Letter to the Editor: Adverse event classification in gastrointestinal endoscopy

Abstract

Standardized reporting of endoscopy-related adverse events (AE) remains challenging. Definitions, severity thresholds, and attribution windows vary across units, and documentation practices substantially influence observed event rates. A recent registry-based comparison of two AE classification systems illustrates that strong concordance in severity grading does not, in itself, resolve implementation challenges at the boundary between incidents and non-events, nor does it address under-ascertainment. In this letter, we trace the evolution of endoscopy AE nomenclature, from earlier documentation-intensive approaches to contemporary frameworks that prioritize clinical consequence and treatment-based grading. Drawing on external real-world validation data and published commentaries, we argue that endoscopy units should prioritize a dual-stream safety framework: Severity-graded adverse-event reporting for clinically consequential harms, alongside structured capture of incidents and near-misses for prevention-oriented learning. We further highlight the need for procedure-stratified interpretation, explicit documentation of local post-procedural pathways, robust ascertainment workflows, and separate monitoring of diagnostic delay and procedural appropriateness.

Key Words: Adverse event; Colonoscopy; Endoscopy; Gastroscopy; Healthcare quality; Patient safety

Core Tip: Adverse event (AE) grading is useful only when paired with robust documentation and a clear review pathway. A practical model is to track AEs for benchmarking while retaining an incident or near-miss stream for prevention signals, and to measure diagnostic delay and procedural appropriateness with separate indicators rather than forcing them into AE registries.

TO THE EDITOR

We read with great interest the study by Corsi et al[1] published on the World Journal of Gastrointestinal Endoscopy comparing the adverse events (AEs) in gastrointestinal endoscopy (AGREE) and the American Society for Gastrointestinal Endoscopy (ASGE) classification systems for grading gastrointestinal endoscopy-related AEs using a long-standing registry from a Latin American academic center. The authors analyzed 176655 procedures performed between 2009 and 2022, and identified 235 recorded AEs (0.13%), most commonly cardiorespiratory events, bleeding, and perforation. They demonstrated strong cross-system correlation (Spearman ρ = 0.89; P < 0.001) with high interobserver agreement (κ = 0.83 for ASGE and 0.74 for AGREE), supporting AGREE as a feasible alternative for grading AEs in heterogeneous clinical settings. The study is valuable because it provides real-world data from a large registry and it tests two complementary but conceptually distinct statistical properties that are often conflated: Cohen’s kappa and Spearman’s rank correlation. While Cohen’s kappa addresses inter-rater reliability (whether different observers apply the scale consistently), Spearman’s rank correlation addresses inter-scale concordance (whether two scales behave similarly when grading severity). This distinction is clinically relevant. In addition, kappa values may be influenced by the low prevalence of AEs and by category distribution, so apparently modest or strong agreement should be interpreted in context rather than as a standalone measure of clinical interchangeability. For endoscopy units, statistical concordance between scales does not by itself determine whether a taxonomy will be easier to apply prospectively, more reproducible across users, or better aligned with local review thresholds. Two classification systems may demonstrate strong correlation in severity ranking yet differ in ease of consistent application across users. Conversely, a scale may show strong inter-rater reliability but behave differently when mapped onto another system’s severity ranking. Notably, Crispino et al[2] were the first to externally validate AGREE system and reported 226 AEs among 84863 procedures (0.3%). Compared to Corsi et al[1], they reported a more modest correlation between ASGE and AGREE grades (Spearman ρ = 0.61), but higher interobserver agreement for AGREE (weighted κ = 0.80) compared with ASGE (κ = 0.60). This variation supports the conceptual promise of AGREE’s intervention-anchored grading to reduce subjectivity, while also showing that correlation and agreement depend on context, procedure types, and methods of event capture.

