Machine learning-based nonlinear regression-adjusted real-time quality control modeling: a multi-center study.

Area of Science:

• Clinical chemistry
• Laboratory medicine
• Machine learning in healthcare

Background:

• Patient-based real-time quality control (PBRTQC) is crucial for monitoring laboratory testing performance.
• Concerns exist regarding the generalizability of current PBRTQC methods across different settings and analytes.
• There is a need for more robust and widely applicable PBRTQC tools.

Purpose of the Study:

• To develop a machine learning, nonlinear regression-adjusted, patient-based real-time quality control (mNL-PBRTQC) model.
• To create an mNL-PBRTQC with enhanced generalizability for real-world laboratory settings.
• To evaluate the performance of mNL-PBRTQC against existing PBRTQC models.

Main Methods:

• Computer simulation was used to introduce artificial biases into patient population data for 10 measurands.
• The mNL-PBRTQC model was trained on data from eight hospital laboratories and validated on independent datasets from three other hospitals.
• Performance was compared with the IFCC PBRTQC model and linear regression-adjusted real-time quality control (L-RARTQC).

Main Results:

• The mNL-PBRTQC model demonstrated superior performance compared to IFCC PBRTQC and L-RARTQC across all measurands and introduced biases in three independent test datasets.
• For platelet analysis with a 20% bias (positive and negative), mNL-PBRTQC exhibited the smallest error detection uncertainty at median and maximum values.
• The model showed significant improvements in error detection accuracy, particularly for unstable analytes.

Conclusions:

• The mNL-PBRTQC framework is a robust machine learning approach for laboratory quality control.
• It enables accurate detection of errors, especially for analytes prone to instability.
• The model is effective in identifying even small biases, enhancing overall testing reliability.