Will machines decipher colonoscopy quality from endoscopists’ notes?

Colonoscopy has been shown to reduce incidence and mortality of colorectal cancer; however, its effectiveness is highly dependent on the quality.1Zauber A.G. Winawer S.J. O’Brien M.J. et al.Colonoscopic polypectomy long-term prevention colorectal-cancer deaths.N Engl J Med. 2012; 366: 687-696Crossref PubMed Scopus (1744) Google Scholar, 2Pilonis N.D. Bugajski M. Wieszczy P. al.Long-term cancer after a single negative screening colonoscopy.Ann Intern 2020; 173: 81-91Crossref (19) 3Shaukat A. Mongin Besser M.S. for cancer.N 2013; 369: 1106-1114Crossref (515) Scholar Therefore, it widely recognized that quality assessment, assurance, improvement tools in colonoscopy are essential ensure effectiveness. There clearly defined measurable indicators such as endoscopist’s adenoma detection rate (ADR), adequate bowel preparation, cecal intubation rate, mean withdrawal time, which have proved be associated with important outcomes patients.4Kaminski M.F. Regula J. Kraszewska E. al.Quality risk interval 2010; 362: 1795-1803Crossref (1224) Scholar,5Rex D.K. Schoenfeld P.S. Cohen colonoscopy.Am Gastroenterol. 2015; 110: 72-90Crossref (246) However, collecting data these clinical setting often accompanied by an additional administration burden, mostly because lack structured reporting system linking from multiple sources. For colonoscopy, particularly high-volume procedure, manual outside normal healthcare process unfeasible. In this issue Gastrointestinal Endoscopy, Laique al6Laique S.N. Hayat U. Sarvepalli S. al.Application optical character recognition natural language processing large-scale metric extraction reports.Gastrointest Endosc. 2021; 93: 750-757Abstract Full Text PDF (6) demonstrate use 2 technologies, (OCR) (NLP), extract variables scanned procedure pathology reports image format accuracy >95%. Analysis 589 records 262 corresponding center showed when compared extraction, hybrid OCR/NLP approach achieved nearly 100% detecting presence polyps, adenomas, sessile serrated advanced inadequate failed intubation. The obtained enabled calculation following results: ADR 23.1%, preparation 83.1%, 98.4%. OCR technology extracts converts images into unstructured text. NLP method electronically analyzing extracting information text.7Harkema H. Chapman W.W. Saul al.Developing application measuring procedures.J Am Med Inform Assoc. 2011; : i150-i156Crossref (51) Several studies evaluated role assessing examination quality, showing high quality-related variables. applicability was limited need costly expert programming their development validation at center.8Imler T.D. Morea Kahi C. al.Multi-center measurement utilizing processing.Am 543-552Crossref (40) 9Mehrotra Dellon E.S. Schoen R.E. al.Applying tool electronic health assess performance measures.Gastrointest 75: 1233-1239.e14Abstract (54) 10Raju G.S. Lum P.J. Slack R.S. al.Natural alternative metrics.Gastrointest 82: 512-519Abstract (38) Recently, Lee al11Lee J.K. Jensen C.D. Levin T.R. al.Accurate identification polyp findings using processing.J Clin 2019; 53: e25-e30Crossref (9) commercially available tools, did not require background, were adaptable report formats. These accurate identifying key related across medical centers different formats.11Lee Each year, >19 million colonoscopies performed United States.12Gastrointestinal Endoscopes Market Analysis, States, 2017-2023, MedCore.https://idataresearch.com/product/gastrointestinal-endoscopes-market-analysis-united-states-2017-2023-medcore/Date: 2016Google Because crucial procedures meet criteria quality. national setting, lacks central automated registry heterogenous sources data, will extremely difficult. initial results, show great potential expedient reports, promise create new frontier collection indicators. can times unreliable inaccurate, added complication varying formats may result low-quality text outputs. Poor make difficult subsequent derive meaningful insights analysis. Another huge financial investment programming. This leads consideration whether evolving machine learning endoscopic systems could panacea problems. Let’s imagine objectively during endoscopy, without bias resulting endoscopists’ self-reporting. Recognition cecum artificial intelligence (AI) enough completeness colonoscopy. If calculate proportion clean mucosal surface, scoring would no longer task. Finally, AI effective histologic diagnosis detected polyps.13Lui T.K.L. Guo C.G. Leung W.K. Accuracy histology prediction polyps: systematic review meta-analysis.Gastrointest 92: 11-22.e6Abstract (23) Scholar,14Shahidi N. Rex D.R. Kaltenbach T. al.Use impression, intelligence, pathologist interpretation resolve discrepancies between endoscopy analyses diminutive polyps.Gastroenterology. 158: 783-785.e1Abstract (14) future improvements used automatically ADR. incorporated endoscopes record digitalized database overcome burdensome implementation software systems. we cannot let perfection enemy progress. Multiple commercial parties, including leading manufacturers many renowned research groups, currently developing computer-aided control all introduced game-changers field endoscopy; yet, results detection15Hassan Spadaccini Iannone al.Performance detection: 77-85Abstract (31) control16Su J.R. Li Z. Shao X.-J. al.Impact real-time automatic prospective randomized controlled study (with videos).Gastrointest 91: 415-424.e4Abstract (63) still scarce. Bench-to-bedside requires strong multidisciplinary collaboration. To avoid production “black-box” supporting scientific evidence, lengthy collaboration scientists parties essential. should also focus engaging less endoscopists. meantime, millions unknown before revolutionary solution possible, going back existing notes seems good compromise. Dr Kaminski recipient speaking consultation fees Olympus; Fujifilm, Boston Scientific, Medtronic, Norgine; travel grants Alfa Sigma ERBE; advisory board Sigma. other author disclosed relationships. Application reportsGastrointestinal EndoscopyVol. 93Issue 3PreviewColonoscopy commonly States. Reports generated non-standardized always integrated records. Thus, readily streamlining management, participating registries, or patient- center-specific factors predictive outcomes. We aim charts elucidated (OCR/NLP hybrid) obtain relevant co-developed Cleveland Clinic eHealth Technologies (West Henrietta, NY, USA). Full-Text