多结局生存分析模型与Cox模型的随机模拟比较  

Comparison Between Multivariate Generalization of the Cox Proportional Hazards Models and Cox Proportional Hazards Model by Simulation

作　　者：高峻[1] 董伟[1] 高尔生[2] 赵耐青[1] 

机构地区：[1]复旦大学公共卫生学院社会医学与卫生统计教研室,200032 [2]上海市计划生育科学研究所,200032

出　　处：《中国卫生统计》2007年第3期248-250,254,共4页Chinese Journal of Health Statistics

摘　　要：目的通过随机模拟评价不同多结局生存分析模型的特点。方法利用随机模拟数据,比较多结局生存分析模型、将多个结局视为一种结局的单结局Cox模型及将各个结局分开拟合多个单结局Cox模型的回归系数的估计精度。结果多结局生存分析模型回归系数的估计最准确,95%可信区间包含回归系数的百分比最高,95%可信区间不包含0的百分比也最高。结论采用将各个结局分开单独拟合Cox模型和将多种结局视为单一结局拟合Cox模型分析多结局生存数据会导致回归系数估计不准及检验效能的降低。Evaluate the characteristics of multivariate generalization of the Cox proportional hazards models by simulation. Methods Compare estimate biases of three strategies of analysis （fitting multivariate generalization of the Cox proportional hazards models, fitting Cox proportional hazards model by assuming all the outcomes are same, fitting Cox proportional hazards model by dividing different outcomes） by simulated data. Results Estimators of multivariate generalization of the Cox proportional hazards models are the most closed to it＇s real values, the proportion of including estimator and the proportion of not-including zero in 95%CI of according estimator is the highest in three analysis strategies. Conclusion Fitting Cox proportional hazards model by assuming all the outcomes are same and fitting Cox proportional hazards model by dividing different outcomes will conduce great biases in estimating coefficients and lower the efficiency.

关 键 词：多结局生存分析 夹心方差估计 风险区间 

分 类 号：R195[医药卫生—卫生统计学]