Prediction models for postoperative anastomotic stricture in rectal cancer: a systematic review

Systematically evaluate the performance and methodological quality of predictive models for the risk of anastomotic stricture after rectal cancer surgery to provide evidence-based guidance for clinical practice. We comprehensively searched nine Chinese and English databases (CNKI, Wanfang, VIP, Sinomed, PubMed, Web of Science, Cochrane Library, CINAHL, and Embase) for relevant studies published until June 7, 2025. Two independent researchers screened the literature, extracted data, and assessed methodological quality using the Prediction Model Risk of Bias Assessment Tool (PROBAST). Nine eligible studies were included, demonstrating AUC values ranging from 0.739 to 0.911 in derivation cohorts. Internal validation results were reported in three studies (AUC: 0.905–0.930), while two studies provided test-set validation (AUC:0.818–0.888). The most significant predictors included anastomotic leakage, the use of a protective stoma, and the anastomotic distance. All studies exhibited high risk of bias according to PROBAST criteria, primarily due to retrospective designs, inappropriate handling of continuous variables (dichotomization), inadequate sample sizes, improper management of missing data, univariate-based predictor selection, and insufficient model validation. However, two studies demonstrated good clinical applicability. Research on predictive models for anastomotic stricture after rectal cancer surgery is still in its early stages, constrained by methodological flaws and a lack of external validation that limit their clinical applicability. Future studies should prioritize external validation across diverse populations to confirm their generalizability.