8th Global Gastroenterology and Artificial Intelligence Summit Separating AI Hype from Clinical Reality | September 2026-1

1 - Garg Quality Metrics for GI

Pdf Summary

This presentation describes an internal Mayo Clinic workflow, “Colon-Pilot,” that uses generative AI to automate measurement of gastrointestinal quality metrics and support colonoscopy surveillance recommendations. The central problem is that quality measurement has traditionally relied on sampled chart review, which is slow and incomplete, while GI quality targets require accurate, repeatable, case-level tracking.<br /><br />Colon-Pilot separates tasks into two lanes: automated quality measurement and clinician-reviewed follow-up recommendations. The AI model extracts structured data from procedure reports, pathology, and prior GI history, while deterministic guideline logic applies USMSTF/ACG-ASGE rules. Complex or ambiguous cases are deferred to clinicians rather than auto-resolved. This design emphasizes that the AI reads, rules decide, and clinicians remain responsible.<br /><br />In validation, 596 consecutive colonoscopies were reviewed with no manual exclusions. For non-deferred interval recommendations, the system matched expert consensus in 97.5% of cases (509/522). Quality indicator extraction accuracy was very high, generally 99.1–100%, including adenoma detection, bowel preparation, cecal intubation, and advanced histology measures. The 13 mismatches were analyzed to improve the system, highlighting guideline gaps, extraction threshold errors, and ambiguous pathology as key failure modes.<br /><br />The system now runs as a fully unattended 6 AM refresh, processing large volumes of colonoscopies and producing role-based dashboards for clinicians, division leaders, and administrators. It also supports review of discordant orders before scheduling, but it does not change orders autonomously.<br /><br />The key message is that AI can enable a census-style quality system at scale, but clinical decision-making must remain human, with continuous validation, transparent error reporting, and governance.

Keywords

Colon-Pilot

generative AI

gastrointestinal quality metrics

colonoscopy surveillance

workflow automation

USMSTF guidelines

ACG-ASGE rules

quality indicator extraction

clinician-reviewed recommendations

Mayo Clinic