Better Disease Control with the Covid Prediction Dashboard
A Pilot project in cooperation with EIThealth
“One way to think of AI,” says Peter Norvig, the co-author of Artificial Intelligence: A Modern Approach, “is as a process of optimization — finding the course of action, in an uncertain world, that will result in the maximum expected utility.” The Covid Prediction Dashboard, developed by us, is an example of this kind of refinement. By combining artificial intelligence with spatial technology, the dashboard, a newly-developed technology, provides a visualisation of SARS-CoV-2 reported cases, and a 14-day forecast of predicted cases in various municipalities.
The AI technology is funded by the EU, and the data reflects conditions in Belgium. Besides case numbers, the platform also includes ICU or hospitalisation rates, plus forecasts. Members of the public can log in to view the relative burden of their community compared to the municipalities around them. Various indicators, such as the forecast gradient, provide a simple representation of whether the disease burden is increasing or decreasing, and at what rate. This helps people make important decisions, such as where or when they can travel.
As well as informing the public, the dashboard can also assist with policy and planning, including provincial and facility-level planning with regards to ICU admissions.
Neural networks for short-term forecasting
Because Covid-19 is a rapidly-changing disease with frequently inaccurate public data, we wanted to develop new technology to build a Covid Prediction Dashboard. Neural networks, a form of machine learning based on how the brain works, was the best way to achieve this. These networks, specifically Long Short-Term Memory (LSTM) networks (frequently used in medicine, weather and finance), are able to learn hidden relationships in data. With a set of inputs, short-term temporal patterns reveal themselves, making short-term forecasts or predictions possible.
The digital dashboard
It also includes reproducible benchmarks as well as error scores, which serve a comparative function: anyone can compare their own model with this one. In addition, error values ensure continuous improvement. The ICU case incidence and forecasting tends to be particularly accurate compared to municipal data, because of the greater quantity of data gathered at the provincial level.
Discovery of novel risk factors
By using data which is contextually relevant to the epidemic - for example, population density, climate and air pollution data, and location - the platform can help with identification of novel risk factors. This is achieved by running a relevant data set through the engine of the dashboard to determine its predictive ability.
Although the Covid Prediction Dashboard is a pilot project and not the final product, there are exciting plans ahead. First up, the Covid Prediction Dashboard could help to establish links between historical interventions - such as restrictions on public gatherings, school closures, etc. - and their effects on the epidemic. In addition, there are plans to model the effects of events, such as festive season periods or elections, on outcomes.
Ultimately, the goal is to produce a forecasted report or visualisation on the basis of inputting a potential intervention into the dashboard. Because the model is constantly learning, the integration of further data continuously improves the dashboard’s accuracy and predictive abilities.
There is every reason to believe that technology can make ever greater contributions in the fight against health crises, and that AI can be a force for public good. The Covid Prediction Dashboard, available free of charge to members of the public, uses cutting-edge technology and automatic data collection in pursuit of better control over the Covid-19 scourge.