Volume 49, Issue 6 - Special Issue: Entering a Newer of Laboratory Data Processing and Interpretation; Issue Editors: Georg Hoffmann and Frank Klawonn

Laboratory findings provide essential information for assessing the state of health. Two concepts for the classification and evaluation of the findings are central: reference intervals (RI) and clinical decision limits (CDL), which also include therapeutic target limits. The IFCC (International Federation of Clinical Chemistry and Laboratory Medicine) with the Committee for Reference Intervals (C-RIDL) and the working group on the Global Reference Interval Database (TF-GRID) presented concepts and their sustainability with a focus on reference limits at EuroMedLab Brussels 2025. Sustainability for RIs and CDLs represents a new concept that has prerequisites like a data infrastructure and a quality and information framework. Not all of these prerequisites have been fully characterized yet, but the essential aspects are addressed in this article. Standardization with metrological traceability or, where not otherwise possible, harmonization forms the basis for sustainability of RIs and CDLs. Sustainable RIs and CDLs enhance the use and value of electronic patient records and help to establish cumulative reports.

Objectives There are currently two publicly available R packages for the indirect determination of reference intervals: reflimR and refineR. We aimed to optimize the reflimR algorithm to achieve greater concordance with refineR, including in challenging data sets. Methods We modified the truncation algorithm of the three-stage reflimR method so that the approximation to the central 95 % of the presumably non-pathological values is performed consecutively in five steps rather than all at once. Results Using the livertests dataset included in the reflimR package as an example, we demonstrate that our refined approach can effectively handle datasets containing a substantial proportion of pathological values near the reference limit. The results are more robust and coincide better with the refineR results. Conclusions The updated reflimR version 1.1.0 is available on CRAN (Comprehensive R Archive Network) and delivers reliable results in an extremely short time with moderate requirements for the number of cases.

Objectives Verification of reference intervals (RIs) is mandatory for clinical laboratories. In this context indirect methods of RI estimation may be of assistance, yet these are often limited to experts with programming or statistical skills due to the lack of standardized solutions in laboratory information systems. Methods This study evaluates two web-based platforms, goCrunch and ReferenceRangeR, developed to simplify RI estimation with indirect methods for laboratory professionals without requiring expertise in R or other statistical software. RIs estimated from real-world platelet and anion gap datasets were compared across the platforms. Results The platforms were assessed regarding data handling, statistical methods, visualization, and quality assurance. goCrunch supports both classical (Hoffmann, Bhattacharya) and modern methods (refineR, reflimR), with options for filtering, transformation, verification, and graphical outputs. ReferenceRangeR focuses on newer methods (refineR, TML, TMC, kosmic, reflimR) and includes additional features such as stratification by age and sex, age-drift detection, and RI comparison against external values. The comparison of a simple analyte (platelet count) and a complex one (anion gap) highlighted the strengths and limitations of each platform. Conclusions Both tools enable efficient RI generation and verification with minimal user intervention. While goCrunch is particularly suitable for routine applications and educational purposes, ReferenceRangeR provides enhanced analytical and comparative functionalities, thereby both tools support laboratories in meeting accreditation requirements.

Objectives In this study we evaluated the performance of the decision and regression tree–based algorithm rpart for age- and sex-partitioning of reference intervals (RIs) in an indirect setting, and compared its results with those derived from the CALIPER study. Methods Routine laboratory data for γ-glutamyltransferase (GGT), creatinine (CREA), and alkaline phosphatase (AP) were analyzed. Indirect RIs were estimated using the reflimR package, with partitions defined either by CALIPER or suggested by rpart . Corresponding partitions were compared using the Harris and Boyd method and assessed for clinical plausibility. Results rpart produced clinically and statistically meaningful partitioning schemes in mixed populations with unknown proportions of pathological values. Compared with CALIPER, rpart often yielded comparable RIs but avoided non-significant splits. Notable differences were observed in specific age groups, particularly in early childhood partitions for GGT and neonatal partitions for AP. Validation against original CALIPER RIs showed good agreement in most cases, with some broader intervals in the indirect approach. Conclusions These findings support the use of rpart not only in carefully preselected reference cohorts but also in heterogeneous routine laboratory datasets. Combined with indirect RI estimation via reflimR, automated partitioning offers a practical and reproducible approach to establishing clinically relevant RIs. Further multicenter studies should be conducted to confirm these results.

Objectives The standardization and normalization of laboratory parameters to a fixed scale, such as the zlog value, may offer significant advantages for interpretation. This method not only allows for the direct comparison of different parameters but also facilitates a way for advanced visualizations beyond the classic table. Methods We employ spider charts in conjunction with zlog transformation to enable laboratory parameters that are jointly evaluated for diagnostic purposes to be summarized in a unified visualization. Results With anemia and blood gas parameters we present two examples where specific constellations of laboratory parameters create distinguishable diagnostic patterns in the spider charts. Conclusions The zlog spider combines zlog-transformation standardization with spider diagram visualization to facilitate laboratory parameter comparison and create recognizable diagnostic patterns, potentially improving clinician comprehension and efficiency, though further case studies are needed to confirm its efficacy.

Objectives Therapeutic drug monitoring (TDM) reports are critical for precision medicine but remain labor-intensive. This study attempts a proof-of-concept approach to investigate the possibilities of large language models (LLMs) in generating TDM reports. Methods Using fictional cases based on real cases and a questionnaire-based framework, we evaluated completeness, lack of false information, evidence, appropriateness and relevance of the generated reports using the CLEAR framework. Results While basic constellations were acknowledged by the models and the TDM reports followed the structure of the prompt, challenges remain in context understanding and domain-specific nuance. Importantly, we examined only standalone LLM outputs rather than human-AI collaboration. Conclusions Our findings suggest LLM-assisted TDM reporting can reduce clinical workload and enhance standardization. Further research should focus on improving model interpretability, refining data integration, and designing systems that effectively incorporate human oversight.

The EU AI Act came into force on August 1, 2024, and since February 2, 2025, employees must be sufficiently trained in the use of AI systems (Article 4 of the EU AI Act). But what does “sufficient” AI literacy mean in the context of the EU AI Act? This article highlights the statements of the EU AI Act on AI literacy and aims to help in the development of training curricula. Medical laboratories offer numerous areas of application for AI, including improving the organization of laboratory processes and knowledge management. Even though there are no mandatory training requirements yet, basic data literacy is considered a prerequisite for AI Literacy. Three major topics are identified for AI literacy training: a general knowledge on the content of the EU AI Act, fundamentals of artificial intelligence, and context-specific training. This combination of fundamental data expertise and fundamental AI expertise will enable Article 4 of the EU AI Act to be fulfilled “to the best of ability” and lay the foundation for the transformation of medical laboratories.