Portal de Periódicos da CAPES

Artificial Intelligence Improves Detection at Colonoscopy: Why Aren’t We All Already Using It?

2022; Elsevier BV; Volume: 163; Issue: 1 Linguagem: Inglês

10.1053/j.gastro.2022.04.042

1528-0012

Douglas K. Rex, Tyler M. Berzin, Yuichi Mori,

Gastric Cancer Management and Outcomes

See "Impact of artificial intelligence on miss rate of colorectal neoplasia," by Wallace MB, Sharma P, Bhandari P, et al, on page 295. See "Impact of artificial intelligence on miss rate of colorectal neoplasia," by Wallace MB, Sharma P, Bhandari P, et al, on page 295. In this issue of Gastroenterology, Wallace et al1Wallace M.B. Sharma P. Bhandari P. et al.Impact of artificial intelligence on miss rate of colorectal neoplasia.Gastroenterology. 2022; 163: 295-304Abstract Full Text Full Text PDF PubMed Scopus (70) Google Scholar report a well-designed randomized tandem study of CADe for colonoscopy. CADe refers to artificial intelligence (AI) programs that highlight precancerous lesions for detection by colonoscopists. Among 230 randomized subjects in 8 centers, the adenoma miss rate (AMR) was 15.5% when CADe was used in the first colonoscopy, and 32.4% when colonoscopy without CADe was used first. Similarly, the false-negative rate for any adenoma was 6.9% with AI first versus 29.6% with standard colonoscopy first. The AMR reduction with AI resulted from fewer misses of lesions ≤5 mm, with no difference in miss rates of 6–9 mm or ≥10 mm lesions.1Wallace M.B. Sharma P. Bhandari P. et al.Impact of artificial intelligence on miss rate of colorectal neoplasia.Gastroenterology. 2022; 163: 295-304Abstract Full Text Full Text PDF PubMed Scopus (70) Google Scholar Other tandem studies also showed AI lowers AMR.2Glissen Brown J.R. Mansour N.M. Wang P. et al.Deep learning computer-aided polyp detection reduces adenoma miss rate: a United States multi-center randomized tandem colonoscopy study (CADeT-CS Trial) [published online ahead of print September 14, 2021]. Clin Gastroenterol Hepatol.https://doi.org/10.1016/j.cgh.2021.09.009Google Scholar These studies support parallel design randomized trials showing the adenoma detection rate (ADR) increases by approximately 10% with AI,3Hassan C. Spadaccini M. Iannone A. et al.Performance of artificial intelligence in colonoscopy for adenoma and polyp detection: a systematic review and meta-analysis.Gastrointest Endosc. 2021; 93: 77-85.e6Abstract Full Text Full Text PDF PubMed Scopus (232) Google Scholar including 2 studies performed in Western populations with the same CADe software used by Wallace et al.4Repici A. Badalamenti M. Maselli R. et al.Efficacy of real-time computer-aided detection of colorectal neoplasia in a randomized trial.Gastroenterology. 2020; 159: 512-520.e7Abstract Full Text Full Text PDF PubMed Scopus (300) Google Scholar,5Repici A. Spadaccini M. Antonelli G. et al.Artificial intelligence and colonoscopy experience: lessons from two randomised trials.Gut. 2022; 71: 757-765Crossref PubMed Scopus (82) Google Scholar Thus, we can conclude that CADe increases the ADR and decreases the AMR in colonoscopy. In the current study, sessile serrated lesions were counted as adenomas. For uncertain reasons, the prevalence of sessile serrated lesions was <3%, below the expected rate with modern high-level detection skills and experienced pathology.5Repici A. Spadaccini M. Antonelli G. et al.Artificial intelligence and colonoscopy experience: lessons from two randomised trials.Gut. 2022; 71: 757-765Crossref PubMed Scopus (82) Google Scholar However, a meta-analysis reported CADe improves sessile serrated lesion detection,3Hassan C. Spadaccini M. Iannone A. et al.Performance of artificial intelligence in colonoscopy for adenoma and polyp detection: a systematic review and meta-analysis.Gastrointest Endosc. 