Requiem for the Passing-Bablok nonparametric regression in assessing the agreement between two measurement methods

Area of Science:

• Biostatistics
• Laboratory Medicine
• Analytical Chemistry

Background:

• Regulatory guidelines often suggest non-parametric Passing-Bablok regression for assessing agreement between two measurement methods.
• However, the standard interpretation of Passing-Bablok confidence intervals for slope and intercept is statistically inappropriate for equivalence testing.

Purpose of the Study:

• To evaluate the accuracy of the non-parametric Passing-Bablok regression method in assessing measurement agreement.
• To compare the performance of Passing-Bablok regression with other regression methods (Theil, Ordinary Least Squares, Deming) under various simulated conditions.

Main Methods:

• Exhaustive simulations of structural models with varying slope, intercept, and measurement error.
• Calculation of slope and intercept bias for four regression techniques.
• Assessment of agreement percentages using the Passing-Bablok method and comparison with equivalence thresholds.

Main Results:

• Passing-Bablok regression provides unbiased estimates, comparable to Deming's regression.
• The method's Type I error rate is near 0.05 for slopes close to 1 but deviates significantly for slopes outside the 0.950-1.05 range.
• Confidence intervals from Passing-Bablok regression are frequently too wide to meet standard equivalence thresholds.

Conclusions:

• The standard interpretation of Passing-Bablok regression confidence intervals is not suitable for robust measurement agreement assessment.
• The wide confidence intervals limit the method's utility in establishing equivalence according to population or individual equivalence models.
• Alternative statistical approaches may be necessary for reliable agreement evaluation in laboratory settings.