An appraisal of statistical procedures used in derivation of reference intervals

When conducting studies to derive reference intervals (RIs), various statistical procedures are commonly applied at each step, from the planning stages to final computation of RIs. Determination of the necessary sample size is an important consideration, and evaluation of at least 400 individuals in each subgroup has been recommended to establish reliable common RIs in multicenter studies. Multiple regression analysis allows identification of the most important factors contributing to variation in test results, while accounting for possible confounding relationships among these factors. Of the various approaches proposed for judging the necessity of partitioning reference values, nested analysis of variance (ANOVA) is the likely method of choice owing to its ability to handle multiple groups and being able to adjust for multiple factors. Box-Cox power transformation often has been used to transform data to a Gaussian distribution for parametric computation of RIs. However, this transformation occasionally fails. Therefore, the non-parametric method based on determination of the 2.5 and 97.5 percentiles following sorting of the data, has been recommended for general use. The performance of the Box-Cox transformation can be improved by introducing an additional parameter representing the origin of transformation. In simulations, the confidence intervals (CIs) of reference limits (RLs) calculated by the parametric method were narrower than those calculated by the non-parametric approach. However, the margin of difference was rather small owing to additional variability in parametrically-determined RLs introduced by estimation of parameters for the Box-Cox transformation. The parametric calculation method may have an advantage over the non-parametric method in allowing identification and exclusion of extreme values during RI computation.