Our Take
Upper GI endoscopy has been treated as the junior procedure in combined sessions, and the evidence shows it: 69% of missed upper GI cancers were already visible to the endoscopist but not examined carefully enough.
Why it matters
Health systems are combining procedures to address a projected 1,400-endoscopologist shortage by 2037, but the infrastructure to maintain upper GI quality during combined sessions does not yet exist. AI can fill that gap now, before the procedural volume makes quality recovery harder.
Do this week
Gastroenterology leadership: audit your combined endoscopy protocols this quarter to map where upper GI examination time is being compressed, and establish landmark-completion checkpoints before scope withdrawal.
Combined endoscopy is becoming standard, but quality infrastructure is not
Combined upper and lower GI endoscopy (esophagogastroduodenoscopy followed by colonoscopy in a single session) is now routine across US and European health systems. The clinical appeal is straightforward: one preparation, one sedation, less downtime for patients. For health systems facing a projected specialist shortage of nearly 1,400 gastroenterologists by 2037, consolidation is shifting from convenience to necessity.
The problem: upper GI quality is lagging. Lower GI has decades of established metrics (adenoma detection rate, cecum detection, withdrawal time) and AI-assisted tools validated across 28 randomized controlled trials involving nearly 24,000 patients, showing a 20% increase in adenoma detection and a 55% decrease in adenoma miss rate (per a 2024 meta-analysis). Upper GI has neither the standardized metrics nor the proven detection infrastructure.
The clinical consequence is measurable. Upper GI cancer miss rates remain above 8% for esophageal and gastric cancers, with some estimates reaching 11.3% across three years. More striking: in 69% of missed upper GI cancer cases, the endoscopist had already recorded an abnormality at the exact site where cancer was later diagnosed. The cancer was not invisible. It was inadequately examined.
Cognitive load during combined procedures is eroding upper GI performance
The sequencing of combined procedures amplifies the problem. Colonoscopy typically dominates clinical attention in a combined session because it carries cancer-screening guidelines, the most developed quality metrics, and the highest procedural volume. Upper GI is often performed first, quickly, before the procedure both specialist and patient are primarily there for.
Research into bidirectional endoscopy sequencing confirms that procedural fatigue and endoscopist preference materially affect upper GI findings. Upper GI examinations are performed faster than clinical best practice recommends, and combined-session time pressure accelerates that trend. The reimbursement structure reinforces it: US endoscopy billing is largely binary, paying a base fee for the diagnostic procedure plus an increment if at least one biopsy or resection occurs, regardless of how thorough the examination was. There is no financial signal to encourage careful landmark coverage.
Operator specialization adds another layer. US endoscopy training emphasizes colonoscopy. Many practitioners develop deeper competency in lower GI over their careers, yet combined procedures require the same specialist to perform both, regardless of where their experience is concentrated.
AI landmark verification is the first step, not the final one
The credibility argument for AI-assisted endoscopy has already been made in colonoscopy. That precedent now applies to upper GI, where the clinical community has enough experience with AI to evaluate evidence on its merits. Institutional reluctance that slowed early lower GI adoption no longer carries the same weight.
The development roadmap for upper GI AI must follow the gap. Landmark verification comes first: establishing in real time that all anatomically required areas were examined and creating the quality baseline that detection work depends on. Think less "spotter" and more "sophisticated GPS," tracking whether the endoscopist passed through every required checkpoint and viewed it at appropriate image quality.
From there, focus shifts to early detection of precancerous conditions: dysplasia in Barrett's esophagus and gastric intestinal metaplasia. These are the upper GI counterparts to adenoma detection in colonoscopy, with similarly high clinical stakes and significantly higher room to improve on current unassisted performance.
A secondary benefit: when AI findings are injected automatically into procedure reports rather than documented manually after each case, time between procedures shrinks and throughput increases. In a combined session where documentation requirements are effectively doubled, those gains compound. For health system operators weighing adoption cost, this is the ROI argument: quality improvement and operational efficiency run on the same roadmap.