2021; 93: 77-85.e6Abstract Full Text Full Text PDF PubMed Scopus (232) Google Scholar as did a different tandem study.2Glissen Brown J.R. Mansour N.M. Wang P. et al.Deep learning computer-aided polyp detection reduces adenoma miss rate: a United States multi-center randomized tandem colonoscopy study (CADeT-CS Trial) [published online ahead of print September 14, 2021]. Clin Gastroenterol Hepatol.https://doi.org/10.1016/j.cgh.2021.09.009Google Scholar Second, tandem studies are more often positive than parallel design studies.6Zimmermann-Fraedrich K. Pohl H. Rosch T. et al.Designs of colonoscopic adenoma detection trials: more positive results with tandem than with parallel studies – an analysis of studies on imaging techniques and mechanical devices.Gut. 2021; 70: 268-275PubMed Google Scholar In a parallel design study, endoscopist bias toward any study arm is mitigated by the clinical and medical-legal demands to protect patients from colorectal cancer in a single withdrawal. Another limitation is that detection gains for AI are largely for diminutive lesions.3Hassan C. Spadaccini M. Iannone A. et al.Performance of artificial intelligence in colonoscopy for adenoma and polyp detection: a systematic review and meta-analysis.Gastrointest Endosc. 2021; 93: 77-85.e6Abstract Full Text Full Text PDF PubMed Scopus (232) Google Scholar This is generally true for detection gains from ancillary devices, because powering trials for improved advanced lesion detection is not practical. Better detection of advanced adenomas was shown in a meta-analysis that included larger numbers of AI trials.3Hassan C. Spadaccini M. Iannone A. et al.Performance of artificial intelligence in colonoscopy for adenoma and polyp detection: a systematic review and meta-analysis.Gastrointest Endosc. 2021; 93: 77-85.e6Abstract Full Text Full Text PDF PubMed Scopus (232) Google Scholar However, although increases in the ADR associated with endoscopist education decrease post colonoscopy colorectal cancer risk,7Kaminski M.F. Thomas-Gibson S. Bugajski M. et al.Performance measures for lower gastrointestinal endoscopy: a European Society of Gastrointestinal Endoscopy (ESGE) quality improvement initiative.United European Gastroenterol J. 2017; 5: 309-334Crossref PubMed Scopus (131) Google Scholar it is not yet proven that device associated ADR gains, including gains from AI, will decrease the post colonoscopy colorectal cancer risk. A recent microsimulation model found CADe during screening colonoscopy produced additional 11% and 7% relative decreases in CRC incidence and mortality compared to screening colonoscopy without CADe,8Areia P.M. Mori Y. Correale L. et al.Cost-effectiveness of artificial intelligence for screening colonoscopy: a modelling study.Lancet Digital Health. 2022; 4: e436-e444Abstract Full Text Full Text PDF PubMed Scopus (60) Google Scholar but real-world observational data of the type available for endoscopist education7Kaminski M.F. Thomas-Gibson S. Bugajski M. et al.Performance measures for lower gastrointestinal endoscopy: a European Society of Gastrointestinal Endoscopy (ESGE) quality improvement initiative.United European Gastroenterol J. 2017; 5: 309-334Crossref PubMed Scopus (131) Google Scholar are still lacking for AI. If CADe increases the ADR and CADe is commercially available in both the United States and Europe, why is the buzz that early uptake in both regions is slower than expected? Several considerations seem likely relevant. First, other adjunctive detection devices have received approval from the US Food and Drug Administration and then failed to reach widespread use. Examples include ultra-wide angle endoscopy [full spectrum endoscopy, Third Eye Retroscope (Avantis Medical Systems, San Jose, CA)], Endocuff, Endocuff Vision (Olympus America Corp, Center Valley, PA), and distal colonoscope hoods. Although each device had certain unique barriers to uptake, the limited adoption of this entire category suggests that physicians attach a relatively low price point to the value of detection gains produced by add-on devices. Incorporating detection devices with add-on cost is particularly problematic in US ambulatory surgery centers and office practices, where the device cost is born by the practice. Short term profit reductions from purchase of detection devices, which may be offset by higher polypectomy charges and future increased procedure volume (owing to shorter surveillance intervals), seems to not resonate with colonoscopists in these settings. The explanation for low valuation of add-on detection devices likely lies in the common reimbursement models for performing colonoscopy. The most common model is fee-for-service payment per procedure completed, with extra payment per patient with ≥1 polyp or in some settings per polyp removed. There is generally no reimbursement for specifically using image-enhanced endoscopy, chromoendoscopy, or add-on devices. For now, this is also the case for AI. As with some other adjunctive detection devices, AI detection gains and the associated potential reduction in cancer care costs indicate that CADe is cost effective8Areia P.M. Mori Y. Correale L. et al.Cost-effectiveness of artificial intelligence for screening colonoscopy: a modelling study.Lancet Digital Health. 