There was a period when the daily lives of millions of people revolved around a single number. Every afternoon attention turned to new COVID-19 cases, hospital admissions and intensive care unit (ICU) occupancy. Those figures showed where the pandemic stood, but they largely described events that had already happened. What remained uncertain was what would happen in the days and weeks ahead.
Six years after the outbreak of COVID-19, researchers continue to search for tools that can help healthcare systems move from reacting to crises to anticipating them. A new study from the University of Cyprus forms part of that effort, using artificial intelligence to estimate the course of an epidemic as much as seven days in advance.
The research draws on experience gained during the coronavirus pandemic. It does not attempt to predict when the next pandemic will emerge, nor does it seek to identify a new virus before it appears. Instead, it examines whether the progression of an infectious disease outbreak already under way can be forecast early enough to allow authorities additional time to prepare.
The study, titled Cross-Country Learning for National Infectious Disease Forecasting Using European Data, was conducted by researchers Zacharias Komodromos, Kleanthis Malialis, Artemis Kontou and Panayiotis Kolios.
Its key innovation lies in the way the model is trained. Rather than relying solely on data from a single country, it incorporates information from across Europe, searching for common epidemiological patterns that can improve the accuracy of predictions.
Learning differently
The development of artificial intelligence tools depends heavily on the quantity and quality of the data available.
For larger countries, obtaining sufficient data is not always a challenge. For smaller states such as Cyprus, however, the limited number of cases can make it more difficult to train predictive models effectively.
To overcome this obstacle, the researchers adopted a different approach. Instead of relying exclusively on Cypriot data, they gathered information from numerous European countries, allowing the model to identify patterns that recur even when countries differ in terms of population size, public health measures or outbreak intensity.
The results proved striking.
The highest level of accuracy was not achieved when the model used data only from countries with characteristics similar to those of Cyprus. Instead, forecasts improved when information from almost the entire European dataset was utilised.
The findings suggest that even countries with very different epidemiological experiences can contribute valuable insights.
In other words, the model does not look for countries that resemble one another. It looks for recurring patterns in how epidemics evolve, patterns that may be difficult for the human eye to identify when examining national datasets individually.
A seven-day head start
The research team focused on forecasting the trajectory of infections seven days in advance.
To achieve this, they found that the model performed best when it analysed data from the preceding 14 days.
Although the finding may appear technical, its practical implications are significant.
If a healthcare system can estimate with reasonable accuracy how an outbreak is likely to develop during the following week, authorities gain precious time to organise services before hospitals come under increased pressure.
Before the pressure arrives
In practice, the value of such forecasts is not measured solely by the precision of an algorithm.
It is measured by the time that can be gained.
Seven days may not sound like a long period, but during an epidemic it can be enough time to:
- Better organise staffing levels;
- Reinforce hospital units;
- Secure necessary stocks of medicines and medical supplies; and
- Introduce targeted public health measures.
The researchers do not present artificial intelligence as a replacement for epidemiologists or public health experts.
Instead, they describe it as an additional decision-support tool.
Responsibility for final decisions would remain with scientists and public authorities, who must evaluate all available evidence and determine what action is required in each situation.
The next step
The study does not represent a ready-made system already being used by health authorities.
Rather, it is a research proposal demonstrating the potential role of artificial intelligence in public health planning.
The University of Cyprus team is already looking ahead to the next phase of development.
Their objective is to expand the methodology so that it forecasts not only infection numbers but also indicators that directly affect healthcare operations, including:
- Hospital admissions;
- Intensive care unit admissions; and
- Other measures of pressure on healthcare systems.
If future studies confirm the model's effectiveness, the benefits could be considerable.
Not because artificial intelligence would replace scientists, but because it could allow them to make decisions with more information and, perhaps most importantly, with more time at their disposal.
Lessons from COVID-19
One of the clearest lessons of the COVID-19 pandemic was that every day matters during a public health emergency.
The University of Cyprus research suggests that artificial intelligence may help health systems gain those days before pressure reaches hospitals and intensive care units.
And that, perhaps, is the study's most significant message: not that AI can predict the future with certainty, but that it may help authorities prepare for it sooner.
