Hospitals are now utilizing “decision support tools” that are AI-powered and can diagnose illnesses, recommend treatments, or forecast the results of surgeries. But since no algorithm is perfect, how do doctors decide when to believe the AI’s advice?
According to a recent study led by Qian Yang, assistant professor of information science at Cornell’s Ann S. Bowers College of Computing and Information Science, AI tools can counsel doctors like colleagues. Moreover, it also points out relevant biomedical research supporting the decision. It helps doctors weigh the benefits of the recommendation.
In the past, many AI researchers have tried to assist physicians in evaluating recommendations from decision-support tools. They did so by describing how the underlying algorithm functions or what data was used to train the AI. However, Yang claimed that understanding how AI generates its predictions was insufficient. Moreover, many clinicians inquired about the tool’s validation in clinical trials, which is uncommon for such tools.
Yang said,
A doctor’s primary job is not to learn how AI works
If we can build systems that help validate AI suggestions based on clinical trial results and journal articles, which are trustworthy information for doctors, then we can help them understand whether the AI is likely to be right or wrong for each specific case.
The researchers first interviewed three clinical librarians and nine clinicians from a variety of disciplines to design this approach. They discovered that when clinicians differ on the best course of action, they look up outcomes from pertinent case studies and biomedical research, taking into account the caliber of each study and how closely it pertains to the issue at hand.
Yang said,
We built a system that basically tries to recreate the interpersonal communication that we observed when the doctors give suggestions to each other, and fetches the same kind of evidence from clinical literature to support the AI’s suggestion
The decision support tool’s user interface lists patient data, medical history, and lab test results on one side. In addition, the AI’s personalized diagnostic or therapy recommendations and pertinent biomedical papers In response to comments from physicians, the researchers provided a brief summary for each study that highlights information on the patient demographic and the medical intervention. Moreover, they also emphasized that patient outcomes are important so physicians may rapidly take in the most crucial details.
Yang said,
It’s a highly generalizable method,
I would hope to see it embedded in different kinds of AI systems that are being developed, so we can make them useful for clinical practice.
