Forecasting the daily number of near-term Covid-19 admissions has been a vital part of hospital planning during the pandemic.
New research has now shown that artificial intelligence (AI) and mobile network data can help deliver an accurate forecasting model of near-term Covid-19 patient admissions 14-to-21 days ahead of time.
The AI model developed by telecommunications companies Ericsson, and Telia was put to trial in Swedish hospitals.
The AI experts first developed and tuned predictive models through a training phase to identify how each predictive model performed across various forecasting windows.
Once the best performing predictive models and settings were identified, the experts retrained the models using all data, producing a reliable 14-to-21 day Covid-19 admissions forecast, which could be delivered every week.
The predictive models are also tuned to consider external factors such as vaccination data.
The forecast indicators produced throughout the trial have been considered, combined, and weighed against multiple other sources of information to form the basis of the hospitals’ Covid-19 prognoses.
Other relevant indicators are the fraction of positive PCR tests, the change in commute pattern (using data from the Vasttrafik transport authority), and the change in the overall mobility in heavy-traffic areas such as shops and offices.
Its “a reliable tool for short term prediction of admissions to the hospital due to Covid-19 and has been vital for the planning of effective resource allocation to ensure that all patients receive the care needed”, said Thomas Brezicka, Chief Medical Officer, Sahlgrenska University Hospital.
“In this past year, our hospitals and healthcare resources have been stretched to their very limits, and it’s been important that we join to support in every way we can. This trial has demonstrated the lasting impact that disruptive technologies such as AI can have across our societies,” added Peter Laurin, Head of Managed Services, Ericsson.
