Monday, November 30 | 9:30 a.m.-10:30 a.m. | M3-SSIN02-8 | Room E450B
How radiologists utilize AI and how they document that use could sway jurors in a malpractice case.
Presenter Elizabeth Song of Brown University and colleagues examined how different AI-radiologist workflows affect perceived malpractice liability, and which factors might reduce it when diagnostic errors occur.
In an online experiment, 2,347 U.S. adults acted as hypothetical jurors in a case involving a missed intracranial hemorrhage. Participants were randomized to either a no-AI control or one of 21 AI scenarios. The scenarios varied by factors such as single- versus double-read workflow, whether the radiologist documented an initial interpretation, and whether jurors were told the AI’s error rates.
Double-read workflows, documented initial interpretations, and disclosure of AI error rates all lowered perceived liability. Liability was highest, however, when radiologists initially spotted the abnormality but then changed their interpretation after the AI missed it, according to the researchers. In those scenarios, more than 80% of participants found the radiologist liable, compared with 62% in the control group. The researchers call this effect an “erroneous change penalty.”
“Radiologist liability is strongly influenced by workflow design, documentation practices, and inclusion of AI error rates,” the researchers wrote. “Double-read workflows with documented initial interpretations and disclosure of AI error rates reduce perceived liability, though changing a correct initial interpretation increases it.”
Take in this Monday talk to hear about a proposed workflow that reconciles these findings.
