Difference- and regression-based approaches for detection of bias.

Area of Science:

• Biostatistics
• Laboratory Medicine
• Analytical Chemistry

Background:

• Bias detection is critical for method evaluation and reagent lot verification in clinical laboratories.
• Commonly used statistical criteria for bias detection vary in their effectiveness.
• Simulation studies are valuable for assessing the performance of these criteria under different conditions.

Purpose of the Study:

• To statistically assess the performance of six common rejection criteria for bias detection.
• To identify the most reliable criteria for identifying bias in laboratory measurements.
• To provide evidence-based guidance for selecting appropriate statistical methods in method validation.

Main Methods:

• A simulation study was conducted to evaluate false rejection rates and bias detection probabilities.
• Six criteria were assessed: individual paired difference, mean paired difference, t-statistics (paired t-test), regression slope, regression intercept, and coefficient of determination (R²).
• Ordinary least squares, weighted least squares, and Passing-Bablok regression models were used for analysis.

Main Results:

• Rejection criteria based on regression slope, intercept, and individual sample paired differences exhibited high false rejection rates or low bias detection probability.
• T-statistics (α=0.05) performed optimally in scenarios with low measurement range ratios and low imprecision.
• Mean difference (10%) demonstrated superior performance across other range ratio and imprecision scenarios.

Conclusions:

• Objective evidence was provided on the performance of commonly used bias detection criteria.
• The findings guide laboratories in experimental design and statistical assessment for bias detection.
• Appropriate selection of statistical methods enhances the reliability of method evaluation and reagent lot verification.