Compliance and Internal Auditing Machine Learning Framework for GAHAR clinical laboratory standards (edition 2021) Based on the Twelve Quality System Essentials in Egyptian Medical Laboratories

Abstract Background Clinical laboratories are central to safe and effective health care, and their performance depends on robust quality management and alignment with accreditation standards. Clinical laboratory standards of the General Authority for Healthcare Accreditation and Regulation (GAHAR) in Egypt, GAHAR clinical laboratory standards (Edition 2021) provide national requirements, while the Quality System Essentials (QSE) framework offers an internationally recognized structure for laboratory quality. The potential of machine learning to support internal auditing and accreditation readiness in this context remains underexplored. Methods A qualitative cross-sectional comparative mapping was conducted between the Evidence of Compliances (EOCs) in GAHAR clinical laboratory standards (Edition 2021) and the twelve components of the QSE framework (4,8). Each and every EOC in GAHAR clinical laboratory standards (Edition 2021) was assigned to a one or more QSE category, with counts and percentages calculated to describe coverage and distribution of all EOCs across the QSE framework. Mapping decisions were validated by structured peer review from senior laboratory experts and accreditation consultants. In parallel, a hybrid machine learning internal auditing tool was developed and piloted on an internal audit question set labeled for compliance status. The prototype used sentence-level embeddingsas input to different model architectures and generated structured audit outputs including a tabular report and narrative recommendations. Results A total of 468 Evidence of Compliance (EOCs)in the GAHAR clinical laboratory standards (Edition 2021) were mapped to the twelve QSE components. Overall, 93% of QSE were addressed. In regard of distribution of EOCs across the 12 QSEs, coverage was highest for process management, organisation, and facilities and safety, while assessments and continual improvement were least represented. These patterns indicate a strong emphasis on operational processes, with limited explicit requirements for formal assessments and iterative improvement cycles. The hybrid machine learning prototype achieved its best performance with an embedding based large language model (LLM), which reached an accuracy of about 93%, a precision of 0.90, and an F1 score of 0.91 in classifying audit questions by compliance state. The prototype also produced structured internal audit reports with linked evidence and suggested follow-up actions. Conclusion GAHAR clinical laboratory standards (Edition 2021) demonstrate substantial alignment with the 12 QSE framework but underrepresent assessment and continual improvement functions that are critical for long-term quality maturity. A machine learning enabled internal auditing tool can provide accurate, scalable support for identifying compliance gaps and generating actionable recommendations. Careful attention to governance, transparency, and local validation is required before such tools are adopted at scale.