Prediction Models for Delayed Bleeding After Colorectal Endoscopic Submucosal Dissection: A Systematic Review and Critical Appraisal.

Background: Although many predictive models have been developed or validated for delayed bleeding after colorectal endoscopic submucosal dissection (ESD), their methodological quality and clinical utility remain uncertain. We systematically reviewed and critically appraised all available predictive models for this outcome. Methods: We searched five databases through March 2026 for observational studies that developed or validated prediction models or scoring systems for delayed bleeding after colorectal ESD in adults (≥18 years). Two reviewers independently extracted data and assessed risk of bias using PROBAST. Results: Of 25,177 records screened, five studies reporting five predictive models were included. The incidence of delayed bleeding ranged from 2.9% to 9.9%. The most commonly used predictors were lesion location and lesion size. Discrimination in derivation cohorts ranged from 0.726 to 0.890, internal validation areas under the curve [AUCs] ranged from 0.739 to 0.870, and the externally validated model achieved an AUC of 0.634. All studies were judged to have a high risk of bias, mainly because of inadequate reporting in the predictor, outcome, and analysis domains, although most had low concerns regarding applicability. Conclusions: This systematic review identified five predictive models for delayed bleeding after colorectal ESD. Lesion location and size were the most frequently reported predictors. Discriminative performance varied across derivation and validation cohorts, with suboptimal external-validation results. All models had a high risk of bias per PROBAST (low concerns regarding applicability); therefore, their findings should be interpreted cautiously.