Tian Lu, Durazo-Arvizu Ramón A, Myers Gary, Brooks Steve, Sarafin Kurtis, Sempos Christopher T
Department of Health Research and Policy, Stanford University, Palo Alto, CA, U.S.A.
Stat Med. 2014 Nov 10;33(25):4420-36. doi: 10.1002/sim.6235. Epub 2014 Jun 17.
In clinical chemistry and medical research, there is often a need to calibrate the values obtained from an old or discontinued laboratory procedure to the values obtained from a new or currently used laboratory method. The objective of the calibration study is to identify a transformation that can be used to convert the test values of one laboratory measurement procedure into the values that would be obtained using another measurement procedure. However, in the presence of heteroscedastic measurement error, there is no good statistical method available for estimating the transformation. In this paper, we propose a set of statistical methods for a calibration study when the magnitude of the measurement error is proportional to the underlying true level. The corresponding sample size estimation method for conducting a calibration study is discussed as well. The proposed new method is theoretically justified and evaluated for its finite sample properties via an extensive numerical study. Two examples based on real data are used to illustrate the procedure.
在临床化学和医学研究中，常常需要将从旧的或已停用的实验室程序获得的值校准为从新的或当前使用的实验室方法获得的值。校准研究的目的是确定一种变换，该变换可用于将一种实验室测量程序的测试值转换为使用另一种测量程序会获得的值。然而，在存在异方差测量误差的情况下，没有可用的良好统计方法来估计这种变换。在本文中，当测量误差的大小与潜在真实水平成比例时，我们提出了一组用于校准研究的统计方法。还讨论了进行校准研究的相应样本量估计方法。通过广泛的数值研究，对所提出的新方法进行了理论论证并评估了其有限样本特性。使用两个基于实际数据的例子来说明该过程。
