Comparative performance and external validation of three different models in predicting inadequate bowel preparation among Chinese inpatients undergoing colonoscopyFirst Name Last

Background and Aims Inpatient colonoscopy frequently fails because bowel cleansing is inadequate. Magnesium sulfate is the cathartic of choice in most Chinese hospitals because it costs only a few cents, yet no externally validated tool exists to flag inpatients at high risk of poor preparation. We compared three in-house prediction models to identify the one that best helps nurses and physicians optimize bowel preparation before colonoscopy. Methods Using three previously derived models—Model-1 (seven static variables), Model-2 (nine two-stage variables), and Model-3 (twelve integrated variables)—we conducted a prospective cohort study. Consecutive inpatients aged ≥ 18 years who received a split-dose magnesium sulfate regimen for elective colonoscopy between July 2024 and March 2025 were enrolled. Inadequate bowel preparation was defined as a total Boston Bowel Preparation Scale score < 6 or any segment score < 2. Discrimination (area under the ROC curve, AUC), calibration (calibration plot and Hosmer-Lemeshow test), and clinical utility (decision-curve analysis) were evaluated. Results Among 977 patients, 107 (10.95 %) had inadequate preparation. Model-3 achieved an AUC of 0.785 (95 % CI 0.746–0.824), outperforming Model-1 (AUC 0.679) and Model-2 (AUC 0.681) by 0.106 (P < 0.001). At the optimal cut-off, Model-3 provided 83.8 % sensitivity, 60.6 % specificity, and 79.7 % accuracy. Calibration was good (Hosmer-Lemeshow P = 0.286), and decision-curve analysis showed the widest net-benefit range (threshold probability 0.401–0.982). Conclusions In hospitalized patients receiving magnesium sulfate, Model-3—combining baseline risk factors with real-time nursing assessments—offers superior discrimination, calibration, and clinical utility. Embedding this simple score in the electronic health record flags high-risk patients early, triggers tailored education, and improves preparation quality without added cost. Multi-centre implementation studies are warranted to confirm generalizability.