A comparative study of the quality control efficacy of multiple error-introduction methods for patient-based...

This study aims to compare the quality control efficacy of multiple error-introduction methods for patient-based real-time quality control.

A total of 25 routine testing items covering clinical biochemistry, immunology, and hematology domains were included, using patient test data collected from January 2024 to December 2024 at the Xi'an Area Medical Laboratory Center. Exponentially Weighted Moving Average (EWMA) PBRTQC models were constructed using the AI-PBRTQC intelligent platform, with three weighting coefficients (λ = 0.02, 0.03, 0.05) evaluated for each item. Three error introduction approaches were compared: (1) real-world quality risk event validation, (2) manual experimental error creation, and (3) intelligent platform simulated systematic error addition. Model performance was assessed using four core metrics: error detection rate (Ped), false positive rate (FPR), false negative rate (FNR), and average number of patient samples before error detection (ANPed).

Real-world quality risk event validation for the total thyroxine (TT4) item demonstrated 100% concordance between PBRTQC alerts and IQC results during calibration events, confirming its value as the reference standard for model validation. The intelligent platform simulation method successfully identified optimal λ parameters for all 25 items, with λ = 0.05 achieving the best balance of high Ped (>90%) and low FPR (<5%) for 10 items including ALP, Urea, TT3, TT4, FT3, FT4, WBC, RBC, MCH, MCHC, while the remaining 15 items showed optimal performance at λ = 0.02 or 0.03. In contrast, the manual error creation method yielded valid parameters for vitamin B12 (VitB12, λ = 0.05, Ped = 100%, FPR = 5.27%) but failed to detect errors for TT3, TT4, and free thyroxine (FT4) across all tested λ values, suggesting limited generalizability.

The joint application of the intelligent platform simulation error method and the real-world event method can significantly improve the screening efficiency of the PBRTQC model; however, the human-made error method requires further optimization and validation. The study provides a parameter-setting basis and a validation method reference for clinical laboratories to independently construct PBRTQC models, promoting the application of PBRTQC in practical scenarios.