Gastroenterology AI Standard of Care: Colonoscopy AI, Polyp Detection, and Liability

AI at the Scope: The New Frontier of GI Liability

#

Gastroenterology has become the second major clinical frontier for AI in medicine, following radiology. With multiple FDA-cleared computer-aided detection (CADe) systems now in routine use during colonoscopies, endoscopists face novel liability questions: What happens when AI misses a polyp that becomes cancer? What if AI misclassifies a polyp, leading to inadequate follow-up? And critically, does AI assistance create a new standard of care that makes non-AI colonoscopy legally indefensible?

This guide examines the emerging standard of care for AI use in gastroenterology: what FDA clearance means for colonoscopy AI, how liability is allocated between endoscopists and AI systems, what the American Gastroenterological Association recommends, and how GI practitioners can protect themselves while leveraging AI’s genuine benefits.

FDA-Cleared Colonoscopy AI: The Current Landscape

#

Timeline of Major Approvals

#

What These Systems Do

#

Computer-Aided Detection (CADe):

• Real-time video analysis during colonoscopy
• Highlights suspected polyps with visual markers
• Alerts endoscopist to areas requiring attention
• Does NOT make diagnostic decisions

Computer-Aided Diagnosis (CADx):

• Predicts polyp histology (hyperplastic vs adenomatous)
• Still emerging, not yet standard of care
• Could enable “diagnose and leave” strategies
• Higher liability exposure than CADe

What FDA Clearance Does NOT Mean

#

As with radiology AI, FDA 510(k) clearance for colonoscopy AI does NOT guarantee:

No Generalizability Proof:

• Performance may vary across patient populations
• Training data may not represent diverse populations
• Real-world performance often differs from trial conditions

No Comparative Superiority:

• Cleared as “substantially equivalent” to predicate device
• Does not mean AI outperforms expert endoscopists
• Individual endoscopist skill still matters

No Outcome Improvement Guarantee:

• Detects more polyps, but impact on cancer prevention unclear
• Increased detection may include clinically insignificant lesions
• Long-term mortality benefit not yet established

Clinical Evidence: What the Research Shows

#

The 2024 Meta-Analysis

#

A comprehensive meta-analysis published in the Annals of Internal Medicine (2024) by Soleymanjahi et al. provides the strongest evidence base:

Scope:

• 44 randomized controlled trials
• Comparing standard colonoscopy vs CADe-assisted colonoscopy
• Platforms: GI Genius, Fujifilm CAD EYE, EndoScreener, YOLO-based systems

Key Findings:

• CADe significantly increases adenoma detection rate (ADR)
• Improved detection of sessile serrated lesions
• Little to no adverse events directly attributable to CADe use
• Evidence supports conditional recommendation for CADe use

Limitations and Gaps

#

Despite positive findings, significant gaps remain:

AGA Guidelines: The Emerging Standard of Care

#

2024-2025 Draft Clinical Guidelines

#

The American Gastroenterological Association (AGA) released draft clinical guidelines on AI-assistance in colonoscopy:

Conditional Recommendation:

• AGA conditionally recommends CADe for polyp detection in adults
• “Conditional” reflects still-developing evidence base
• Acknowledges potential benefits while recognizing limitations

Key Guideline Points:

1. CADe is appropriate for routine screening colonoscopy
2. Not required to meet standard of care (yet)
3. Training recommended before clinical use
4. Documentation of AI use should be standard practice
5. Human judgment remains primary:AI is advisory

Implementation Requirements

#

If using CADe, the AGA and general medical AI standards require:

Pre-Deployment:

• Ensure equipment meets manufacturer specifications
• Train all operators on system use
• Establish protocols for AI alerts
• Inform patients about AI assistance

During Procedure:

• Use AI as intended by manufacturer
• Maintain standard withdrawal times
• Don’t over-rely on AI for comprehensive inspection
• Document AI alerts and responses

Post-Procedure:

• Report outcomes for quality monitoring
• Track adenoma detection rates with/without AI
• Report adverse events or failures to vendor and FDA

The Liability Framework

#

The Core Liability Scenario

#

A detailed liability scenario from Clinical Gastroenterology and Hepatology (2024) illustrates the emerging risk:

A 50-year-old patient presents for screening colonoscopy. The endoscopist identifies polyps, and a CADx tool provides a “hyperplastic” diagnosis prediction. Based on this prediction and the endoscopist’s clinical impression, one polyp is left in place with follow-up recommended in 10 years. Seven years later, the patient is diagnosed with metastatic colon cancer. A lawsuit contends that the interval cancer was caused by misdiagnosis of the polyp.

Key Questions:

• Who is liable, the endoscopist, the AI vendor, or both?
• Did relying on CADx meet the standard of care?
• Should the polyp have been removed regardless of AI prediction?

