A learning agent called the Artificial Intelligence (AI) Clinician is presented in a paper published online this week in Nature Medicine. The AI Clinician could help doctors make better decisions in real time to improve the outcomes of patients with sepsis.
Sepsis - a life-threatening condition in which the body’s extreme response to severe infection damages tissues and organs - is the third leading cause of death worldwide and the main cause of mortality in hospitals. Critical to treating sepsis is the correct administration of fluids and medicines to help a patient keep their blood pressure stable. However, current clinical practices are suboptimal.
To help improve clinical decision-making, Aldo Faisal, Anthony Gordon, Matthieu Komorowski and colleagues developed the AI Clinician - an artificial intelligence agent that learns optimal treatment approaches for sepsis by analysing thousands of real-life treatment decisions made by doctors. The AI Clinician uses reinforcement learning, a type of machine learning common to economics and game theory, to extract useful information from a collection of patient data that exceeds manyfold the lifetime experiences of human doctors.
The authors find that the AI Clinician is able to select the optimal treatment for patients with sepsis more reliably on average than human doctors. Moreover, in a large validation cohort independent of the training data, mortality rates were lowest in patients whose actual, clinician-prescribed dose matched the AI’s recommendation.
This work will require evaluation using real-time data and decision-making in clinical trials alongside testing in different healthcare settings. However, the authors note that if only a small percentage reduction in sepsis mortality rates could be achieved, this would equate to several tens of thousands of lives worldwide being saved annually.