CORRECTION OF LOGISTIC-REGRESSION RELATIVE RISK ESTIMATES AND CONFIDENCE-INTERVALS FOR MEASUREMENT ERROR - THE CASE OF MULTIPLE COVARIATES MEASURED WITH ERROR
CORRECTION OF LOGISTIC-REGRESSION RELATIVE RISK ESTIMATES AND CONFIDENCE-INTERVALS FOR MEASUREMENT ERROR - THE CASE OF MULTIPLE COVARIATES MEASURED WITH ERROR
复制标题
DOI：
10.1093/oxfordjournals.aje.a115715
复制
发表时间：
1990-10-01
影响因子：
5
通讯作者：
WILLETT, WC
中科院分区：
文献类型：
--
作者：
ROSNER, B;SPIEGELMAN, D;WILLETT, WC
如果在同一多元逻辑回归模型中考虑疾病的几个危险因素，并且其中一些危险因素的测量存在误差，则与这些因素相对应的相对风险的点和区间估计值可能偏向于零值或远离零值。提供了一种方法，用于校正从一个或多个连续变量的测量误差的逻辑回归中获得的相对风险的点和区间估计。该方法需要单独的验证研究来估计多变量线性回归模型的系数，该模型将替代变量与真实危险因素的向量联系起来。其他作者也提出了类似的方法，但没有人提供一种校正置信区间的方法，其中包括由于验证研究中测量误差参数估计而产生的变异性成分。一项关于膳食脂肪、卡路里和酒精与乳腺癌关系的前瞻性研究以及一项用于评估这些营养素的问卷调查的验证研究提供了一个例子。在校正测量误差之前，酒精摄入量增加25克的年龄调整相对危险度为1.33(95%可信区间（CI） 1.14-1.55)；校正测量误差后，相对危险度增加到1.62 （95% CI 1.23-2.12）。同样，饱和脂肪摄入量每增加10克，年龄调整后的相对风险为0.94 (95% CI 0.83-1.06)；校正测量误差后，相对危险度为0.84 （95% CI 0.59-1.20）。这些结果表明，未能发现乳腺癌风险与饱和脂肪摄入量之间存在实质性的正相关关系，不能用脂肪、卡路里或酒精的测量误差来解释。
If several risk factors for disease are considered in the same multiple logistic regression model, and some of these risk factors are measured with error, the point and interval estimates of relative risk corresponding to any of these factors may be biased either toward or away from the null value. A method is provided for correcting point and interval estimates of relative risk obtained from logistic regression for measurement error in one or more continuous variables. The method requires a separate validation study to estimate the coefficients from the multivariate linear regression model relating the surrogate variables to the vector of true risk factors. Similar methods have been suggested by other authors, but none provides a means of correcting the confidence intervals which include a component of variability due to estimation of the measurement error parameters from a validation study. An example is provided from a prospective study of dietary fat, calories, and alcohol in relation to breast cancer, and from a validation study of the questionnaire used to assess these nutrients. Before correcting for measurement error, the age-adjusted relative risk for a 25 g increment in alcohol intake was 1.33 (95% confidence interval (CI) 1.14-1.55); after correcting for measurement error, the relative risk increased to 1.62 (95% CI 1.23-2.12). Similarly, for a 10 g increment in saturated fat intake, the age-adjusted relative risk was 0.94 (95% CI 0.83-1.06); after correcting for measurement error, the relative risk was 0.84 (95% CI 0.59-1.20). These results indicate that the failure to find a substantial positive association between breast cancer risk and saturated fat intake cannot be explained by measurement error in fat, calories, or alcohol.