Short-term prognostic model of spontaneous cerebral hemorrhage based on XGboost
10.3760/cma.j.cn115354-20230314-00148
- VernacularTitle:XGboost模型预测自发性脑出血短期预后的价值分析
- Author:
Hong YUE
1
;
Aimei WU
;
Zhi GENG
;
Zhaoping YU
;
Ye YANG
;
Chi ZHANG
;
Xuechun LIU
;
Juncang WU
Author Information
1. 安徽医科大学附属合肥医院，合肥市第二人民医院神经内科，合肥 230011
- Keywords:
Intracerebral hemorrhage;
Predictive model;
Machine learning
- From:
Chinese Journal of Neuromedicine
2023;22(7):706-710
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To develop a short-term prognostic model of spontaneous cerebral hemorrhage based on eXtreme Gradient Boosting (XGBoost) machine learning, and to compare its predictive performance with a Logistic regression model.Methods:Patients with sICH admitted to Department of Neurology, Second People's Hospital of Hefei from January 2018 to March 2022 were chosen; their general demographic characteristics, medical history, laboratory indices and cranial imaging data were retrospectively collected. The prognoses of patients 90 d after discharge were evaluated according to modified Rankin Scale (mRS) scores (good prognosis: mRS scores<3; poor prognosis: mRS scores≥3). XGboost and multiple Logistic regression models were used to screen out the factors for prognoses of patients 90 d after discharge, and area under receiver operating characteristic (ROC) curves, sensitivity, specificity and prediction accuracy of the 2 models were analyzed and compared.Results:A total of 413 patients with sICH were included; 180 patients(43.6 %) had poor prognosis and 233 (56.4%) had good prognosis 90 d after discharge. Multivariate Logistic regression results showed that age≥65 years, hemorrhage into the ventricle, hematoma volume of 20-40 mL, hematoma volume>40 mL and National Institutes of Health Stroke Scale (NIHSS) scores were independent influencing factors for short-term prognoses of sICH ( P<0.05). The variables in the XGBoost model were ranked in order of importance: NIHSS scores, systolic blood pressure at admission, Glasgow coma scale (GCS) scores, age≥65 years, hemorrhage into the ventricle, hematoma volume of 20-40 mL, and hematoma volume>40 mL. AUC of XGBoost model in predicting the prognosis was 0.895 (95% CI: 0.820-0.947), enjoying sensitivity of 68.89%, specificity of 94.83%, and prediction accuracy of 83.5%. AUC of Logistic regression model in predicting the prognosis was 0.894 (95% CI: 0.818-0.946), enjoying sensitivity of 93.33%, specificity of 70.69%, and prediction accuracy of 80.58%. Conclusion:The short-term prognostic model based on XGboost for sICH patients has high predictive efficacy, whose predictive accuracy is better than traditional Logistic regression.
