Indirect methods for the verification of reference intervals in laboratory medicine.

Abstract

Despite the obligation imposed by national and international accreditation standards to periodically verify reference intervals against local conditions, this task is performed infrequently in most clinical laboratories due to the perceived complexity relative to the resources available. The CLSI/IFCC C28-A3 standard recommends a simple 20-sample binomial procedure whose statistical power is, however, fundamentally inadequate: it cannot reliably detect intervals that are too wide, requires sample sizes approaching n = 100 to achieve acceptable sensitivity for shifted or too narrow limits, and offers laboratories a false sense of compliance. Fifty years after the foundational conceptual framework for reference intervals was established, a critical and timely reassessment is therefore warranted. This review evaluates the limitations of the CLSI/IFCC 20-sample verification approach and systematically examines methodological advances published since the last comprehensive review in 2018, with emphasis on indirect statistical methods capable of exceeding guideline performance. A structured literature search identified 22 publications meeting inclusion criteria, and about the same number of references were added from the authors' literature collections to ensure complete coverage of the topic, especially for more recent literature that may not yet be fully indexed. Four converging lines of evidence support replacing the binomial procedure with indirect estimation methods applied to routine laboratory data. Two open-source R packages (reflimR and refineR) implement quantitative acceptance criteria based on equivalence limits (EL) and uncertainty margins (UM), respectively. Independent multi-center and multi-method studies confirm robust performance and comparable results of indirect methods over a wide range of analytes. Machine learning techniques, particularly Gaussian mixture deconvolution and regression tree modeling, extend verification capability to datasets with high pathological prevalence or complex biological confounders. Web-based platforms now make these tools accessible without programming expertise. In conclusion, indirect verification methods have reached sufficient methodological maturity for guideline endorsement. A stepwise workflow that includes rapid screening with reflimR, confirmatory analysis with refineR, and machine learning-assisted decomposition for complex cases is proposed as a practical, statistically defensible standard for routine laboratory verification. This workflow has been implemented in a new web tool at trillium.de/VeRIf.