LUMC first hospital with AI admissions predictor for Acute Admissions Unit
High demand at emergency departments is not a reliable indicator of the number of patients that will be sent to the Acute Admissions Unit. The LUMC has therefore developed an admissions predictor. This tool uses AI to provide instant insight into how many beds will be needed.
‘Most patients in the Acute Admissions Unit come from the emergency department’, says Britt de Boer, a nursing admissions coordinator at the Acute Admissions Unit (AAU). ‘The patients at our unit can go home or to a specialist department within 48 hours. Before we had the admissions predictor, we never knew exactly how many patients to expect. This meant we couldn’t always free up beds in time. It would sometimes seem quiet and then we’d suddenly get a call from the emergency department: “It’s busy, we’re expecting five admissions for you.”’
‘High demand at emergency departments is not a reliable indicator of the number of patients that will be sent to the Acute Admissions Unit.’
Demand not a reliable indicator
‘It is important that our beds are freed up quickly for new patients’, says Jennifer Smit, an emergency department nurse. ‘And we don’t want patients here waiting unnecessarily long either. We only have 13 beds available so only make the diagnosis and get the first treatment started. Then we refer patients to other departments, such as the AAU. High demand at the emergency department is not a reliable indicator of the number of patients that will be sent to the AAU. Some patients go straight home from the emergency department or to other hospital departments.’
AI tool: likelihood of admissions
CAIRELab, the LUMC’s AI expertise centre, developed an admissions predictor for the AAU. This tool uses AI to predict how many patients will be sent from the emergency department to the AAU. The AI model includes a range of variables from the patient’s notes such as a record of vital signs and requests for blood or radiology tests. ‘If a patient comes in, the tool makes a first prediction within 10 minutes’, Smit explains. ‘And the prediction is adjusted as soon as we enter new data, such as a lab request.’
Instant insight into new patients
‘At the AAU, the admissions predictor dashboard tells us how many patients are expected at any time from the emergency department’, says De Boer. ‘We’ve already had a capacity dashboard for some time that shows how many of our patients can be sent to other departments. The new admissions predictor completes the picture. This makes it easier for us to anticipate whether we have enough space for the expected number of new patients or need to free up beds. This makes surprises less likely. Nor do we have to constantly bother the emergency department with phone calls.’
More reliable figures
De Boer says other nursing admission coordinators are enthusiastic about the admissions predictor too. ‘It is clear how many patients we can expect in the short term. This means that in our daily “bed meetings” with other departments we have more reliable figures about how many of our patients need to be moved on. If the AAU is full, then other departments arrange beds. But we can also ensure that departments don’t unnecessarily free up beds. Sometimes if it’s busy at the emergency department, we free up beds by moving patients to other departments. Only for it to turn out later that no one comes to us from the emergency department.’
More effective use of beds
‘This admissions predictor is a great co-creation between departments, CAIRELab and the LUMC’s capacity centre’, says project leader and head of AI Marijke de Vrijes. ‘We need AI tools like this to increase efficiency in healthcare. The pressure on the number of beds and health professionals increases every year. The LUMC is the first hospital to use the admissions predictor. Nurses have told me that the prediction often matches the number of patients who are actually sent to the AAU and that this makes it quicker and easier for them to organise the department. A full evaluation will follow but the admissions predictor seems to work well for both nurses and patients.’