January is the busiest month of the year for the NHS – with patients often queuing in corridors and ambulances.
In 2019 Emergency Department waiting times in England were the worst on record, with 2000 patients waiting for more than 12 hours for a hospital bed in December. At the same time latest research shows that over the past three years almost 5500 patients have died in emergency departments while waiting for a hospital bed.
Part of the problem is that patients who are admitted as emergencies to hospital can be really sick and unstable. So making the decision as to when they are getting better and are safe to go home (and the bed is free) is complicated and risky.
Indeed for every five patients sent home from hospital, one will be brought back as a medical emergency within a month. But our research might have found a way to help unblock hospital beds and help doctors and nurses know quickly which patients are safe to go home.
In our latest research, we used machine learning – or artificial intelligence (AI) – to help doctors and nurses be confident as to which patients are ready to leave hospital and which should stay. We used changes in vital signs such as blood pressure and heart rate to highlight those patients who might be well enough to leave hospital.
In the unit in Bangor where we tested this system with 790 seriously ill patients, we found that using AI in this way would have meant 2500 less days in hospital for these patients.
Reading the signs
Vital signs such as blood pressure, heart rate, speed of breathing, temperature, oxygen level, need for additional oxygen and level of consciousness are already commonly used by doctors and nurses to find out how sick someone is. These are taken two to six times a day while patients are in hospital. The more abnormal each measurement is the more likely it is that the patient may need intensive care or die.
Our new study builds on research from our unit at Wrexham Maelor Hospital, published in 2001, which tested a system that summarised all vital signs in a single number. A very similar system is now used in most ambulances and hospitals in the UK, making it easier for doctors and nurses to quickly assess a patient and communicate how seriously a patient is ill.
The system basically gives each vital sign a score between zero and three points – zero points for normal measurements and up to three points for very abnormal measurements. All points are added up: if the total score is zero the patient is likely to be well. If the total score rises the patient is at higher risk – with 20 being the highest score.
For this new study we teamed up with the Bevan Commission and researchers from health electronics company Philips Healthcare. In our study, the computer looked at vital signs of sick patients who were admitted to hospital as emergencies with medical conditions such as asthma, heart failure and stomach ulcers. The computer observed and learned the scores for two days and then started to looks at trends – analysing how often the scores went up and down and how high or low the scores went.
During the study we found that patients with total scores of three or less for more than 96 hours are usually “stable”. They were unlikely to become unwell again during the rest of their hospital stay and could most likely safely leave hospital. We calculated that implementing this simple rule alone would have saved 2143 days in hospital for the 790 patients.
But when the computer used AI it was possible to tell who the “stable” patients were much earlier – after just 12 hours. Working out which patients were ok to go home at this point would have meant 2652 less days in hospital for the patients in our study.
Deciding whether a patient can go home is complicated – and doctors and nurses often opt to keep patients in hospital longer because they are unsure whether they are really getting better. But the longer patients stay in hospital the more likely they will suffer from complications.
Indeed, it is estimated that 300,000 patients acquire infections in hospital in England alone each year. And being in hospital is particularly risky for elderly patients.
Decisions about discharge from NHS hospitals are usually done by senior hospital specialists, who see patients twice a week – though they might review patients briefly on other days. In contrast our AI system could check whether a patient is getting better several times each day.
The system we developed used data from real patients in a typical UK hospital. The data used by the team in Bangor is available at every hospital bedside in the UK, so could be easily rolled out to all hospitals. This would not only help to make hospital discharges safer but also help to make sure patients are getting the right care. And if the system can stop patients from having to queue for days in Emergency Departments or in ambulances then it might also ultimately help to save lives.
Christian P Subbe has undertaken consultancy work for Philips Healthcare. His employer BCUHB has received funding from the Bevan Commission and Philips Healthcare to undertake the research reported in this article. He is affiliated with Bangor University as a Senior Clinical Lecturer, with Betsi Cadwaladr University Health Board as a Consultant in Acute, Respiratory & Critical Care Medicine and the Health Foundation as an Improvement Science Fellow.