回归校准和模拟外推对测量误差的校正效果研究  

The Correction Effect Analysis of Regression Calibration and Simulation Extrapolation for Measurement Error

作　　者：陈霞[1] 张桥[2] 陈永杰[1] 李慧婷[1] 张秋菊[1] 刘美娜[1] 

机构地区：[1]哈尔滨医科大学公共卫生学院卫生统计学教研室,150081 [2]哈尔滨医科大学医务处

出　　处：《中国卫生统计》2014年第5期741-745,共5页Chinese Journal of Health Statistics

基　　金：国家科技支撑计划(2011BAIO9B02)

摘　　要：目的探讨回归校准法(RC)和模拟外推法(SIMEX)对logistic回归中测量误差的校正效果。方法通过SAS软件产生有测量误差的模拟数据,用RC和SIMEX对测量误差进行校正,对比设定的真实β值和校正后的β*值之间的差别,以评价校正效果。结果当X可精确测量时,在设定的σ2u条件下,RC1的校正效果较好;P-SIMEX仅当σ2u很小时校正效果较好。当X不可精确测量时,随着测量误差的增大,E-SIMEX的校正效果降低,而RC2的校正效果相对较稳定。结论无论X是否可测,在经典测量误差模型前提下,RC对logistic回归模型中测量误差的校正效果优于SIMEX,建议应用RC校正测量误差。Objective To estimate the correction effect of regression calibration（ RC） and simulation extrapolation（ SIMEX） for measurement error in logistic regression model. Methods We simulated datasets including given measurement error by SAS software,and applied RC and SIMEX to correct measurement error. Then evaluated the correction effect by comparing the difference between true β value and corrected β＊value. Results If X could be accurately measured,RC1 could achieve preferable correction effect under the given σ2u,while P-SIMEX could only do well when the value σ2uwere rather minimal. If X was unobservable,the effect of E-SIMEX appeared worse with the increase of measurement error,while RC2 performed relatively stable with different σ2uvalues. Conclusion Whether or not X could be accurately measured,RC worked better than SIMEX in the classical measurement error model condition. We recommend RC on the basis of our simulation results.

关 键 词：测量误差 回归校准法 模拟外推法 LOGISTIC回归 回归稀释 

分 类 号：R195.1[医药卫生—卫生统计学]