Researchers at UCL and UCLH have shown in a proof-of-concept study that artificial intelligence (AI) can predict which neurosurgery patients will need to go to the intensive care unit (ICU) after surgery.
At the moment, the decision about who will need an ICU stay is based on judgement by a patient's clinical team - but this judgement can sometimes be wrong.
Around 14-28% of ICU admissions are unplanned, meaning they were not anticipated by the clinical team in advance. Better prediction in advance would improve care, and the study results suggest the tool could reduce unplanned ICU admissions after surgery tenfold.
Dr Julia Ive (UCL Institute of Health Informatics) and Mr Olatomiwa Olukoya (Barts Health NHS Trust) are joint first authors of the paper, which is published in npj Digital Medicine.
The BRC at UCLH supported the work through support to last author Hani Marcus (NHNN at UCLH, and UCL).
The work was done using CogStack - software developed by UCLH BRC, UCL Institute of Health Informatics, NIHR Maudsley BRC and Kings College London.
The team used AI to read through free-text medical notes in patient healthcare records - in over 2,200 elective neurosurgery patients.
The AI attempted to predict - based on patient histories and characteristics - which patients needed ICU care after surgery.
The best AI model developed by the research team could predict who needed ICU admission with very high accuracy, identifying 89% of cases of where ICU admission was needed, but had not been planned in real life.
The results suggest the tool would be of great help if used as a planning tool ahead of surgery.
The AI’s decision making was checked by clinicians, who confirmed it was basing its decisions on medically sensible patterns.
The model performed well across sex and ethnicity groups.
Dr Ive said: “For patients, use of the model should mean safer care, fewer unexpected ICU admissions, and therefore better recovery and survival. Hospitals should see more efficient ICU use - saving money. Doctors and nurses would benefit as it would make decision making easier.”
In terms of next steps, Mr Olukoya said: “As a proof of concept study, the model needs further work before it can be used as a planning tool. The next steps would be to validate the tool at other hospitals, and incorporate it into the electronic health records system at hospitals where it is used. There is also scope to refine the AI further too.”
Image: Adobe Stock / al-sultan
