As the third leading cause of death globally and the main cause of death in hospitals, sepsis is a challenge that needs to be solved.

New research published on 22 October in Nature Medicine by Komorowski et al. demonstrated the positive contribution AI could make to improve treatment of sepsis in patients.

The AI tool, developed by researchers at Imperial College London, learnt the best treatment strategy for a patient. Through analysing the records of about 100,000 hospital patients in intensive care units and the doctor’s decisions affecting them.

The AI used reinforcement learning to analyse a range of treatment decisions made by human doctors so that it was able to suggest the most appropriate treatment for an individual patient.

Aldo Faisal, senior author from the Department of Bioengineering, Department of Computing at Imperial and an Associated Investigator at the MRC LMS, said:

“Sepsis is one of the biggest killers in the UK – and claims six million lives worldwide – so we desperately need new tools at our disposal to help patients. At Imperial, we believe that AI for Healthcare is the solution. Our new AI system was able to analyse a patient’s data – such as blood pressure and heart rate – and decide the best treatment strategy. We found that when the doctor’s treatment decision matched what the AI system recommended, they had a better chance of survival.”

Critically ill patients with sepsis are very closely monitored in an intensive care unit. Every heart beat and the treatment decisions are now recorded electronically too. Therefore, this data rich environment allowed the researchers to use reinforcement learning to find optimal treatment strategies to tackle this important health problem. It was the clinical need and amount of data collected which led to the collaboration between engineers and clinicians for this project.

There is a major unmet healthcare need around the world for better sepsis treatment. As many as 6 million people die each year from sepsis including ~45,000 in the UK. The uncertainty around better treatment strategies for critically ill patients is due to the optimal doses of fluid and vasopressor treatments requiredfor each patient. This is balanced alongside the use of antibiotics for the treatment of the infection.

The results from this research are very encouraging as the AI learnt from information and decisions made in one collection of data and was tested on entirely independent data. This was additionally supported by the observations that the best patient survival rates occurred when human doctors’ actual decisions matched the decision of the AI doctor.

The research team are yet to change any decisions on patient care, but are planning to trial the AI doctor in UK hospitals. They will compare doctors’ usual decisions to the decisions made by doctors aided by the AI doctor. The ultimate aim being the development of a ‘AI doctor’ tool which will help doctors make better more informed decisions, and improve patient care.

This project shows how the basic science of understanding behaviour is closely linked to medical science applications, an area of core interest to the LMS. The cognitive AI approach used in this project could be applied to other conditions in the future.

The research was supported by National Institute of Health Research Imperial Biomedical Research Centre and the UK Research and Innovation’s Engineering and Physical Sciences Research Council.

The full article can be read here: https://www.nature.com/articles/s41591-018-0213-5

 

Showcasing AI for Healthcare
Aldo Faisal is speaking at the Imperial Global Science Policy Forum: AI showcase on 30 October. Head of the MR Facility at the LMS, Declan O’Regan, will also be presenting how his group are using machine learning to understand heart failure in the AI exhibition. Declan’s group use a form of perceptual AI, trained on MRI heart images.