暴露测量错分对研究真实性的影响

周艺彪; 杨美霞; 姜庆五

Abstract

周艺彪,杨美霞,姜庆五.暴露测量错分对研究真实性的影响[J].Chinese journal of Epidemiology,2005,26(11):919-923

暴露测量错分对研究真实性的影响

Study on the impact of exposure misclassification on the validity of a study

Received:April 16, 2005

DOI：

KeyWord: 二分暴露变量错误分类偏倚

English Key Word: Dichotomous exposure variable Misclassification Bias

FundProject:

Author Name	Affiliation	E-mail
ZHOU Yi-biao	Department of Epidemiology, School of Public Health, Fudan University, Shanghai 200032, China
YANG Mei-xia	上海市徐汇区疾病预防控制中心
JIANG Qing-wu	Department of Epidemiology, School of Public Health, Fudan University, Shanghai 200032, China	qwjiang@shmu.edu.cn

Hits: 5860

Download times: 1387

Abstract:

目的探讨二分暴露变量错分对暴露与疾病结局变量联系效应的影响. 方法利用错分概率、联系效应以及暴露比例之间的函数, 采用R 2. 0. 1软件作图来讨论错分对研究真实性的影响. 结果无差异性暴露错分使普通线性模型系数的估计值偏低, 当r=0. 5时, 灵敏度和特异度对错分偏倚产生的影响是相同的, 而r≠0. 5时, 它们对错分偏倚产生的影响是不同的. 当错分是无差异性时, 暴露比例r0趋于0或1, 估计的OR值都趋近于1. 差异性错分所致的偏倚远比无差异性错分所致的偏倚来得复杂得多, 可使OR值趋于或远离无效值. 结论暴露错分对暴露与结局变量之间联系效应的影响相当复杂, 在研究中对错分偏倚的识别、控制、评估以及正确地对结果进行解释均十分必要.

English Abstract:

Objective To explore the impact of misclassification in binary explanatory variables on the effects associated with 'exposure-disease'. Methods Based upon the functions of probabilities on misclassification, effects of association and proportions of exposure, the ' R Project for Statistical Computing' method was used to analyze the impact of misclassification on the validity of a study. Results To the linear model case, the effect of nondifferential misclassification serves as an attenuating bias. When r = 0. 5, the bias is symmetric in both sensitivity and specificity but when r≠0. 5, the bias is not symmetric in sensitivity and specificity. When misclassification is nondifferential, estimated odds ratio tends to be 1 while the exposure prevalence in the control tends to be 0 or 1. Bias seems to be very complex in differential misclassification than in nondifferential misclassification that can make OR tend to or be away from the null value. Conclusion The impact of exposure misclassification on the effect associated with exposure-disease is complicated, hence necessary to understand, to control, and to assess bias of misclassification in order to correctly interpret the Results of a study.

View Fulltext Html FullText View/Add Comment Download reader