ARTIFICIAL INTELLIGENCE IN THE PRE-ANALYTICAL PHASE

Mario Plebani
Department of Medicine-DIMED, University of Padova-Italy

Despite decades of evidence highlighting its inherent vulnerability, the pre-analytical phase remains the Achilles’ heel of laboratory medicine, even though it represents the foundational component of the total testing process (TTP), encompassing all procedures that occur prior to sample analysis. In fact, in the preanalytical phase of laboratory testing, several critical processes ensure the integrity and accuracy of a patient’s results.

The pre-analytical phase is critical, as 60–70% of laboratory errors originate in this step. It determines sample quality and represents the most vulnerable, human-dependent phase, involving multiple actors (e.g., nurses, clinicians, transport personnel, and wards). Consequently, it is a primary target for harmonization and quality indicators, with a direct impact on diagnosis, therapy, patient outcomes, and value-based laboratory medicine (VBLM). To address these challenges, substantial efforts are underway to strengthen the pre-analytical phase through the adoption of advanced technologies, including automation, artificial intelligence (AI), and robotics. These innovations enable improvements in critical processes such as automated sample labeling and vein detection, thereby enhancing accuracy, efficiency, and patient safety. AI tools have been developed and validated for improving: a) appropriateness of test request, b) specimen handling, c) identification of wrong blood in tubes, d) clot detection. In addition, AI should allow laboratory professionals to improve sample dilution management and serum quality assessment.

However, a translational gap persists in currently developed AI tools, which are often characterized by single-center or simulated study designs, retrospective validation, and heterogeneous levels of explainability and integration into clinical workflows. The recently published European Federation of Clinical Chemistry and Laboratory Medicine (EFLM) checklist provides a rigorous, standardized, and transparent framework for the development and validation of AI/ML models in clinical laboratories and should be adopted to support their safe and effective implementation in routine clinical practice.