Automated tool determines colonoscopy surveillance intervals
Click Here to Manage Email Alerts
LAS VEGAS — Researchers have developed a natural language processing algorithm which has proven effective at determining post-polypectomy colonoscopy surveillance intervals, according to a presenter at the ACG Annual Scientific Meeting.
“We successfully developed an automated pipeline that uses natural language processing to automatically determine the appropriate protocol effective surveillance interval from free text colonoscopy and free text pathology reports,” Camille Soroudi, MD, first-year gastroenterology fellow at the David Geffen School of Medicine at University of California, Los Angeles, said during her presentation. “A more immediate use of this tool that we hope to implement within our own health care system is to use it to track surveillance intervals across our healthcare system and to help ensure that we're meeting certain quality metrics, such as the ASGE and ACG taskforce goal of having greater than 90% of screening colonoscopies with documented guideline supported surveillance interface.”
Soroudi told Healio a better job needs to be done to ensure patients regularly get screened on time.
“Our tool can help accomplish this by assisting providers when determining the appropriate surveillance interval and allowing health systems to track how often patients undergo timely follow-up after screening colonoscopy,” Soroudi said. “Use of guideline-concordant colonoscopy surveillance intervals is one of the ASGE/ACG colonoscopy quality indicators, and our tool can help health systems meet this recommendation.”
The natural language processing algorithm works by taking in data, such as polyp number, size and histology, and then extracting and structuring this data for system analysis in order to determine the appropriate post-polypectomy surveillance interval.
In a retrospective manual chart review, researchers analyzed the data of 469 patients (50.3% women; mean age, 57.9 years) who underwent average-risk screening colonoscopies from June 2020 to February 2021. The natural language processing algorithm was then utilized, and its performance was compared to that of the manual chart review.
The algorithm was determined to have a sensitivity of 82.6% and a specificity of 98.3% in identifying the appropriate surveillance interval after a screening colonoscopy, with an overall accuracy of 97.6%. Of the 469 cases, 58 were misclassified by the algorithm.
“Among the 58 cases misclassified by automated system, the majority (75.9%; n = 44) were allocated to a shorter surveillance interval than the manual gold standard, which is reassuring that the current algorithm would not contribute to delays in colorectal cancer screening follow-up. We expect that we will be able to improve the system further with additional rounds of adjustments to the algorithm,” Soroudi told Healio.
“We think this is an important tool that has many applications looking forward. It's possible this may be able to be integrated into the healthcare system to allow us, as a decision aid tool, to help physicians determine the appropriate surveillance interval after a screening colonoscopy,” she concluded in her presentation.
Collapse
Soroudi C. Validation of machine learning tool to determine post-polypectomy surveillance intervals after screening colonoscopy in a large health system. Presented at: ACG Annual Scientific Meeting; Oct. 22-27, 2021; Las Vegas (hybrid meeting).
Read more about
Click Here to Manage Email Alerts