The manuscript is timely, as it is increasingly important for endoscopy units to have AE metrics that are standardized, feasible, and oriented towards quality improvement. Despite longstanding efforts to systematize AE reporting, contemporary practice remains vulnerable to knowledge gaps, inconsistent thresholds, and challenges in delayed attribution. These limitations undermine cross-center comparability and hinder meaningful benchmarking. Several prior attempts have been made to define and clarify the AEs following endoscopy. Fleischer et al[3] proposed the ODD system (outcome, disability, death) as a scoring approach for endoscopic procedures based on detailed documentation of these three adverse domains. However, it was considered overly cumbersome for routine clinical use and therefore was not adopted widely. Cotton et al[4] (ASGE lexicon) defined an AE as an event that prevents completion of the planned procedure and/or results in hospital admission, prolongation of an existing hospitalization, performance of another procedure requiring sedation/anesthesia, or subsequent medical consultation. This framework emphasized temporal categorization (pre-procedural, intra-procedural, post-procedural up to 14 days or later) and deliberately distinguished incidents from AEs. Incidents are unplanned events that do not alter procedural completion or care plan but should nonetheless be recorded for quality improvement. Nass et al[5] highlighted practical limitations of earlier endoscopy AE classifications, including subjectivity inherent in severity labeling and the potential of misclassification when length of stay is used as a proxy for severity, as hospitalization duration may reflect observation practice, logistical constraints, or social determinants rather than true clinical harm. They therefore introduced the AGREE classification, which grades AEs primarily according to the treatment or intervention required and defines an AE as a negative outcome that either prevents procedural completion or results in deviation from the standard post-procedural course. The intent was to provide a clearer, more reproducible framework aligned with modern therapeutic endoscopy and suitable for cross-institutional comparison. While Awadie and Gralnek[6] supported AGREE as a meaningful step toward standardized, reproducible AE reporting, the authors highlighted persistent underreporting as a major limitation, driven by time constraints, duplication of documentation, institutional culture, fear of scrutiny, and accountability concerns. They further cautioned that admission status or length of stay may not reliably reflect clinical severity, as they may be influenced by social factors or local policies. They advocated refinement stratified by endoscopic procedure type and AE category, alongside broader external validation. Similarly, Facciorusso et al[7] argued that AGREE’s reliance on deviation from the standard post-procedural course introduces subjectivity as standards vary across health systems, potentially limiting large-scale applicability. They also cautioned that inclusion of events occurring within 30 days irrespective of a clear causal relationship may overestimate AE rates and noted that lack of blinding in acceptability assessments could inflate perceived advantages of newer tools. Our purpose is not only to comment on the registry findings, but also to propose a practical implementation framework for endoscopy units; in our view, AGREE is best introduced through prospective dual-coding alongside ASGE rather than abrupt replacement, with particular relevance to therapeutic endoscopy where intervention-based grading may offer added practical value. Against this backdrop, we identify five practical issues which merit further evaluation. The proposed endoscopy safety framework is summarized in Figure 1, which integrates definitions and scope, documentation processes, and learning-and-response mechanisms.

Figure 1 Safer endoscopy through measurable learning. AE: Adverse event; EMR: Electronic medical record; ED: Emergency department.

Reclassification to expand the ‘non-AE’ category

Reclassifying a greater proportion of cases as “non-AE” may have both beneficial and unintended consequences. On one hand, narrowing the AE definition can reduce signal-to-noise distortion within registries, allowing limited registry and peer-review resources to focus on events with clear clinical consequences. This prioritization is particularly valuable in resource-constrained settings. On the other hand, excessive reclassification risks excluding potentially modifiable events from formal review pathways. Physiological perturbations such as transient hypoxemia or hypertension, and patient-experience harms such as inadequate sedation may not meet strict AE criteria but pose important opportunities for systems-level learning and prevention. Accordingly, the practical solution does not lie in debating the correct label, but rather in designing registries that preserve complementary reporting streams. Specifically, robust quality frameworks should maintain: (1) Graded AEs for severity benchmarking; and (2) Structured documentation of incidents/near-misses for prevention-orientated learning. Such a dual-stream approach ensures that events classified as “non-AEs” remain visible for audits and quality improvement initiatives.

Procedure-dependent concordance

The observed differences in inter-scale performance between diagnostic and therapeutic procedures warrant careful consideration. Therapeutic endoscopy is associated with a higher baseline risk profile and more frequent downstream interventions, including radiologic procedures, endoscopic rescue, surgical intervention and hospital admission. Consequently, even modest divergence between grading systems may become magnified in quality dashboards and case review triggers. It is therefore important to avoid interpreting overall instrument concordance as evidence of uniform performance across procedure categories. We also encourage procedure-stratified reporting by modality [e.g., upper gastrointestinal endoscopy, colonoscopy, endoscopic retrograde cholangiopancreatography (ERCP)]. This ensures that the AE classification systems are transparently evaluated within clinically relevant subgroups. Such stratification would allow more precise alignment and hence enhance interpretability.

