“Hedge” statements were found in nearly a third of radiology reports using a computer-based review system described in an award-winning paper presented Tuesday at the RSNA meeting.
"Hedge" statements were found in nearly a third of radiology reports using a computer-based review system described in an award-winning paper presented Tuesday at the RSNA meeting.
The system, developed by Stanford researchers, used natural language processing to comb through more than 2.8 million sentences from 900,600 radiology reports prepared over a five-year period ending in 2009. It found uncertainty or hedge statements in 30.9% of the reports. It also probed for recommendations, finding them in 19.2% of the reports.
The uncertain statements were higher in reports by residents than when the same report was dictated by an attending radiologist alone. But when a resident dictated a report for a radiologist, the uncertainty level went from 26% to 38%.
The paper was presented by Dr. Bao Do. He was given a young researcher award by the RSNA.
A validation process for the natural language processing program found its sensitivity and specificity at 97.1% and 99.6%. The natural language processing program probed only the impression of the report, not the findings.
Do gave two examples of sentences that contained uncertainty: "Findings could represent pneumonia," and "Right breast BI-RADS 3."
"This is important because these expressions of uncertainty, or hedging, if captured by computer algorithms and extracted, can potentially be used for billing, quality control, and decision support," Do said.
Said one member of the audience, "It's nice to finally have validation for what everyone has known is our favorite plant, the hedge."
