A Novel Tool for the Rapid and Transparent Verification of Reference Intervals in Clinical Laboratories

Abstract

Background/Objectives: We present a software package called reflimR (Version 1.0.6), which enables rapid and transparent verification of reference intervals from routine laboratory measurements. Our method makes it easy to compare the results with specified target values and facilitates the interpretation of deviations using traffic light colors. Methods: The algorithm includes three procedural steps: (a) definition of an appropriate distribution model, based on Bowley's quartile skewness, (b) iterative truncation, based on a modified boxplot method to obtain the central 95% of presumably inconspicuous results, and (c) extrapolation of reference limits from a truncated normal quantile-quantile plot. Results: All algorithms have been combined into one consolidated library, which can be called in the R environment with a single command reflim (x). Using an example dataset included in the package, we demonstrate that our method can be applied to mixed data containing a substantial proportion of pathological values. It leads to similar results as the direct guideline approach as well as the more sophisticated indirect refineR software package. As compared to the latter, reflimR works much faster and needs smaller datasets for robust estimates. For the interpretation of the results, we present an intuitive color scheme based on tolerance ranges (permissible uncertainty of laboratory results). We show that a relatively high number of published reference limits require careful reevaluation. Conclusions: The reflimR package closes the gap between direct guideline methods and the more sophisticated indirect refineR method. We recommend reflimR for the rapid routine verification of large amounts of reference limits and refineR for a careful analysis of unclear or doubtful results from this check.

Keywords: color coding; indirect method; reference interval; refineR; reflimR; verification.

Figure 1

Graphical output of the reflim function. The vertical lines represent the observed and theoretical reference limits with their respective tolerance ranges.

Figure 3

Graphical output of the lognorm function (step 1 of the reflim algorithm). The graphics represent density curves and boxplots of 1000 simulated values with arbitrary units (black: original, blue: logarithms). The left side shows normally distributed values (mean = 100, sd = 10), whereas the right side is an example of a lognormal distribution (meanlog = 4.0, sdlog = 0.4). For the calculation of Bowley’s skewness delta see Table 2.

Figure 4

Left graphic: histogram of the original values with the corresponding boxplot in black. The blue curve represents the density of the truncated values with corresponding boxplot in blue. Right graphic: number of values remaining after each step of the iterative truncation.

Figure 5

Graphical output of the truncated Q-Q plot function.

Figure 2

Typical density curves and boxplots for reference individuals (green) and patients (red) in the male cohort of the livertests dataset. The histogram in the background represents the distribution of all male individuals (n = 374).

Figure 6

Comparison of results obtained with reflimR (solid lines) and refineR (dashed line). The reflimR method is applied to healthy controls (thick line) and to all individuals (thin line). The green boxes indicate the target values determined with the direct quantile-based CLSI/IFCC method applied to the healthy controls, and the empty rectangles represent the target values derived from the literature. * = slight deviation from the literature. ** = marked deviation from the literature. The terms slight and marked refer to the yellow and red traffic light colors shown in Figure 1. For analyte-specific units on the x-axis, see Table 2.

Figure 7

Computation times of reflimR and refineR as a function of sample sizes. Each point represents the mean of 30 simulations of normally distributed values without outliers.