老年脑卒中患者卒中后认知功能障碍风险预测模型的决策曲线分析

Risk prediction model for post-stroke cognitive impairment in elderly patients:a decision curve analysis

目的基于血清代谢指标建立老年脑卒中患者卒中后认知功能障碍的风险预测模型,采用决策曲线分析其风险模型的预测价值。方法选取2019年8月至2022年8月青岛市中心医院神经外科收治的老年脑卒中患者297例,根据随访结果,最终纳入294例,按照3:1比例将其分为训练集206例和验证集88例。训练集根据是否发生认知功能障碍分为认知障碍组88例和非认知障碍组118例。采用logistic回归分析老年脑卒中患者卒中后发生认知功能障碍的影响因素,并构建logistic回归预测模型,通过R4.1.3绘制列线图对logistic回归预测模型进行可视化处理,并采用ROC曲线、决策曲线分析logistic回归模型预测效能,通过验证集验证该模型的预测效能。结果多因素logistic回归分析显示,训练集三酰甘油(TG,OR=1.266,95%CI:1.089~1.471,P=0.002)、低密度脂蛋白胆固醇(LDL-C,OR=1.321,95%CI:1.136~1.537,P=0.000)、血清胱抑素C(Cys C,OR=1.847,95%CI:1.421~2.401,P=0.000)、淀粉样蛋白A(SAA,OR=1.120,95%CI:1.057~1.187,P=0.000)是老年脑卒中患者卒中后发生认知功能障碍的独立危险因素。ROC曲线显示,训练集中TG、LDL-C、Cys C以及SAA对老年脑卒中后认知功能障碍预测的曲线下面积(AUC)分别为0.732、0.726、0.756、0.736,联合预测的AUC为0.891。验证集中TG、LDL-C、Cys C以及SAA对老年脑卒中后认知功能障碍预测的AUC分别为0.759、0.703、0.769、0.756,联合预测的AUC为0.914。2种模型预测的一致性较好,联合预测模型在训练集和验证集中的准确性分别为83.98%、86.36%,均高于单项指标预测模型。决策曲线分析显示,训练集和验证集的阈值概率分别为11%~48%和13%~45%,此时对老年脑卒中患者进行临床干预后可能受益最大。结论TG、LDL-C、Cys C以及SAA是老年脑卒中患者卒中后发生认知功能障碍的独立危险因素,基于血清代谢指标建立的联合决策曲线预测模型具有较高的预测效能。

bjective To construct a risk prediction model of post-stroke cognitive impairment in elderly patients based on serum metabolic indicators and assess its predictive value by decision curve analysis.Methods From August 2019 to August 2022,297 elderly stroke patients admitted to our department were prospectively enrolled,and according to the follow-up results,294 of them were finally included and then randomly assigned into a training set(206 cases)and a verification set(88 cases)in a ratio of 3:1.The patients in the training set were divided into cognitive impairment group(88 cases)and non-cognitive impairment group(118 cases)according to whether cognitive impairment occurred or not.Logistic regression analysis was used to analyze the influencing factors of cognitive dysfunction in elderly stroke patients,and a logistic regression prediction model was constructed.The logistic regression prediction model was visualized by drawing a nomogram in R4.1.3.Its prediction efficiency was analyzed by ROC curve and decision curve analyses,and verified with the verification set.Results Multivariate logistic regression analysis showed that in the training set,TG(OR=1.266,95%CI:1.089-1.471,P=0.002),LDL-C(OR=1.321,95%CI:1.136-1.537,P=0.000),Cys C(OR=1.847,95%CI:1.421-2.401,P=0.000)and SAA(OR=1.120,95%CI:1.057-1.187,P=0.000)were independent risk factors for post-stroke cognitive impairment in elderly stroke patients.ROC curve analysis indicated that the AUC value of TG,HDL-C,Cys C and SAA for predicting cognitive dysfunction in the elderly after stroke in the training set was 0.732,0.726,0.756 and 0.736,respectively,and the AUC value of above indicators combined together in the prediction was 0.891.In the verification set,the AUC value of above indicators for the prediction was 0.759,0.703,0.769 and 0.756,respectively,and the AUC value of the combined indicators was 0.914.The two models had good prediction consistency.The accuracy of the combined prediction model in the training set and the verification set were 83.98%and 86.36%,respectively,and all of them were higher than the prediction of single indicator.Decision curve analysis showed that the threshold probabilities of the training set and the validation set were 11%-48%and 13%-45%,respectively,which may benefit the most from clinical intervention in elderly stroke patients.Conclusion Elevated levels of TG,HDL-C,Cys C and SAA are independent risk factors for cognitive dysfunction in elderly stroke patients after stroke.The combined decision curve prediction model based on serum metabolic indicators shows higher predictive efficacy.