When a new virus with the potential to spread all over the world is discovered, things need to happen quickly. The period from the discovery of the virus to it having spread across large parts of the world and affecting huge numbers of people in a pandemic can be as short as two months. Several epidemiological models were used in Sweden during 2020 to help planning in healthcare regions and decision-making on the national level.
Researchers at Linköping University and Lund University have been commissioned by the Public Health Agency of Sweden to draw up the report: “Sammanställning och utvärdering av modeller för pandemiprediktion i Sverige under 2020" (English title: “Summary and evaluation of models for pandemic prediction in Sweden during 2020”). Researchers at Chalmers University of Technology and the University of Gothenburg have also participated.
In the report, the researchers examine the models used to predict the spread of Covid-19 and the load on the medical care system in Sweden. They have evaluated 22 models developed by Swedish and international researchers, the Public Health Agency of Sweden, and other Swedish bodies.
Models are useful in the planning of measures
“Prediction models attempt to predict how something, in this case a pandemic, will probably develop, based on the information available at a certain point in time. The idea is that the models can be used as a basis on which different actors can decide which measures to take to avoid negative consequences”, says Toomas Timpka, professor at Linköping University and consultant for Region Östergötland.
The authors of the report conclude that several of the prediction models helped to understand how the pandemic developed. These models were useful in planning the measures to take, and showed that the spread of infection would probably differ significantly between different parts of the country. Further, models of various scenarios showed that changes in patterns of social contact would affect the rate of the spread of infection.
However, the report found consistent deficiencies in many of the models.
“One deficiency was that in several cases it was not clear which data had been used, and what the true intention had been of the information. It’s important for the recipient that this is unambiguous, such that decisions can be taken based on the model to the extent that this is possible”, says Anna Jöud, docent at Lund University.
Important to evaluate the models
Only a few of the published models had been evaluated to determine how well the prediction agreed with reality. The report lists recommendations for how work with epidemiological models can be improved.
“Our evaluation shows that it is necessary to standardise documentation and communication of the models and their predictions. It is also important that the assumptions on which the model rests are clearly stated”, says Philip Gerlee, docent at Chalmers University of Technology.
It is important to evaluate the quality and practicality of prediction models, such that they can contribute to preparing society for future pandemics.
“The COVID-19 Forecast Hub in the US is a good example. This allows predictions of the pandemic development to be shared as they are made, such that other analysts and researchers can later evaluate how well the prediction agreed with the outcome. This will help us to find out which methods work well. It would be a good idea to set up a similar programme in Europe”, says Toomas Timpka.
The report (in Swedish)