Beyond Existing Rating Scales: Development of a Novel Nomogram for Predicting Severe Clinical Bleeding Associated with Low-Molecular-Weight Heparin in Hospitalized Medical Patients

Introduction Low-molecular-weight heparin (LMWH) is widely used for thromboprophylaxis and treatment in hospitalized patients; however, LMWH-related severe clinical bleeding (LSCB) remains a major concern. Existing bleeding risk scales have been developed for oral anticoagulants and have limited applicability to LMWH, leaving clinicians without reliable tools for predicting LSCB. Aim This study aimed to evaluate the discriminatory performance of existing scales for predicting LSCB in hospitalized patients and develop a tailored nomogram for individualized risk prediction. Method Hospitalized medical patients prescribed LMWH between July 2021 and August 2024 at three tertiary hospitals in Hangzhou, Chinawere retrospectively analyzed. Each LSCB case was matched with three non-LSCB controls from the same department and period. The prevalence of LSCBs, bleeding sites, and clinical characteristics are described. Receiver operating characteristic (ROC) curves were used to assess the predictive performance of existing scales. Variables with p < 0.05 in univariate analysis were included in logistic regression to identify independent predictors and subsequently incorporated into a nomogram. Discrimination, calibration, and external validation were performed. Results Among 22,096 patients, 369 (1.67%) developed LSCB, most commonly severe gastrointestinal bleeding (74.3%), with a mean onset of 5.68 days. A total of 1,089 patients with non-LSCB were matched as controls. Existing scales performed limited predictive value (AUC 0.52-0.68). Logistic regression identified 12 independent predictors: hypoproteinemia (albumin <30 g/L), anemia (Hb <90 g/L), active gastrointestinal ulcer, thrombocytopenia (platelets <75×10⁹/L), coagulation abnormalities (PT or aPTT >1.2×ULN), cefoperazone/latamoxef exposure >7 days, hypocalcemia ([Ca²⁺] <2.10 mmol/L), aspirin therapy, dual antiplatelet therapy, renal dysfunction (GFR <60 mL/min), hepatic impairment (AST or ALT ≥3 or TBIL ≥2×ULN), and age >65 years. Odds ratios ranged from 6.16 (hypoproteinemia) to 1.47 (age >65 years). A nomogram incorporating these factors achieved AUC 0.890 in the derivation cohort. Calibration was good (Hosmer-Lemeshow p = 0.312), and predictions closely matched the observations. External validation yielded an AUC of 0.876, confirming robustness. Conclusion The existing scales for predicting LSCB lack accuracy in hospitalized patients. Independent predictors were identified and integrated into a validated nomogram with superior performance, offering a practical framework for individualized bleeding risk assessment and safe LMWH management in hospitals.