Objectives: The use of anticoagulants in patients increases the risk of intracranial hemorrhage (ICH). Our aim was to identify factors associated with cerebral hemorrhage in patients using anticoagulants and to develop a predictive model that would provide an effective tool for the clinical assessment of cerebral hemorrhage.
Methods: In our study, indications for patients receiving anticoagulation included AF, VTE, stroke/TIA, arteriosclerosis, peripheral vascular diseases (PVD), prosthetic mechanical valve replacement, etc. Data were obtained from the patient record hospitalization system. Logistic regression, area under the curve (AUC), and bar graphs were used to build predictive models in the development cohort. The models were internally validated, analytically characterized, and calibrated using AUC, calibration curves, and the Hosmer-Lemeshow test.
Results: This single-center retrospective study included 617 patients treated with anticoagulants. Multifactorial analysis showed that male, leukoaraiosis, high risk of falls, APTT ≥ 45.4 s, and FIB ≥ 4.2 g/L were independent risk factors for cerebral hemorrhage, and β-blockers were protective factors. The model was constructed using these six factors with an AUC value of 0.883. In the validation cohort, the model had good discriminatory power (AUC = 0.801) and calibration power. Five-fold cross-validation showed Kappa of 0.483.
Conclusion: Predictive models based on a patient's medical record hospitalization system can be used to identify patients at risk for cerebral hemorrhage. Identifying people at risk can provide proactive interventions for patients.