The local influence analysis is an important problem in statistical inference and some models have been discussed in many literatures This paper deals with the problem of assessing local influences in a multivariate t...The local influence analysis is an important problem in statistical inference and some models have been discussed in many literatures This paper deals with the problem of assessing local influences in a multivariate t-model with Rao's simple struc-ture(RSS). Based on Cook's likelihood displacement, the effects of some minor perturbation on the statistical inference is assessed. As an application, a common covariance-weighted perturbation is thoroughly discussed.展开更多
文摘The local influence analysis is an important problem in statistical inference and some models have been discussed in many literatures This paper deals with the problem of assessing local influences in a multivariate t-model with Rao's simple struc-ture(RSS). Based on Cook's likelihood displacement, the effects of some minor perturbation on the statistical inference is assessed. As an application, a common covariance-weighted perturbation is thoroughly discussed.
文摘目的:探讨心脏生物标志物N-末端B型脑钠肽前体(N-terminal pro-brain natriuretic peptide,NT-proBNP)和高敏肌钙蛋白T(high-sensitivity cardiac troponin T,hs-cTnT)在急性缺血性卒中(acute ischemic stroke,AIS)患者长期死亡风险预测中的价值,并构建和验证相关预测模型。方法:本研究为单中心回顾性研究,连续入选2022年1—12月在南京医科大学第一附属医院接受取栓治疗AIS患者,随访2年。通过Cox回归和LASSO回归筛选全因死亡相关因素,构建3种预测模型,分别为基础模型、模型1(基础模型+NT-proBNP)和模型2(基础模型+hs-cTnT),并比较不同模型的预测能力。结果:最终纳入230例患者,按3∶2比例随机分为训练集(n=146)和测试集(n=84)。随访期间共发生83例全因死亡事件,死亡率为37.2%。多因素Cox回归显示,NTproBNP每升高1 000 pg/mL,2年全因死亡风险增加27%(HR=1.27,95%CI:1.15~1.40,P <0.001);而ln(hs-cTnT)升高与死亡风险无显著关联(HR=1.11,95%CI:0.89~1.38,P=0.372)。通过Cox回归和LASSO回归最终筛选出以下与全因死亡风险相关的变量:既往房颤、术后美国国立卫生院卒中量表(National Institutes of Health Stroke Scale,NIHSS)评分、基线血红蛋白、白细胞计数以及随机血糖,并基于此构建基础模型。基础模型训练集和测试集的受试者工作特征曲线下面积(area under the curve,AUC)分别为0.816和0.778。模型1的训练集和测试集的AUC分别为0.866和0.799,提高了对全因死亡风险的预测能力。模型2的训练集和测试集的AUC分别为0.811和0.788,对全因死亡风险的预测能力提升不明显。结论:NT-proBNP是AIS患者全因死亡的独立预测因子,可提高基于传统临床指标模型的死亡风险预测能力,辅助AIS患者的个体化管理。
