Abstract
Abstract
Semen analysis is the primary diagnostic tool for male fertility evaluation, yet its subjective nature introduces substantial interlaboratory variation that compromises result reliability; despite over three decades of external quality assessment (EQA) practice, coefficients of variation (CV) for core semen parameters remain unacceptably high worldwide. Synthesising evidence from 34 years of global EQA programmes, this review documents interlaboratory CVs of 12.7–138.0 %, 17.0–127.0 %, and 7–375 % for concentration, motility, and morphology, and identifies three structural dilemmas: (1) peer self-referencing bias in assigned value determination, (2) disconnection of allowable total error (TEa) from clinical need, (3) a cognitive threshold effect in morphological interpretation. The principal international schemes – CAP (USA), UK NEQAS, NCCL (China), and QUIP (Germany) – diverge markedly in specimen type, assigned value strategy, and acceptance criteria. We critically evaluate the transition from state-of-the-art to biological variation (BV)–based analytical quality specifications (APS), examine performance differentials between assisted reproductive technology (ART) centres and general clinical laboratories, and assess emerging technologies including AI-assisted morphology assessment and digital EQA platforms. The field is undergoing a paradigm shift from statistical peer consensus toward clinically driven BV-based APS, derived from intra-individual (CV1) and inter-individual (CVG) variation of semen parameters; However, AI systems exhibit systematic analytical bias compared with manual microscopy, which stems from training-set composition, annotator subjectivity, and visual fatigue caused by high-throughput workloads. Independent calibration and version-control validation should be completed prior to clinical application. Priority efforts encompass EQA framework harmonisation aligned with WHO Sixth Edition (2021) and ISO 23162:2021, measurement uncertainty incorporation into TEa derivation, prospective AI algorithm validation, split-sample parallel testing, and disaggregation of morphological APS by defect category (head, midpiece, and tail).