2022; 4: e436-e444Abstract Full Text Full Text PDF PubMed Scopus (60) Google Scholar and warrants reimbursement. However, it is widely understood that the path to reimbursement of add-ons is long and complicated. Developers of AI should acknowledge the limited incentives for endoscopists to use these devices. Second, CADe is generally being sold as an add-on to standard endoscopic equipment. This is a substantial change from previous electronic tools that alter or improve the white light endoscopic image. For example, high definition increases the ADR,9Tziatzios G. Gkolfakis P. Lazaridis L.D. et al.High-definition colonoscopy for improving adenoma detection: a systematic review and meta-analysis of randomized controlled studies.Gastrointest Endosc. 2020; 91: 1027-1036.e9Abstract Full Text Full Text PDF PubMed Scopus (26) Google Scholar and is now included as standard in commercial reusable colonoscopes. Electronic chromoendoscopy has been included routinely as part of commercial colonoscopes since its introduction. This made electronic chromoendoscopy available to all colonoscopists from its inception, and although many colonoscopists seldom use it, the uses and limitations of this technology were sorted out without endoscopists paying "extra" for using it, because the technology was a standard endoscope feature. As the functionality of AI expands beyond detection (Table 1), AI tools may seem more indispensable and worth separate purchase. In the meantime, the add-on cost of AI is easier to turn down when AI only offers CADe, with its lower valuation. The situation hints that an endoscope company could gain advantage by offering AI and AI improvements as standard equipment with their endoscopes.Table 1Expected Functions of Artificial Intelligence in ColonoscopyCADe – highlighting lesionsCADx – predictions of histologyCAQ – assessment of quality of examination technique and bowel preparationSemi-automated completion of procedure report Open table in a new tab This discussion of the seeming slow early uptake of CADe in clinical practice is speculative. Of the various potential drivers for slow uptake (Table 2), perhaps surveys of endoscopists can sort the relative importance. Although the scientific achievement of AI is exciting, both developers and endoscopists should keep in mind that AI is fundamentally just another tool for colonoscopists, and one that colonoscopists currently receive no direct payment for using. Many colonoscopists seem to understand this point already.Table 2Potential Reasons for Slow Early Uptake of Commercial CADe DevicesNo legal or regulatory mandate to measure ADR in many countries, regionsEndoscopists measuring ADR believe their ADR is adequate or limited benefit in increasing ADRHigh cost of CADeCosts of CADe are immediate, return on investment is uncertain, delayedCADe systems sold as add-onsLack of direct reimbursement for AIEndoscopists view current systems as early in technical developmentEndoscopists expect prices to drop with increased competitionEndoscopists sense little pressure to purchase from patients or local competitorsEndoscopists waiting to see what experts doEndoscopists believe AI competes with physician expertiseEndoscopists mistrust AI development, monitoring processesADR, adenoma detection rate; AI, artificial intelligence. Open table in a new tab ADR, adenoma detection rate; AI, artificial intelligence. Impact of Artificial Intelligence on Miss Rate of Colorectal NeoplasiaGastroenterologyVol. 163Issue 1PreviewWhen the endoscopist is assisted by artificial intelligence, the risk of missing colorectal neoplasia decreased substantially, reassuring a better stratification of the neoplastic risk of the individual patient, and a higher degree of colorectal cancer prevention. Full-Text PDF Open Access