

Data providers that collect COVID-19 related information that would help improve our predictor and prescriptor models:

• Premise and Facebook collect global survey data on economic, social, and health sentiment regarding the COVID-19.

• Blavatnik School of Government at the University of Oxford collects publicly available information on indicators of government response.

• Center for Systems Science and Engineering at Johns Hopkins University collects global data on COVID-9 cases, deaths, vaccinations, and more.

Partners that provide computing services that would speed up the calculations, and for hosting our web application.

• Microsoft Azure

• IBM Cloud

• Amazon Web Services