脂蛋白磷脂酶A2联合临床相关指标预测缺血性脑卒中模型的建立与验证  

Construction and verification of a prediction model of ischemic stroke with Lp-PLA2 combined with clinical related indicators

作　　者：吴天晨[1] 梁艳[1] 杨卉[2] 王一如 Wu Tianchen;Liang Yan;Yang Hui;Wang Yiru(Department of Encephalopathy,Nanjing Hospital of Traditional Chinese Medicine,Nanjing 210001,Jiangsu Province,China)

机构地区：[1]南京市中医院脑病科,210001 [2]南京中医药大学

出　　处：《中华老年心脑血管病杂志》2022年第10期1013-1018,共6页Chinese Journal of Geriatric Heart,Brain and Vessel Diseases

基　　金：国家自然科学基金(81904112);江苏省自然科学基金(BK20190136);南京市中医药青年人才培养计划项目(ZYQ20047)。

摘　　要：目的 构建以脂蛋白磷脂酶A2(Lp-PLA2)为基础,结合脑卒中危险因素及临床常见指标预测急性及亚急性脑卒中的临床模型。方法 选取2021年1~12月南京市中医院门诊及住院的急性期及恢复期脑卒中患者194例(脑卒中组),以年龄和性别为模糊匹配对象选取健康体检者169例(对照组),将363例研究对象以7?3分类,其中训练集组251例,验证集组112例。利用LASSO回归筛选危险因素进入预测模型,构建logistic回归模型。绘制ROC曲线,计算曲线下面积(AUC),Hosmer-Lemeshow检验进行拟合优度检验,临床决策曲线评判预测模型的临床指导性,最后以列线图,可视化呈现该临床预测模型。结果 脑卒中组与对照组高血压、糖尿病、冠心病、年龄、纤维蛋白原、Lp-PLA2、脂蛋白(a)、载脂蛋白A-Ⅰ、HDL、游离三碘甲状腺原氨酸水平比较,差异有统计学意义(P<0.05)。训练集组与验证集组脂蛋白(a)、载脂蛋白A-Ⅰ、HDL、LDL水平比较,差异有统计学意义(P<0.05)。LASSO回归分析显示,糖尿病、冠心病、年龄、纤维蛋白原、Lp-PLA2、脂蛋白(a)、载脂蛋白A-Ⅰ、HDL 8个系数不为零的变量,构建logistic回归模型,最终纳入4个危险因素,载脂蛋白A-Ⅰ、脂蛋白(a)、Lp-PLA2、年龄。ROC曲线分析显示,建立预测模型,训练集组AUC为0.905(95%CI:0.870~0.941),Youden指数为0.663;验证集组AUC为0.887(95%CI:0.828~0.945),Youden指数为0.636。采用Hosmer-Lemeshow检验对模型训练集组及验证集组的校准度进行分析(P>0.05)。临床决策曲线训练集组及验证集组预测概率区间分别为1%~85.1%和1%~83.2%。结论 利用4个预测变量建立的脑卒中预测模型,可以更好的开展患者的个体化评估,指导临床医师的决策,对缺血性脑卒中患者早期进行干预治疗,减少发病及延缓病程发展。Objective To construct a clinical model for predicting acute and subacute ischemic stroke based on Lp-PLA2 combined with stroke risk factors and common clinical indicators.Methods A total of 194 patients with stroke at acute and recovery stages(stroke group) in Nanjing Hospital of Traditional Chinese Medicine from January to December 2021 were enrolled, and 169 healthy subjects(control group) who taking physical examination were selected with age and gender as fuzzy matching objects.These 363 participates were assigned at a ratio of 7?3 into 251 cases in the training set and 112 cases in the validation set.LASSO regression was used to screen risk factors into the prediction model to establish a logistic regression model.ROC curve was drawn and the area under curve(AUC) was calculated.Hosmer-Lemeshow test was adopted for goodness-of-fit and calibration.The clinical decision curve was used to evaluate the clinical guidance of the prediction model.Finally, the clinical prediction model was visualized with a nomogram.Results There were statistical differences in hypertension, diabetes, coronary heart disease, age, fibrinogen, Lp-PLA2,lipoprotein(a),apolipoprotein A-Ⅰ,HDL and FTbetween the stroke group and control group(P<0.05).Significant differences were seen in lipoprotein(a),apolipoprotein A-Ⅰ,HDL and LDL between the patietns from the training set and the verification set(P<0.05).LASSO regression analysis showed that 8 variables with non-zero coefficients, including diabetes, coronary heart disease, age, fibrinogen, Lp-PLA2,lipoprotein(a),apolipoprotein A-Ⅰ,and HDL,were used to construct a logistic regression model, and finally, 4 risk factors(apolipoprotein A-Ⅰ,lipoprotein(a),Lp-PLA2,and age) were included.ROC curve analysis showed that the AUC of training set was 0.905(95%CI:0.870-0.941) and Youden index was 0.663.The AUC of validation set was 0.887(95%CI:0.828-0.945) and Youden index was 0.636.Hosmer-lemeshow test indicated the calibration of the model between the training set and verification set(P>0.0

关 键 词：1-烷基-2-乙酰甘油磷酸胆碱酯酶 卒中 脂蛋白(A) 载脂蛋白A-Ⅰ 年龄因素 

分 类 号：R743.3[医药卫生—神经病学与精神病学]