Laboratory results are assessed on the basis of reference data. The generic term "reference data" includes reference intervals, decision limits and action limits. By definition, reference ranges include 95% of the results of healthy people. Decision limits enable an efficient separation of healthy and diseased. Action limits are empirically defined limits, the exceeding of which is linked to an action instruction (example: PSA > 4µg/l ? prostate biopsy). Statistical methods are of great importance in determining these intervals and limits and in assessing their validity.
Prof. Dr. Rainer Haeckel, Dipl. Math. Werner Wosniok and Dr. Farhad Arzideh from the Institute of Statistics at the University of Bremen have collaborated on these questions for many years. In 2007, the previously casual cooperation was transferred to the DGKL Working Group on Reference Values.
„The working group usually meets twice a year, offers courses on the Reference Limit Estimator and publishes articles on the algorithms and the data obtained.“
Objectives of the working group
The working group deals with questions related to the evaluation of laboratory results
The aim is to develop solutions that can be transferred to patient care. The results will be published and presented at small conferences and at workshops during the annual congresses.
Reference Limit Estimator (RLE) Permissible uncertainty of measurement
In the clinical-chemical laboratory, a large number of parameters are measured, evaluated and documented daily. A database is set up, in which the measurement results and other relevant information such as the date of sampling, age and sex are stored.
Generally, for each measured parameter the proportion of pathological values is small compared to the total number of all values. In this case, certain model assumptions and statistical methods can be used to separate the distribution of data with pathological values and the distribution of data with non-pathological values. From the distribution of the non-pathological values, the reference limits can then be calculated as the 2.5th and 97.5th percentiles. If the data is first filtered by age or gender, age and gender related reference limits can be determined.
The advantage over the conventional method, in which a previously selected reference group is examined, is substantial: The data is already available in the laboratories’ databases and the tedious selection and examination of subjects for a reference group is no longer necessary. Furthermore, due to organisational, cost and time reasons it is almost impossible to find a sufficient number of subjects to allow for further stratifications other than by gender (e.g. according to age). In addition, the representativeness of a selected group is often questionable. If the measurement method is poorly standardized, the reference limits determined in one laboratory with the help of volunteers cannot be adopted by other laboratories. If, on the other hand, the measurement data from the laboratory's own database are used, laboratory reference limits are determined. Universally valid reference limits can only be obtained with these methods if several laboratories are involved in the determination and the measurement procedure has a high degree of standardization.
The user operates the program via an Excel interface. We recommend the use of Microsoft Excel Version 2010 or higher (32Bit or 64Bit). For users of very old Excel versions (2007 and older) a special 32Bit version is included, which however will not be further developed in the future. Since the statistical calculations are very extensive and complex in parts, the statistical analysis is carried out with the program R and using some additional modules (packages). The statistical results and graphical representations are then transferred back to the Excel user interface.
Note: Currently only 32-bit versions of Microsoft Excel are supported!
The DGKL Working Group on Reference Values has developed a concept for deriving the acceptable measurement uncertainty (acceptable imprecision and acceptable bias) from the reference interval. This concept and the underlying algorithms were described in Clin Chem Lab Med 2015 (Acceptable limits of uncertainty in laboratory medicine). The Excel program automatically calculates the permissible coefficient of variation (pCVA) and the permissible limits of the deviation of the individual value according to RiliBÄK 2008 (∆max). The table contains almost all measured variables of the RiliBÄK 2008.