Existing risk prediction models designed to forecast delayed bleeding following colorectal endoscopic submucosal dissection show variable discrimination, limited independent external validation, and a widespread high risk of bias that restricts confidence in their routine clinical use, according to a systematic review published in BMC Gastroenterology. The investigation, conducted to systematically evaluate these prediction tools and establish clear priorities for their external validation and refinement, analyzed ten published studies encompassing ten newly developed predictive models alongside one updated model.
To identify eligible literature, investigators executed comprehensive literature searches across nine major medical and scientific databases: PubMed, Embase, Web of Science, CINAHL, the Cochrane Library, CNKI, Wanfang, VIP, and SinoMed. The search parameters covered the entire indexed timeline of each respective database from inception up through July 1, 2026. From this retrieval process, the review identified ten primary studies that detailed ten newly created risk prediction models and documented one updated model, establishing the current evidence base for predicting delayed post-procedure bleeding in patients undergoing colorectal endoscopic submucosal dissection.
The systematic review applied structured evaluation frameworks to assess the methodology, reporting transparency, and methodological rigor of all identified prediction models. Study characteristics and model data were extracted using the formal CHARMS checklist. The authors subsequently evaluated model development quality, risk of bias, and clinical applicability using the PROBAST+AI assessment tool. Furthermore, the completeness of reporting across the included studies was scrutinized against the standardized TRIPOD+AI guidelines, with the extracted findings synthesized descriptively based on individual model characteristics, the specific prediction time point utilized, and the validation methods applied.
An analysis of the timing and composition of predictor variables revealed meaningful differences in how clinical data are leveraged to forecast delayed bleeding after colorectal endoscopic submucosal dissection. Among the eleven evaluated models, four relied entirely upon preprocedural predictors, attempting risk stratification prior to the initiation of the procedure. In contrast, seven models required intraprocedural information, indicating that procedural variables and operative observations played a necessary role in determining post-procedure bleeding risk estimates.
Rigorous external testing of the prediction tools remained sparse among the evaluated literature. Only three of the models underwent geographical external validation across independent settings. In these validation cohorts, discrimination varied substantially, with reported areas under the receiver operating characteristic curve spanning from 0.550 to 0.780. This wide performance range highlights that discrimination fluctuates significantly across different validation environments, demonstrating that existing predictive tools do not reliably achieve high diagnostic accuracy across diverse external patient cohorts.
The systematic evaluation highlighted pervasive methodological limitations in both the construction and validation of the models. Concerns regarding development quality were classified as high in nine out of the eleven development or updating assessments performed. In addition, 24 of the 31 evaluated model performance evaluations were judged to carry a high risk of bias under PROBAST+AI criteria. While full model specifications were adequately documented for eight of the eleven models, reporting deficiencies remained widespread. Crucial elements were incompletely documented across multiple studies, specifically surrounding the exact model-building procedures, methods for handling missing data, and the availability of analytical code.
Given the combination of high risk of bias, incomplete calibration assessment, and a marked shortage of independent external validation, the authors concluded that current models cannot yet be recommended with clinical confidence. The investigators noted that researchers should update existing models where necessary and conduct formal assessments of their practical clinical usefulness to establish dependable tools for delayed bleeding following colorectal endoscopic submucosal dissection.
Subject of Research: Systematic review of risk prediction models for delayed bleeding after colorectal endoscopic submucosal dissection
Article Title: Risk prediction models for delayed bleeding after colorectal endoscopic submucosal dissection: a systematic review
Article References: Liu, L., Li, J., Wu, Z., Zhou, Y., Wang, P., & Chen, K. (2026). Risk prediction models for delayed bleeding after colorectal endoscopic submucosal dissection: a systematic review. BMC Gastroenterology. https://doi.org/10.1186/s12876-026-05494-x
Image Credits: AI Generated
DOI: 10.1186/s12876-026-05494-x
Keywords: colorectal endoscopic submucosal dissection, delayed bleeding, risk prediction models, systematic review, PROBAST+AI, TRIPOD+AI, CHARMS checklist, external validation, risk of bias, preprocedural predictors, intraprocedural predictors, area under the receiver operating characteristic curve, model discrimination, calibration assessment, multicenter cohorts, BMC Gastroenterology
News Source: Ophelia Keating. (November 11, 2026). Colorectal Bleeding Prediction Models Show High Bias in Systematic Review. Scienmag.



