Minimizing false-positive errors in hormonal pulse detection.

Abstract

To explore the problem of type I (false-positive) statistical errors associated with enumerating endocrine pulses, we used the analysis of immunoactive luteinizing hormone (LH) pulses as a paradigm. In this system, the distribution of measurement error was found to approximate a Gaussian pattern. Moreover, the choice of a fixed threshold criterion to identify a hormone pulse (as generally undertaken to date) was shown to yield significantly different false-positive error rates under different experimental conditions. Therefore, we developed a technique to minimize the detection of false-positive signals and to maintain an essentially uniform error rate among different experimental groups. This simple technique requires that the pulse-detection threshold be adjusted in relation to the degree of intra-assay measurement error present. Our method should facilitate the valid comparison of endocrine pulse frequencies in normal physiological states vs. those associated with hormonal deficiency when measurement error is typically greater. In addition, this approach will aid in a more meaningful assessment of pulse concordance between two or more different hormonal species (i.e., in different assays) and will assist in comparisons of pulse properties quantitated in different laboratories. In summary, rates of false-positive pulse detection using conventional fixed-threshold criteria are materially influenced by even small differences in within-assay variance. We present a technique to minimize this type I statistical error and to maintain an essentially uniform error rate among different experimental groups.