To investigate the risk factors related to the recurrence in patients with hemorrhagic stroke and to establish a corresponding risk prediction model to provide reference for reducing the recurrence rate of hemorrhagic stroke.

The patients with hemorrhagic stroke treated in the Department of Encephalopathy, the First Affiliated Hospital of Henan University of Chinese Medicine from January 1, 2020 to December 31, 2022 were included in the study. The clinical data of patients with recurrent hemorrhagic stroke were collected to explore the risk factors related to recurrence of hemorrhagic stroke. The original data were divided into modeling sequence and verification sequence by random grouping method, and the modeling sequence was analyzed by univariate and multivariate analyses and model construction. The C index, receiver operator characteristic (ROC) curve, standard curve, and clinical decision curve were used to evaluate the discrimination, accuracy, and clinical utility of the model, and the prediction accuracy of the model was verified by verification sequence.

A total of 1 571 patients with hemorrhagic stroke were included, including 253 patients with recurrent hemorrhagic stroke and 1 318 patients with initial hemorrhagic stroke. The original data set was constructed and randomly divided into modeling sequence and verification sequence according to 7:3 ratio. Logistic regression analysis showed that course of disease (OR=3.548, 95%CI: 2.852-4.415), smoking (OR=1.499, 95%CI: 1.136-1.978), drinking (OR=3.330, 95%CI: 2.356-4.709), male (OR=1.424, 95%CI:1.006-2.016), hyperlipidemia (OR=4.372, 95%CI: 2.227-8.584), cerebral infarction (OR=2.254, 95%CI: 1.294-3.928), high density lipoprotein cholesterol (OR=2.034, 95%CI: 1.220-3.389), prothrombin time (OR=1.103, 95%CI: 1.004-1.211), and homocysteine (OR=1.03, 95%CI: 1.014-1.047) were the influencing factors of recurrence in patients with hemorrhagic stroke, and a risk prediction model was established. The area under the ROC curve of the modeling sequence and the verification sequence was 0.834 (95%CI: 0.810-0.859) and 0.842 (95%CI: 0.804-0.880), respectively, and the standard curves were highly consistent. The results of Hosmer Lemeshow goodness-of-fit test were P=0.900 and P=0.736, respectively, and the thresholds of clinical decision curves were < 0.9 and <0.95, respectively, indicating that the model had high discrimination, calibration, and clinical effectiveness.

Course of disease, smoking, drinking, sex, hyperlipidemia, cerebral infarction, high density lipoprotein cholesterol, prothrombin time and homocysteine are independent risk factors of hemorrhagic stroke. The risk prediction model established in this study can timely identify high-risk patients with recurrent hemorrhagic stroke and prevent the occurrence of adverse events.