Why is low oxygen during surgery a concern?
During an operation under general anaesthesia, anaesthetists watch the patient's condition continuously. One of their goals is to avoid hypoxaemia, meaning too little oxygen in the blood.
Anticipating such an episode is still hard: according to the authors, reliably forecasting a future hypoxaemia is not currently possible. Seeing the problem coming a few minutes earlier would leave time to act.
Being able to anticipate rather than react is exactly the authors' stated aim: to gain time before oxygen levels fall.
How was the system built?
The team used machine learning: a program that finds regularities in large amounts of data. It was trained with minute-by-minute data from the electronic medical records of over fifty thousand operations.
The key point is that it does not only give a risk: it gives explanations. It shows which features of the patient or the operation push the risk up, so the anaesthetist can understand and check what it says.
For example, the system can show that certain features of the patient or of the procedure raise the risk at a given moment, which gives the team a concrete lead to act on rather than a bare alarm.
What did the study show?
The system improved the performance of anaesthetists when they were shown interpretable risks with their contributing factors. The explanations it gives broadly agree with the medical literature and with the anaesthetists' own knowledge.
The authors offer an order of magnitude: if anaesthetists currently anticipate 15% of hypoxaemia events, with the system's help they would anticipate 30%. A large share of these episodes may benefit from early action, because they are linked to factors that can be changed.
In other words, the study does more than signal danger: it helps understand which changes in risk are caused by patient or procedure characteristics, which can enrich the understanding of hypoxaemia during anaesthesia.
Key figures
- 50,000+operations whose records were used to train the system
- 15% → 30%share of episodes anticipated by anaesthetists, without then with the system (authors' estimate)
What does this change for patients?
The goal is to help the anaesthesia team see low oxygen coming before it happens, and so act sooner. The system can also give general insight into how certain patient or procedure characteristics change the risk.
The anaesthetist stays in charge: the tool is an aid to vigilance, not an autopilot.
Used well, such a tool gives the team more information at the bedside without adding a new task: the risk and its reasons are simply shown alongside what they already watch.
What are the limits?
The 15% and 30% figures are a projection by the authors (an "if" statement), not the result of a trial in operated patients. The system was built and tested from past records.
It remains to be checked, in real operating theatres, whether using it live truly reduces hypoxaemia episodes and improves patient safety.
Before an anaesthetic, ask your questions, with Mima
This research system helps the anaesthetic team see a drop in oxygen coming during surgery. The team monitors and acts: you have nothing extra to do, but you can be informed.
Before an operation, Mima gathers your questions for the anaesthetic consultation and keeps track of your prescriptions. You can ask: "Does my medical history change how you will monitor me under anaesthetic?" She also explains the words in your notes.
Frequently asked questions
What is hypoxaemia during surgery?
It is too little oxygen in the blood. Anaesthetists try to avoid it, and this system is designed to help them see it coming.
Does the AI replace the anaesthetist?
No. In the study, it improves anaesthetists' performance by giving them a risk and its reasons. The decision stays medical.
Why does the explanation matter so much?
Because a doctor cannot rely on an alert without understanding where it comes from. The system's explanations broadly match what anaesthetists already know.
Is it used in operating theatres today?
The study does not say so. It describes the development and testing of the system, and the authors discuss what it could bring.
The original study
Lundberg SM, Nair B, Vavilala MS, et al. “Explainable machine-learning predictions for the prevention of hypoxaemia during surgery.” Nature Biomedical Engineering, 10 October 2018. DOI : 10.1038/s41551-018-0304-0
Read the full study in Nature Biomedical EngineeringThis article summarises a scientific study and is not medical advice.