Predicting the Unpredictable – How Data is Reshaping Pandemic Healthcare Planning

From Chaos to Control: How Multivariable and Simple Models Helped Hospitals Stay Ahead of COVID-19

When COVID-19 swept through the world, hospitals faced an overwhelming surge in admissions. Decision-makers needed accurate, timely data to avoid critical bed shortages and staffing crises. Behind the scenes, a pragmatic predictive model built using real-time COVID-19 case numbers and vaccination rates provided Northwest London actionable insights to NHS leaders twice a week.

Turning Data into Action

A study by researchers from our Information and Intelligence Theme  revealed that simple and multivariable models can be highly effective in predicting hospital bed occupancy. During periods of stability, the multivariable model achieved a mean absolute percentage error (MAPE) as low as 10.8%, making it a valuable decision-support tool. Predictions from the model helped open and close COVID-19 wards strategically, ensuring resources were allocated where they were needed most.

However, when the Omicron variant emerged, the model’s accuracy declined, with MAPE increasing to over 20% in some cases. Predictive models must be continuously adapted to evolving public health crises, as demonstrated during Omicron's rapid surge in cases. 

A Model for Rapid Response

Unlike complex epidemiological simulations, this model was designed for real-world use by health system analysts with limited modelling experience. It relied solely on routinely collected NHS data, making it scalable, cost-effective, and adaptable for future pandemics or seasonal healthcare pressures. The model incorporated age-stratified hospitalisation rates, recognising that older adults faced a disproportionate risk of severe COVID-19. This allowed NHS planners to anticipate surges in vulnerable populations and ensure appropriate levels of care. However, the model faced a challenge due to multicollinearity, which is when some of the age groups in the model showed very similar patterns of COVID-19 cases. This meant the model had difficulty separating the effect of age on hospitalisation because the data from these age groups were closely related. In simpler terms, the model couldn’t distinguish the influence of similar factors because they were too closely linked. This made the predictions less reliable in certain periods, especially when new variants emerged, like Omicron.

A Case for Healthcare

The study underscores the need for sustained investment in healthcare data analytics. With further refinements and ongoing support, models like this—both simple and multivariable—could be expanded to predict winter hospital pressures, optimise ambulance dispatching, or even anticipate primary care demands. This highlights how adaptable data models empower NHS decision-makers to act swiftly and effectively. Predictive analytics should not be seen as a luxury but as a core tool for building a resilient health system that anticipates crises rather than merely reacting to them.

Call to Action

The NHS must invest in scalable, real-world predictive models to fully realise the potential of data-driven healthcare planning. The challenge is clear: to future-proof the health system, we must act now to embed these tools into routine healthcare operations, ensuring they can be swiftly deployed during emerging health crises and effectively support decision-making when needed most.