基于LASSO-Logistic回归构建脑出血术后脑积水的风险预测模型  

Construction of risk prediction model for postoperative hydrocephalus after spontaneous intracerebral hemorrhage based on LASSO-Logistic regression

作　　者：郝广志 孙琳琳 张冰莹 霍达 董玉书 HAO Guangzhi;SUN Linlin;ZHANG Bingying;HUO Da;DONG Yushu(Department of Neurosurgery,General Hospital of the Northern Theater Command of the Chinese People's Liberation Army,Shenyang,Liaoning 110016,China)

机构地区：[1]中国人民解放军北部战区总医院神经外科,辽宁沈阳110016

出　　处：《中华神经外科疾病研究杂志》2025年第1期13-19,共7页Chinese Journal of Neurosurgical Disease Research

基　　金：国家自然科学基金(编号:82071481)。

摘　　要：目的基于LASSO-logistic回归筛选自发性脑出血(SICH)术后脑积水发生的危险因素,同时构建临床预测模型和验证。方法选取2023年8月至2024年3月北部战区总医院收治的724例SICH患者,根据7∶3比例随机分为训练集(n=507)和验证集(n=217)。在训练集中采用Lasso回归结合单因素Logistic回归的方法共同进行危险因素筛选,最后通过多因素Logistic回归确定危险因素并进行模型构建。使用列线图对模型进行可视化处理,通过受试者工作特征(ROC)曲线下面积、校准图分别评估模型的区分度和校准度,通过决策曲线分析(DCA)评估模型临床使用价值。结果SICH后脑积水的发生与年龄、糖尿病、格拉斯哥昏迷量表(GCS)评分、出血范围、脑脊液葡萄糖定量和脑脊液蛋白含量相关(P<0.05),与性别、饮酒、血压、吸烟、凝血指标和是否行腰椎穿刺术无关(P>0.05)。模型的ROC曲线下面积为0.873(95%CI:0.829-0.918)。校准图显示模型具有良好的校准度。DCA分析表明在临床应用方面,预测模型在阈值概率介于0.02和0.85之间具有良好的净收益。结论高龄、糖尿病、GCS评分低、出血破入脑室、脑脊液葡萄糖含量降低和脑脊液蛋白含量增高为影响ICH术后脑积水发生的独立危险因素。以此构建的预测模型具有良好的区分度及校准度,为临床医生提供可靠的诊治依据。Objective To construct and validate a clinical prediction model for postoperative hydrocephalus after spontaneous intracerebral hemorrhage(SICH)based on LASSO-Logistic regression.Methods A total of 724 patients with spontaneous cerebral hemorrhage admitted to the General Hospital of the Northern Theater Command from August 2023 to March 2024 were randomly divided into the training set(n=507)and the verification set(n=217)according to a ratio of 7:3.In the training set,Lasso regression combined with univariate Logistic regression was used to jointly screen the risk factors,and then multivariate Logistic regression was used to determine the risk factors and establish the model.The model was visualized with a nomogram.The area under ROC curve and calibration chart were respectively used to evaluate the differentiation and calibration of the model.The clinical value of the model was evaluated by clinical decision curve(DCA).Results The occurrence of post-ICH hydrocephalus was correlated with age,diabetes,GCS score,bleeding range,CSF glucose quantitation and CSF protein content(P<0.05),but not with sex,alcohol consumption,blood pressure,smoking,coagulation index and lumbar puncture operation(P>0.05).The area under ROC curve of the model was 0.873(95%CI:0.829-0.918).The calibration diagram showed that the model had a good calibration.Decision curve analysis showed that the predictive model had a good net benefit in the threshold probability between 0.02 and 0.85.Conclusions Old age,diabetes,low GCS score,bleeding into the ventricle,decreased glucose content in CSF and increased protein content in CSF are independent risk factors for hydrocephalus after SICH.The prediction model constructed by this method has a good degree of differentiation and calibration,and provides a reliable diagnosis and treatment basis for clinicians.

关 键 词：自发性脑出血 脑积水 预测模型 Lasso回归 诺莫图 

分 类 号：R743.34[医药卫生—神经病学与精神病学]