Managing patient flow in a hospital with a finite number of places and beds is an important and complex task.
Planned procedures have to be balanced against a capacity to treat emergency admissions in order to ensure the quality of care and reduce the number of cancelled surgeries.
The current prediction system relies on looking at the average number of beds that have been required on a particular day over the past six weeks.
Now, a new artificial intelligence (AI) tool promises to provide a much more accurate estimate of the demand for beds at any given time.
The tool, developed by researchers at University College London Hospitals (UCLH), uses patient data as soon as it is recorded.
This allows it to make four forecasts per day, rather than providing a single estimate for the day.
It can also provide a probability distribution for how many beds will be required in four and eight hours’ time, with all the predictions emailed to the relevant planners.
The team trained 12 machine learning models with patient data
The team trained 12 machine learning models with patient data collected at UCLH across a period of just over two years.
The models assessed the probability of a patient in the emergency department being admitted and requiring a bed based on a number of factors.
These included factors such as the patient’s age, how they arrived (on their own or by ambulance), the number of consultations undergone, and certain test results.
The AI made estimates based on all these probabilities, and these were compared with the actual admissions over that two-year period.
The predictions were found to be off by an average of four admissions per day, an improvement on the discrepancy of 6.5 admissions per day arrived at by the conventional method.
The system, which also takes into account planned arrivals in hospitals, was later adapted to take into account atypical patterns caused by the COVID-19 pandemic.
Researchers are now looking at refining and improving the AI so that it can estimate the number of beds required in different parts of the hospital, such as in medical wards or surgical wards.
Alison Clements, head of Operations, Patient Flow & Emergency Preparedness, Resilience & Response at UCLH, said that the next step would be to use the predictions in “daily flow huddles”.
Today’s news was brought to you by TD SYNNEX – the UK’s number one solutions distributor.