Local variability in standard post-procedural course

Defining AEs as deviations from the standard post-procedural course offers an intuitively appealing framework. However, what constitutes a standard course is inherently shaped by local practice environments, including institutional policies, manpower resources, availability of after-hours imaging, bed capacity constraints, and, in some settings, financial or insurance considerations. These contextual factors can influence thresholds for observation vs hospital admission, as acknowledged by Corsi et al[1]. Such variability introduces a potential benchmarking hazard: Identical patient physiology may lead to different classification outcomes across institutions, not because of true differences in harm but due to divergent local care pathways. To preserve interpretability, we recommend institutions which adopt deviation-based frameworks to explicitly document their local post-procedure pathways and decision thresholds. The documentation should include observation duration policies, triggers for imaging, admission criteria, and mechanisms for capturing return-to-emergency department visits. As a pragmatic minimum dataset, units may also record procedure type, indication, planned vs completed procedure, immediate and delayed events, intervention required, unplanned emergency department attendance or admission, need for radiology/endoscopic/surgical rescue, and attribution confidence. Transparent contextualization will enhance the validity of cross-unit comparisons and allow more meaningful benchmarking.

Documentation superseding the grading system

A central implementation lesson is that no AE classification system can compensate for incomplete documentation. Real-world AE surveillance consistently demonstrates that reliance on voluntary endoscopist reporting underestimates AE rates. Additional AEs, particularly milder events, are frequently uncovered only through structured electronic health record review. This under-ascertainment reflects a broader operational challenge: While systematic AE monitoring is widely endorsed, it is hard to sustain in reality due to multiple factors such as time constraints, delayed presentations, and medico-legal concerns. These pressures may inadvertently discourage thorough documentation and consequently, the fidelity of AE surveillance depends less on the sophistication of the grading framework and more on the robustness of documentation workflows. Accordingly, endoscopy unit quality indicators should emphasize practical documentation workflows that sit above any grading system. These include a standardized process to record and track AEs; structured mechanisms to capture delayed AEs within post-procedure contact windows; surveillance of unplanned emergency department visits or admissions; and a formal multidisciplinary review forum that translates documented events into actionable prevention strategies, operator feedback, and protocol refinement.

AEs, diagnostic delay, and procedure appropriateness

Clear delineation of what AE taxonomies are and are not intended to capture is essential to avoid category errors in interpretation and unit benchmarking. The ASGE lexicon explicitly recognizes that AEs may occur prior to scope insertion, including morbidity related to bowel preparation or prophylactic measures, and emphasized that delayed AEs will be under-detected unless units implement structured follow-up rather than relying on spontaneous reattendance. These design elements shape what an endoscopy unit identifies, records, reviews, and ultimately improves. In contrast, outcomes such as missed cancer and delayed cancer diagnosis align more closely with diagnostic-performance and appropriateness domains than with discrete procedure-associated events, unless explicitly operationalized within an AE framework using predefined attribution rules and time windows. Without such structure, these outcomes risk inconsistent classification, retrospective reinterpretation, or omission despite substantial patient impact. Similarly, overuse of endoscopy is not itself an AE under existing frameworks. It represents an indication-level failure that increases exposure to procedural risk and may inflate harm rates without being visible in AE dashboards. Telford summarized earlier evidence suggesting that inappropriate colonoscopy indications were frequent in some settings, particularly in open-access units, and were associated with lower diagnostic yield[8]. More recent evidence also supports the importance of appropriateness, with a 2021 systematic review and meta-analysis showing that appropriate colonoscopy indications increase diagnostic yield and reduce unnecessary examinations[9]. Although complications arising from unnecessary endoscopy remain bona fide AEs and should be graded accordingly, overuse requires separate appropriateness metrics and denominators, rather than reliance on AE registries as proxies for overall quality. In practice, these domains may sit alongside AE dashboards as parallel quality indicators, such as interval cancer review, unplanned post-procedure emergency attendance, unplanned admission, and periodic appropriateness audits. Cotton[10] makes a parallel argument for ERCP, noting that use in patients unlikely to benefit, such as those with obscure abdominal pain without objective evidence of biliary or pancreatic disease, shifts the benefit-risk balance unfavorably, given low diagnostic yield and substantial risk of pancreatitis. He advocates more selective approach with improved noninvasive imaging strategies and transparent risk disclosure. In this context, our unit expanded the use of magnetic resonance cholangiopancreatography and laparoscopic common bile duct exploration in patients with biliary lithiasis to minimise need for ERCP, acknowledging that endoscopy and surgery should be viewed as complementary rather than competing modalities[11,12].

Conclusion

Comparative studies of AE taxonomies are valuable, but their practical impact is ultimately shaped by how endoscopy units define, document, and learn from safety signals. We advocate a dual-stream safety framework that preserves prevention-oriented learning from incidents and near-misses while applying severity grading to clinically consequential harms. On current evidence, endoscopy units considering transition to AGREE should adopt a staged approach, using prospective dual-coding with ASGE, particularly in therapeutic endoscopy, together with explicit documentation of local post-procedure pathways and decision thresholds. Diagnostic delay and procedural appropriateness should be monitored alongside, rather than within, AE registries to avoid overextending the taxonomy beyond its intended scope.