Resampling approach for determination of the method for reference interval calculation in clinical laboratory practice.

Reference intervals (RI) play a key role in clinical interpretation of laboratory test results. Numerous articles are devoted to analyzing and discussing various methods of RI determination. The two most widely used approaches are the parametric method, which assumes data normality, and a nonparametric, rank-based procedure. The decision about which method to use is usually made arbitrarily. The goal of this study was to demonstrate that using a resampling approach for the comparison of RI determination techniques could help researchers select the right procedure. Three methods of RI calculation-parametric, transformed parametric, and quantile-based bootstrapping-were applied to multiple random samples drawn from 81 values of complement factor B observations and from a computer-simulated normally distributed population. It was shown that differences in RI between legitimate methods could be up to 20% and even more. The transformed parametric method was found to be the best method for the calculation of RI of non-normally distributed factor B estimations, producing an unbiased RI and the lowest confidence limits and interquartile ranges. For a simulated Gaussian population, parametric calculations, as expected, were the best; quantile-based bootstrapping produced biased results at low sample sizes, and the transformed parametric method generated heavily biased RI. The resampling approach could help compare different RI calculation methods. An algorithm showing a resampling procedure for choosing the appropriate method for RI calculations is included.