NLP scans free-text reports to track colonoscopy quality

Natural language processing can be used to track the quality of colonoscopies and to determine the appropriate intervals between procedures, according to a study carried out by researchers at the Regenstreif Institute in Indianapolis.

One of the issues that has limited the effectiveness of colonoscopy has been the variation in performance, with the rate at which physicians detect adenomas during a colonoscopy varying significantly from one physician to another. And low adenoma detection rates are associated with subsequent colorectal cancers.

In this study, published in the American Journal of Gastroenterology, the researchers, led by Timothy Imler, MD, a Regenstrief Institute investigator and Indiana University School of Medicine assistant professor of medicine in the Division of Gastroenterology and Hepatology, wanted to see how accurately natural language processing (NLP)  can be in interpreting the complex language found in colonoscopy and pathology reports.

"The presence of adenomas in the colon is predictive of a patient's risk of later developing colon cancer, and the detection rate has been identified as the critical measure of a high-quality endoscopist, the specialist who performs colonoscopy,” said Imler in a press release. “"We found that rapid and inexpensive natural language processing, which utilizes free-text data that was previously unusable for efficient computer-based analysis, was extremely accurate in measuring adenoma detection rate during colonoscopy.”

In the study Imler and his colleagues selected 42,569 colonoscopies that generated pathology reports and were performed at 13 medical centers around the country. Seven hundred and fifty of the paired colonoscopy and pathology reports were randomly sampled and compared by human reviewers using 19 measurements associated with colonoscopy quality measures and surveillance interval determination. The remaining 41,819 paired reports were processed through natural language processing to assess their performance based on the same 19 measures.

Imler and his colleagues found that NLP was accurate regarding type and location of an adenomatous polyp 95 percent of the time compared to the human reviewers.

"Confirming that humans and the computer had similar assessments on procedures performed at facilities across the country gives us a powerful, scalable tool to assess quality and determine the appropriate interval between colonoscopies based on the procedure findings,” Imler said. “Natural language processing will enable comparison of adenoma detection rates across populations to determine geographic, racial, ethnic or gender disparities. It also could be used across health systems or colonoscopy centers or doctors to enable referers or patients themselves to make informed decisions about where they wish to go for a colonoscopy."