Background

RespiCast, the European Respiratory Diseases Forecasting Hub, combines multiple forecasting hubs for several respiratory disease indicators, including influenza-like-illness (ILI), acute respiratory infection (ARI), and indicators related to COVID-19. RespiCast is a collaboration between the European Centre for Disease Prevention and Control (ECDC), the ISI Foundation, and the London School of Hygiene & Tropical Medicine.

Teams from anywhere in the world are invited to submit weekly forecasts for one or more disease indicators and for one or more countries.

For more information on technical details and contributors, visit the project pages on GitHub:

For further insight about respiratory virus surveillance data check the ERVISS website .

Note: COVID-19 indicators forecast are currently hosted at this link: https://covid19forecasthub.eu/

How does RespiCast work?

RespiCast collates probabilistic 1 to 4 week short-term forecasts on indicators (e.g., number of cases) for ILI, ARI, and COVID-19 across Europe generated by independent teams, who use a wide range of approaches of modelling infectious disease dynamics. Individual forecasts are combined to create an ensemble forecast (see "What is an Ensemble Forecast") - an established way to achieve improved forecasting performance for infectious disease dynamics[1].

Through its efforts, RespiCast aims to provide decision makers and the general public with reliable information about the near-term future of the epidemiological situation in Europe in the entire EU/EEA pertaining to respiratory illnesses.

The secondary objectives include understanding predictive performance of various modelling approaches, evaluating the accuracy of forecasts with respect to different diseases and indicators, and fostering a community of infectious disease modellers committed to an open-science philosophy.

What is a Probabilistic Forecast?

Within Respicast, forecasts are presented in the form of predictive intervals rather than point values. To understand the difference, consider the following examples:

Point forecast: "Next week, the ILI incidence will be 23 cases per thousand individuals."
Probabilistic forecast: "Next week, with a 95% probability, ILI incidence will range between 18 and 25 cases per thousand individuals." (In this context, probability is a measure of confidence of the forecast method, i.e., the forecast method is “95% confident” that the ILI cases are between 18 and 25.)

In the first example, the point forecast refers to a specific future value with no associated uncertainty or confidence. In contrast to the point forecast which states a single value, the probabilistic forecast has three fundamental elements: i) a confidence level (95% in the example), ii) a lower limit value (18 in the example), and iii) an upper limit value (25 in the example). For probabilistic forecasts, one can state different confidence levels for the same forecast method. The lower the level of confidence, the closer the lower and upper limit value. For instance, when asked for the 50% instead of 95% confidence interval, the forecast method in the second example above could yield a probabilistic forecast as “Next week, with a 50% probability, ILI incidence will range between 22 and 23 cases per thousand individuals.”

Probabilistic forecasts offer significant advantages over a point forecast. Epidemics depend on numerous factors such as human behaviour and seasonality, contributing to their uncertainty. A point value is therefore insufficient to effectively and comprehensively communicate this uncertainty.

On the RespiCast platforms, it is possible to select three different confidence levels: 50%, 90%, and 95%. By selecting, for instance, 90%, users can visualise the cone of uncertainty within which, according to the model under consideration, the actual incidence value for future weeks will lie with 90% probability. Along with confidence intervals, also the median prediction (i.e., the central value of the forecast) is shown.

What is an Ensemble Forecast?

An ensemble forecast is the result of combining predictions from different models in order to provide a more comprehensive and robust view of the potential future development of a complex system, such as weather phenomena[2] or epidemiological dynamics[3]. Intuitively, the ensemble approach incorporates various models, each representing a possible representation of the system under study. This approach yields future projections that represent this intrinsic uncertainty about the system instead of solely relying on the assumptions and hypotheses of a single model. There are multiple methods for creating ensembles (median, mean, weighted based on different performance metrics), and the performance of those is an active area of research [4] [5]

References

[1]: https://royalsocietypublishing.org/doi/full/10.1098/rsif.2022.0659

[2]: https://www.science.org/doi/full/10.1126/science.1115255?casa_token=O4PlxVx8lsYAAAAA:gO3gqqdRC37srPL1WMsvTIa9DXNPZ9cCgwa8bFyLJZxIRaVCZFmD4M2Vl6XzpHwxWtMqXMDi5gwV3Q

[3]: https://ajph.aphapublications.org/doi/full/10.2105/AJPH.2022.306831

[4]: https://royalsocietypublishing.org/doi/full/10.1098/rsif.2022.0659

[5]: https://elifesciences.org/articles/81916