S412 Perceived Advantages and Disadvantages of Adopting Real Time Artificial Intelligence in Colonoscopy by Providers: A Systematic Review
Bibliographic record
Abstract
Introduction: Although colonoscopy is the gold standard screening modality for colorectal cancer, it is operator-dependent with up to 26% adenoma miss rate (AMR). Using artificial intelligence (AI) in colonoscopy and computed aided detection (CADe), may decrease AMR and improve clinical outcomes; however, uptake of this technology relies on providers’ perspective. We aimed to explore the evidence on providers’ perspective towards CADe. Methods: We performed a systematic review of Cochrane, Google Scholar, Ovid MEDLINE and Embase, PubMed Scopus, and Web of Science to find studies that reported perspectives of providers on AI for colonoscopy. Perspective items included: interest/satisfaction and perceived advantages and concerns for using CADe. In each study, we ranked items based on the proportion of participants voting for them. We grouped the five top-ranked items across studies related to similar advantages or disadvantages into themes. Themes were next sorted based on the number of studies voting for the items (Nv) included in each theme. Additionally, the proportion of participants with a positive response to the questions of interest were combined across studies to report pooled rates. Risk of bias assessment was done using the Joanna Briggs Institute critical appraisal checklist. Results: After screening 1619 titles, we included 7 studies (774 providers) from 4 countries. Majority of the participants were gastroenterologists or gastroenterology fellows. Proportion of participants who were familiar with AI (12%-100%) or used it (25%-100%) were varied across studies. The themes including items related to improved diagnosis (Nv = 5) and improved efficiency (Nv =4) were sorted as top-ranked advantages themes perceived by the providers. On the other side, themes related to lack of responsibility for misdiagnosis/medicolegal concern and cost (Nv =5 for both) were sorted top-ranked disadvantages themes (Table 1). Our meta-analyses revealed 52% (95% confidence interval [CI]: 24-79%, 244/399 responders from five studies) and 38% (95% CI: 9-73%, 132/275 responders from four studies) perceived cost and lack of accountability for misdiagnosis as main concerns for using CADe, respectively (Figure 1). Conclusion: Majority of the providers with different levels of experience and familiarity with AI and from diverse practice settings expressed interest in using CADe. Cost and accountability for misdiagnosis were perceived as main concerns among providers (see Table 1).Figure 1.: Meta analysis of perceived disadvantages of using computed aided detection: cost (A) and lack of responsibility for misdiagnosis (B). Table 1. - Top perceived disadvantages and barriers for adopting artificial intelligence-assisted colonoscopy Study Rank 1 2 3 4 5 Nehme et al 2023 Pre-implementation Too many false-positive signals (69%) Unnecessarily longer procedure time (37%) Too distracting (25%) Not worthwhile improvement in ADR (25%) Medicolegal concern / Too expensive (both 12%) Post-implementation Too many false-positive signals (82%) Too distracting (59%) Prolonged procedure time (47%) Audio beep too load (41%) Only found obvious lesions (12%) Van der Zander et al 2022 Insufficiently developed IT infrastructure (56%) Lack of (technical) knowledge by physicians (50%) Responsibility (uncertainty about laws and regulations) (35%) Costs (25%) Lack of human supervision (25%) Tian et al 2022 Less responsibility for medical negligence Wadhwa et al 2020 Cost (75%) Operator dependence (63%) Increased procedure time (60%) Higher number of false positive detections (34%) Kader et al 2022 Lack of guidelines (92%) Access to AI devices (89%) Availability of Devices with regulatory approval (88%) Accountability for incorrect diagnosis (85%) Evidence for cost-effectiveness (84%) Nazarian et al 2023 Cost (64%) Accessibility (56%) Lack of guidelines (51%) Lack of research (35%) Data ownership (27%) Kochar et al 2021 Replacement of physicians by machines (37%) Data/patient information security (32%) Become less efficient in caring for patients (18%) Increase workload (12%) Make physicians obsolete in caring for patients (8%)
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.018 | 0.077 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.007 | 0.008 |
| Bibliometrics | 0.009 | 0.011 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.001 |
| Insufficient payload (model declined to judge) | 0.007 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".