AI-Driven Patient-Based Real-Time Quality Control System Optimization And Bias Alarming Analysis

Abstract

Patient-based real-time quality control (PBRTQC) is widely used in clinical laboratories, providing a sophisticated approach to monitoring the analytical performance of laboratory instruments using the results generated from actual patient samples. Compared with the traditional quality control (QC), which easily triggers false alarms, delayed error detection, and limited specificity and sensitivity, PBRTQC overcomes critical gaps by proficiently adopting machine learning. Implemented under a set of standardized rules and methods, PBRTQC enables intelligent error detection, handling of complex scenarios, and competitive operational efficiency gains. However, PBRTQC is still facing some challenges due to an imbalance of data resources and a limited ability to identify real-world issues when models are only trained on simulated datasets. This article discusses the development of PBRTQC from the initial Westgard rule to several typical AI-driven machine learning models, including comparing their mechanisms and the disparity in accuracy performance. The compelling advantages of the machine learning models will be highlighted, such as their highly precise model structure and algorithm. At the same time, current limitations of ML models and critical thinking for future projection are also discussed.