As artificial intelligence (AI) use gains ground in gastrointestinal (GI) endoscopy, it raises a host of data privacy, provider liability, and equity and bias concerns. The World Endoscopy Organization issued a consensus statement that aims to help clinicians, technology developers, regulators, and medical societies address these issues.

“Gastrointestinal endoscopy has some of the most translationally mature AI applications, with new use cases emerging rapidly,” corresponding author Omer F. Ahmad, MBBS, PhD, a consultant gastroenterologist and interventional endoscopist at University College Hospital, London, England, told Medscape Medical News.

Computer-aided detection (CADe) of colon polyps, the most validated AI application, already is making its way into clinical practice, and emerging uses — CA diagnosis (CADx), automated report generation, and AI-derived quality metrics — are sparking interest, the authors noted.

“As these tools become increasingly capable and complex, there is an urgent need for expert-informed, consensus-based guidance to support their safe and responsible integration into clinical practice,” Ahmad said.

The 10-point consensus statement, published in Annals of Internal Medicine, was developed by a panel of 14 experts from 11 countries.

Privacy and Transparency

AI tools generate and process vast amounts of patient data, raising questions about data privacy, ownership, and use, as well as algorithm development, the authors noted. To safeguard patient information, AI algorithms for GI endoscopy should adhere to local governance policies related to patient privacy and data-sharing agreements, the statement recommended. In the endoscopy suite, commercial AI algorithms and research initiatives should follow data privacy and security regulations, and providers should establish clear policies on data ownership and usage, so patients know how their data will be used.

Additionally, developers of AI tools for GI endoscopy should implement mechanisms to document and transparently report algorithm modifications, updates, and performance outcomes to regulatory and clinical stakeholders.

“Together, these practices strengthen regulatory oversight, enable postmarket surveillance, and help clinicians understand model changes and limitations in everyday use,” the authors wrote.

Medicolegal Considerations

The use of AI in GI endoscopy requires clinicians to balance AI-generated insights with their own clinical judgement, often in real time, potentially blurring conventional boundaries of medical liability, the authors noted. For example, where does liability lie if a clinician relies on an inaccurate CADx or disregards an accurate one? What risks may arise when an endoscopist dismisses an AI-generated report saying mucosal inspection was inadequate because he or she disagrees?

To help guard against liability, physicians and healthcare organizations should ensure that they’re using AI systems in agreement with the manufacturer’s intended use, according to the statement. Clear guidance from medical societies and legal experts could mitigate liability concerns, especially with the emergence of semi-automated reading of capsule endoscopy and CADx, the authors wrote.

Before AI-driven report generation and AI-enabled performance indicators are widely adopted, it is essential to evaluate their accuracy, understand their clinical relevance, and clarify associated medicolegal implications, they added.

Further research is needed to evaluate the real-world clinical impact of AI in GI endoscopy, ideally with evidence linking AI to meaningful patient outcomes, beyond technical performance, especially regarding automated reporting and AI-generated quality metrics, Ahmad said.

Equity and Bias

Healthcare research has shown that AI tools can yield biased outcomes if training datasets aren’t representative of the patient population. To improve equity and minimize bias, AI algorithms developed for use with GI endoscopy should be trained and validated on datasets that reflect the race, ethnicity, and gender mix of the populations they serve, according to the statement. Similarly, AI research and development should include transparent reporting of the study population, so clinicians can assess the generalizability and equity of the technology, the authors wrote.

Research is needed to determine whether the use of AI in GI endoscopy may exacerbate disparities through underrepresentation in training sets or unequal access to AI technology, they added.

“Although perhaps not every AI model for GI endoscopy will require stratification by race or gender, a default assumption of irrelevance risks overlooking subtle but important sources of inequity. Diversity in data sets should therefore be seen not as a burden but as a safeguard for generalizability,” the authors wrote.

Prospective implementation studies will be essential to understand the many ways AI adoption will change GI endoscopy, including workflow, accountability, and standards of care, Ahmad said. “Research demonstrating AI performance across diverse populations and care setting will be important to avoid potentially widening existing disparities,” he added.

Welcome Addition

The new consensus statement is a welcome addition to recent guidance and guideline statements, including the World Endoscopy Organization position statement on both CADe and CADx,the MAGIC Evidence Ecosystem Foundation and BMJ Living Clinical Practice Guideline, and the American Gastroenterological Association Living Clinical Practice Guideline on CADe-Assisted Colonoscopy, Jeremy Glissen Brown, MD, a gastroenterologist and assistant professor of medicine at Duke University Medical Center, Durham, North Carolina, told Medscape Medical News.

“As the authors rightly point out, this is the perfect time to create and critically appraise structured frameworks for thinking through responsible integration, safe evaluation, and governance of AI in endoscopy,” said Glissen Brown, who was not involved in developing the statement. “CADe for colonoscopy has been in clinical use in Europe since approximately 2019, in Asia since 2020, and in the United States since 2021. In addition, ambient scribe technologies, automatic report generation, CADq systems (computer-aided quality assessment) and many CADx applications are relatively deep into the development-to-deployment pipeline.”

Although additional research is needed to address ongoing questions related to the key areas of medical liability, data governance, privacy, transparency, data ownership, and risk of bias, the consensus statements represent an important starting framework, Glissen Brown said.

The consensus statement was developed as part of the part of the OperA (Optimising Colorectal Cancer Prevention through Personalized Treatment with Artificial Intelligence) project funded by the European Commission. The authors had no relevant financial conflicts to disclose. Glissen Brown disclosed serving as a consultant for Medtronic and Olympus/OdinVision and speaking honorarium for Magentiq.