Gange S J, Muñoz A, Wang J S, Groopman J D
Department of Epidemiology, Johns Hopkins School of Hygiene and Public Health, Baltimore, Maryland 21205, USA.
Cancer Epidemiol Biomarkers Prev. 1996 Jan;5(1):57-61.
In any immunoassay experiment for the detection of molecular biomarkers, a nonlinear calibration curve is constructed to relate fixed biomarker concentrations to observed tracer levels. The biomarker concentration in an experimental sample can then be estimated by projecting the experimental tracer measurements through the inverse of the calibration curve. Once an estimate of the biomarker level has been calculated, it is often of interest to determine its variability. Typically, methods for estimating this variability assume that the biomarker variability is due solely to the uncertainty in the estimation of the calibration curve. A more complete analysis would combine this uncertainty with the variability in the processing and measurements of the sample, including, e.g., measurement error of laboratory procedures or variation in enzymatic activity in enzyme-linked immunosorbent assays or radioactivity counts in RIAs. In this paper, we present a method of estimating the variability of inverse estimates assuming there is variation arising from both the determination of the calibration curve and from the preparation and measurement of the experimental sample. Our method uses a resampling algorithm that avoids requiring many distributional assumptions present in alternative procedures, can be easily implemented, and is generalizable to any immunoassay procedure. Methods for incorporating our results in the estimation of variability for planning and analyzing biomarker experiments are discussed. We provide an example using RIA data for aflatoxin B1 detection. These biomarkers for aflatoxin exposure are used in the analysis of serum aflatoxin adduct levels in human and experimental samples, and they are important in hepatocellular carcinoma research.
在任何用于检测分子生物标志物的免疫分析实验中，都会构建一条非线性校准曲线，以将固定的生物标志物浓度与观察到的示踪剂水平联系起来。然后，可以通过校准曲线的反函数对实验示踪剂测量值进行投影，来估计实验样品中的生物标志物浓度。一旦计算出生物标志物水平的估计值，通常就会想要确定其变异性。通常，估计这种变异性的方法假定生物标志物的变异性仅源于校准曲线估计中的不确定性。更全面的分析会将这种不确定性与样品处理和测量中的变异性结合起来，包括例如实验室程序的测量误差、酶联免疫吸附测定中酶活性的变化或放射免疫分析中的放射性计数。在本文中，我们提出了一种估计反函数估计值变异性的方法，假设校准曲线的确定以及实验样品的制备和测量都会产生变异性。我们的方法使用一种重采样算法，该算法无需许多替代程序中存在的分布假设，易于实现，并且可推广到任何免疫分析程序。讨论了将我们的结果纳入生物标志物实验规划和分析变异性估计的方法。我们提供了一个使用放射免疫分析数据检测黄曲霉毒素B1的示例。这些黄曲霉毒素暴露的生物标志物用于分析人类和实验样品中的血清黄曲霉毒素加合物水平，并且在肝细胞癌研究中很重要。
