Metrological and Algorithmic Traceability of Machine Learning in Laboratory Medicine

BACKGROUND: The integration of machine learning (ML) into clinical laboratory medicine offers new opportunities for quality improvement. Models are now being developed to assist test interpretation, predict clinical outcomes, and support laboratory operations. However, these applications also require traceability that can accommodate both measurement science and data-driven computation. METHODS: We examine how metrological traceability relates to algorithmic traceability. We also examine three broad applications of ML in laboratory medicine and consider their different traceability requirements. RESULTS: Metrological traceability links measurement results to reference systems through documented calibration hierarchies and the evaluation of uncertainty. Algorithmic traceability, by contrast, concerns whether the computational path from source data to model output can be reconstructed and audited. ML outputs therefore require careful interpretation according to their intended use. A model may use metrologically traceable laboratory results as inputs, yet the computational transformation that follows demands its own documentation, validation, and governance. Across the three broad applications, common problems persist. Laboratory data are not always comparable across methods, information leakage can be difficult to detect, uncertainty is frequently underreported, and model performance may shift after deployment. CONCLUSIONS: We propose a functional framework that aligns metrological principles with algorithmic documentation while preserving the distinction between reference materials and training datasets. A tiered approach to regulation and standards, supported by traceability dossiers and international collaboration, may help ensure that AI/ML systems in laboratory medicine are reproducible, comparable, and clinically fit for purpose.