Weather data predict risk of hospital congestion
A team of researchers supported by the Swiss National Science Foundation has developed a mathematical model that anticipates flu peaks in hospitals based on weather data.
When too many people fall ill at the same time, hospitals run the risk of congestion, as the Covid-19 pandemic showed to a certain extent. The flu virus can cause the same problems. A team of researchers supported by the Swiss National Science Foundation (SNSF) has developed a mathematical model that predicts the risk of flu-related hospital congestion based on weather conditions. Its findings are published in the Journal of the Royal Statistical Society (*).
Influenza is a seasonal virus, mainly present during the winter season in our latitudes. The research team therefore compared certain meteorological data – precipitation, humidity, temperature and sunshine – with the cases of influenza recorded daily for three years at Lausanne University Hospital, better known locally as CHUV.
The extremes rather than the average
For the first time, however, the team did not look at the average daily number of influenza cases over the three years. It focused on the extreme values recorded because it is these values that may indicate a risk of congestion for hospitals and are therefore useful for resource planning. They were able to develop a model that uses weather data to predict the risk of congestion three days later – the incubation time of the flu. "Instead of giving hospitals an average figure for expected cases, we tell them the probability that a number of cases exceeding their capacity will be reached, which is more relevant," explains Valerie Chavez, a statistician at the University of Lausanne and co-author of the study.
A warning signal
By tracking this probability each year starting in autumn, hospital officials could anticipate a spike in flu cases which could potentially lead to congestion. The model specifically indicates the number of positive cases that could be exceeded with a probability of 1%, 5% or 10%. It also indicates the maximum number of positive cases that could be observed over a 10-day or 30-day horizon. An increase in these values indicates that the epidemic is approaching its peak. "For hospitals, it's a wake-up call," the scientist sums up.
Applicable to other seasonal viruses – notably coronaviruses and respiratory syncytial virus (RSV), which causes respiratory infections in young children – the model is still subject to some uncertainty in terms of predicting risk, as only three years of data from the CHUV could be analysed to date. Also due to the lack of data, it is not yet applicable to the monitoring of SARS-CoV-2. On the other hand, researchers are already working on models which, in addition to weather data, would also draw on viral propagation processes in order to better monitor contagion.
Observing extremes to quantify risks
Extreme value theory is a statistical field that is concerned with extremely large or extremely small values in a dataset. It allows scientists to quantify risks by estimating the probability of extreme events and was first used in hydrology to calculate the necessary height of dams for flood protection. "You need a certain height to protect against a flood that occurs every 10 years, a greater height for a once-in-a-century flood, and an infinite height if you want to be protected indefinitely," Valerie Chavez explains. But this field of statistics also applies to finance, stock market crash risks or climate events such as heatwaves or melting glaciers.
Why was extreme value theory applied in this research project? "In our model, we treated flu spikes as rare, high-impact events. This is exactly the domain of extreme value theory. Models that work with mean values are based on the central values of the distribution and cannot be used to quantify risks," says Chavez.
Support for research in all disciplines
This work was supported by the SNSF's project funding scheme. After an evaluation process based on competition, the scientists with the best projects are awarded a grant to conduct independent research on a topic of their own choice.