SINGAPORE - To minimise complications after surgery, the Singapore General Hospital (SGH) is turning to a new artificial intelligence (AI) tool to help doctors assess a patient’s suitability for operations.
Developed in-house by the hospital, the Combined Assessment of Risk Encountered in Surgery-Machine Learning (CARES-ML) can trawl through a patient’s medical history, physical status and other investigative test results such as blood tests to more accurately predict the risk of post-operative complications.
It was trained using SGH’s local dataset of nearly 100,000 patients between 2015 and 2022.
Around 300 million major operations are performed each year around the world, and 16.8 per cent result in one or more complications. SGH did not provide a figure for its post-operative complications.
Previously, its anaesthetists and surgeons had to run through a patient’s medical records themselves to determine the chance of complications after surgery. This could lead to less accurate risk assessments if there is an oversight.
With CARES-ML, human error is reduced. It can predict the risk of a patient requiring intensive care or dying within 30 days of surgery with an accuracy of over 90 per cent and 80 per cent respectively.
It is also able to flag the factors that increase risk, such as body mass index and type of anaesthesia used. This helps doctors to understand the reasons behind the AI-generated risk level. The information also means doctors can better anticipate problems that arise from surgery and improve patient outcomes.
They may choose to postpone the operation, for instance, or suggest a prehabilitation programme for patients before the procedure.
With a more accurate gauge of a patient’s needs after surgery, SGH can also allocate its resources more effectively, such as prioritising which patients should be given a space in the intensive care unit.
The AI tool is being used in all surgical risk assessments at the hospital, including for high-risk and major operations such as total knee replacement and colorectal cancer surgery.
Associate Professor Hairil Rizal, senior consultant of anaesthesiology, said that CARES-ML serves as a support for doctors, rather than a substitute.
“Machine learning, and any AI model, helps in terms of providing clinicians with a second opinion. It supports the clinician by looking at more data points than a human clinician can.”
Doctors are still able to change the predicted risk level based on their experience, though this cannot deviate from the AI-generated prediction beyond a certain point.
On the issue of liability in the event of serious complications, such as death after surgery, Prof Hairil assured patients that doctors are always involved in the decision-making process. Since they are the ones determining the patient’s risk level and necessary care, the doctors are still responsible for patient outcomes.
The team of anaesthetists, surgeons and biostatisticians behind CARES-ML are looking to expand their existing model to predict other post-operative outcomes such as length of hospital stay and risk of pneumonia.
