Establishment of a bleeding risk model for low-molecular-weight heparin in cancer-associated venous thromboembolism: a single-center retrospective study

TL;DR

This study created a model to predict bleeding risks in cancer patients using low-molecular-weight heparin, helping doctors make better treatment decisions.

Contribution

The study introduces a new bleeding risk model by integrating LMWH dosage, platelet count, and HAS-BLED score for cancer patients.

Findings

The model achieved high predictive accuracy with an ROC-AUC of 0.90.

The model demonstrated strong sensitivity and specificity for bleeding events.

The model showed good discrimination and strong diagnostic performance.

Abstract

To establish a bleeding prediction model for the use of low molecular weight heparin anticoagulation in cancer patients, aiming to help medical staff to individually evaluate the timing of low molecular weight heparin use in cancer patients.

This retrospective cohort study enrolled 731 cancer patients (aged ≥18 years) from January to December 2021 receiving LMWH for venous thromboe-mbolism (VTE) prophylaxis at a tertiary general hospital in Southwest China. Participants were stratified into bleeding (n = 19) and non-bleeding (n = 712) cohorts based on ISTH-defined clinical outcomes. Risk factors identified through multivariablebinary logistic regression, with subsequent development and internal validation performed using R software (version 4.3.2).

Univariate analysis of bleeding risk factors revealed statistically significant differences (P < 0.05) in body weight, nonsteroidal anti-inflammatory drug (NSAIDs) use, LMWH dosage, prothrombin time (PT), creatinine clearance, platelet count, Padua score, and HAS-BLED bleeding risk score. Based on clinical relevance, the final model incorporated LMWH dosage, platelet count, and HAS-BLED bleeding risk score as assessment items. The model demonstrated excellent predictive ability for bleeding events, with an ROC-AUC (95% CI) of 0.90 (0.82–0.97). The model showed good discrimination (Hosmer-Lemeshow, P = 0.854 > 0.05) and decision making capability, with strong diagnostic performance (accuracy: 0.83, sensitivity: 0.83, specificity: 0.79, positive predictive value: 0.99). The model had a low probability of missed diagnoses and high sensitivity and specificity.

This study developed an optimized bleeding risk prediction model by enhancing the HAS-BLED score through the integration of three key variables: HAS-BLED score, LMWH dosage, and platelet count, demonstrating a robust application prospect for anticoagulation management in cancer patients.

Figures