Back in March, Europe and Finland were waking up to the reality of COVID-19. Many questions and problems started ariising, all needing better answers than were currently available. Enter one part of an unusual connection: Heikki Lukkarinen and his team at Turku University Hospital (Tyks). Their research question was whether we could model and simulate potential medical interventions to Covid-19: “How could we get the maximum effect from a medical intervention?“ We agreed to collaborate on this research topic. In addition to me, Atte Juvonen, Eeva Nikkari, Vilen Looga and Mikko Viikari from Futurice contributed to various streams in this work.
We began by evaluating publicly available models and, again, unusual times created unusual global collaborations across continents, individual experts, organisations, public and private entities - all in order to get things done. The speed and openness with which the connections were created were amazing. They led to several streams: work on different epidemiological models, parameters, visualizations, and a system view. Corosim was the result of some of these streams. We will discuss our work on different models in a later blog post.
Epidemic models are typically written in academic environments, such as R-scripts or computational notebooks. In practice this means that stakeholders ask a team of mathematicians to simulate a scenario, the team retreats into a silo and comes back a week later with a PDF describing their results. Based on our experience working with data-driven management systems, we wanted to do more to operationalize the model. With Corosim, we have achieved the following:
- An interactive visualization with an intuitive interface, allowing a non-technical domain expert to change parameters and simulate different scenarios without coding
- Applied to real-time local conditions; in this case, real historical data in Finland, Finnish population size, local key parameters
- Ability to apply several interventions/changes to the model at specified time points
Time was of the essence, so instead of building the user interface from scratch, we leveraged existing work published by the open-source community. We were familiar with Gabriel Goh’s fantastic Epidemic Calculator, and we realized it would provide a good base to build upon. We contacted Gabriel and started working on it. In the end, we also published our work as open source, to give back to the community we borrowed from.
The more we worked on the original topic, the broader and more systemic the questions we started getting from various organisations and entities - including a request for a written expert opinion from one of the subcommittees of the Finnish Parliament. We addressed the questions as best we could - in a very holistic way, in other words.
Due to the complexity of the topic, this model is mostly relevant for people with domain expertise, but anyone is free to give it a go. Read more about the added functionality, differences to the original, how we calculate various parameters, like the famous R0. The source code is also available.
What we learned
- A collaboration of a multidisciplinary team of data scientists, doctors, epidemiologists, software developers, mathematicians, and systems experts creates magic.
- Although it makes sense to re-use existing models and tools, at the same time it really makes sense to understand those in detail in order to understand their applicability and relevance.
- A model and simulations in isolation provide only limited value unless it is properly connected to a broader context.
- There are interesting deep structural similarities between fields that initially feel distant from each other such as the spread of an epidemic in a community and the flow of knowledge within an organization.
Take a look for yourself at corosim.fi. If you want to have a chat about this simulator or how similar models could be applied in other areas, please get in touch!