Understanding Patients’ Current Acceptability of Artificial Intelligence During Colonoscopy for Polyp Detection: A Single Center Study

Résumé fourni par la source

Aims Detection and accurate classification of neoplastic and non-neoplastic polyps are critical for appropriate management for patients undergoing colonoscopy. [ 1 ] Endoscopist expertise is variable and may be augmented by artificial intelligence (AI), including computer-aided detection (CADe) and diagnosis (CADx). [ 2 ] However, there are reservations about AI implementation into medical care due to concerns about performance (particularly for polyp subtypes), liability, trust, and bias. [ 3 ] [ 4 ] Data is lacking regarding patients’ perspectives. Our aim was to understand patients’ familiarity and acceptability of AI during colonoscopies. Methods In this cross-sectional survey study, we developed and administered a REDCap survey to outpatients undergoing colon cancer screening/surveillance colonoscopies at Mayo Clinic-Rochester from September to November 2023. After defining AI terminology and a brief background on the topic, patients completed the survey prior to colonoscopy. The survey included binary and 5-point Likert-scale questions related to perspectives regarding AI. Patient demographics, obtained with the survey, were summarized and responses assessed univariately using chi-square or Kruskal-Wallis rank sum test. We used multivariable logistic regression with backwards elimination to predict patients’ sentiment towards AI. Results 291 participants (44% male, mean age 59.7(range 28-83), 85.9% white) participated in the survey. Most participants were at least somewhat familiar with AI (50.9%), felt AI was at least somewhat important for physicians to utilize (59.1%), and would lead to better health outcomes (51.4%). Most participants were at least somewhat comfortable with a physician's (96.2%) or CADe's (86.9%) polyp detection, but less were at least somewhat comfortable with physicians (54.0%) or CADx (48.9%) leaving a likely benign polyp. Fewer patients (36.6% & 32.9%) were at least somewhat comfortable with physicians or CADx, respectively, in the resect & discard (R&D) strategy. Those with advanced degrees were more familiar with AI and thought it would improve health (p<0.05), and males were more comfortable with R&D (p<0.05). Interestingly, employment, white race, and a history of cancer did not substantially impact responses. Using multivariate regression, predictors of believing AI will improve health included those with comfort with CADe (p<.001) and CADx R&D (p<.009). Predictors of comfort with R&D included older age, male sex, and believing AI would improve health (all p<0.05). Similarly, those comfortable with CADx R&D had 3.58 times the odds of being comfortable with a leave in situ approach as well (p<0.001). Those undergoing their first colonoscopy were less likely to believe AI is important for healthcare (p=0.028). Those that had a personal/family history of cancer were significantly less likely to believe AI would lead to better health outcomes (OR=0.40, 95% CI: 0.17-0.90, p=0.026). Conclusions Patients are generally familiar and comfortable with AI though demographic differences exist. While many are comfortable with CADe, fewer accept CADx tools in clinical practice. Publication History Article published online: 15 April 2024 © 2024. European Society of Gastrointestinal Endoscopy. All rights reserved. Georg Thieme Verlag KG Rüdigerstraße 14, 70469 Stuttgart, Germany

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.