Liability Allocation

#

The “Diagnose and Leave” Problem

#

CADx systems that predict polyp histology create unique risks:

Current Standard: Remove all polyps, send for pathology

AI-Enabled Future: Leave predicted hyperplastic polyps, resect only adenomas

Liability Risk: If a “left” polyp becomes cancer, the endoscopist faces:

• Claim of negligent reliance on unproven technology
• Allegation of departing from established standard of care
• Difficulty defending AI prediction error as beyond control

The Deskilling Crisis

#

The Lancet 2025 Study

#

A concerning study in The Lancet Gastroenterology & Hepatology (2025) examined how continuous AI exposure affects endoscopist behavior:

Study Design:

• Retrospective, observational study
• Four endoscopy centers in Poland
• Assessed performance when AI was NOT in use after regular AI exposure

Key Concern: The study raises the question: Do endoscopists who regularly use AI become dependent on it, performing worse when AI is unavailable?

Implications:

• “Automation complacency” may affect GI as it has aviation and radiology
• Skills may atrophy with AI reliance
• System failures or unavailability create heightened risk
• Training must maintain non-AI competency

Protecting Against Deskilling

#

For Individual Endoscopists:

• Maintain regular non-AI colonoscopy practice
• Conduct periodic self-assessment without AI assistance
• Document AI-independent competency

For Practices and Hospitals:

• Establish rotation schedules with/without AI
• Track ADR both with and without AI assistance
• Plan for AI system downtime or failure

Reimbursement: The US Gap

#

Current Status (2025)

#

Why This Matters for Liability

#

The lack of US reimbursement creates a problematic dynamic:

Cost Pressure:

• Practices bear full cost of AI systems
• No additional payment for AI-assisted procedures
• Economic incentive to use AI without full training/integration

Adoption Variability:

• Well-resourced practices adopt AI; others cannot
• Creates potential “two-tier” standard of care
• Rural and safety-net hospitals may lag

Documentation Gap:

• Without reimbursement coding, AI use may be poorly documented
• Difficult to track outcomes and quality
• Liability exposure from incomplete records

Practical Guidance for Endoscopists

#

Using CADe Appropriately

#

DO:

• Treat CADe as an adjunct, not a replacement for careful inspection
• Maintain standard withdrawal times (6+ minutes)
• Respond thoughtfully to AI alerts, investigate, don’t just dismiss
• Document AI alerts and your response
• Remove suspicious polyps regardless of AI classification

DON’T:

• Assume AI has detected all polyps
• Reduce your own vigilance because “AI is watching”
• Use CADx predictions to justify leaving polyps in place
• Skip areas because AI didn’t flag them
• Fail to document when you override AI

Documentation Best Practices

#

Every AI-assisted colonoscopy report should include:

1. AI System Used: Name and version
2. AI Alerts: Number and locations
3. Your Response: Actions taken for each alert
4. Overrides: When you disagreed with AI and why
5. Findings: Standard polyp documentation
6. Limitations: Any technical issues affecting AI performance

When AI Isn’t Available

#

Maintain competency for non-AI colonoscopy:

• System downtime happens
• Not all facilities have AI
• Your skills shouldn’t atrophy
• Document that AI was not used if unavailable

Future Directions

#

CADx Evolution

#

Computer-aided diagnosis is advancing rapidly:

Potential Benefits:

• Resect-and-discard for diminutive polyps (cost savings)
• Optical diagnosis reducing pathology burden
• Real-time risk stratification

Barriers to Adoption:

• Liability exposure remains high
• No professional society endorsement yet
• Performance varies across polyp types
• Requires very high accuracy (>90% sensitivity/specificity)

Capsule Endoscopy AI

#

AI is expanding beyond colonoscopy:

• Automated reading of capsule endoscopy images
• Detection of small bowel pathology
• Reduces reading time from hours to minutes
• Liability framework still developing

Integration with EHR

#

Future developments may include:

• Automated procedure documentation
• Risk-adjusted follow-up recommendations
• Population health management integration
• Quality metric tracking

Frequently Asked Questions

#

Related Resources

#

Healthcare AI

#

• AI Medical Device Adverse Events, FDA reporting and MAUDE database
• Healthcare AI Standard of Care, Overview of medical AI standards
• Radiology AI Standard of Care, Parallel liability framework
• Oncology AI Standard of Care, Cancer detection and treatment AI

AI Liability Framework

#

• AI Product Liability, Strict liability for AI systems
• AI Misdiagnosis Case Tracker, Documented AI diagnostic failures
• AI Litigation Landscape 2025, Overview of AI lawsuits

Professional Guidance

#

• AI Insurance Coverage, Professional liability and AI
• AI Standard of Care FAQ, Common AI liability questions