Traceable machine learning real-time quality control based on patient data.

Area of Science:

• Clinical Chemistry and Laboratory Medicine
• Machine Learning in Healthcare
• Metrology and Quality Control

Background:

• Patient-based real-time quality control (PBRTQC) is an emerging tool for internal quality control (iQC) in clinical laboratories.
• Concerns regarding PBRTQC's performance and real-world clinical applicability persist.
• There is a need for robust, accessible, and validated real-time QC methods.

Purpose of the Study:

• To develop an easily accessible patient-based real-time QC method using machine learning (ML).
• To ensure the ML-based QC is traceable to standard reference data from the National Institute of Metrology of China (NIM).
• To compare the performance of the novel ML-based QC with existing PBRTQC methods for clinical validity.

Main Methods:

• Collected over 1.195 million patient testing results for five representative biochemistry analytes.
• Developed a machine learning internal quality control (MLiQC) model using the Random Forest algorithm.
• Validated MLiQC through metrological traceability to NIM reference data and comparison with four IFCC-recommended PBRTQC methods.

Main Results:

• MLiQC demonstrated low uncertainty (0.14% for albumin) when evaluated against NIM standard reference data.
• MLiQC significantly reduced the average number of patient samples needed for bias detection (ANPed) from 600 to 20 compared to PBRTQC.
• MLiQC outperformed all tested PBRTQC methods in median and 95% quantile of sample detection (MNPed and 95NPed) for over 90% of analytes.

Conclusions:

• MLiQC offers superior performance and is highly suitable for real-world clinical laboratory settings.
• The validation confirms MLiQC's satisfactory performance through both algorithmic traceability and clinical effectiveness.
• This ML-driven approach represents a significant advancement in laboratory quality control.