General deming regression for estimating systematic bias and its confidence interval in method-comparison studies

Area of Science:

• Biostatistics
• Analytical Chemistry
• Method Comparison Studies

Background:

• Least-squares regression methods are commonly used to estimate systematic error (bias) and its confidence interval in method-comparison studies.
• Inappropriate regression assumptions can lead to inaccurate statistical estimates.
• An improved, generally applicable regression method for linearly related method-comparison data is presented, free from common simplifying assumptions.

Purpose of the Study:

• To develop and validate an improved regression analysis method for method-comparison studies.
• To address limitations in existing regression techniques, particularly ordinary linear regression (OLR) and simple Deming regression (SDR).
• To ensure statistically unbiased estimates of systematic bias and reliable confidence intervals across diverse data scenarios.

Main Methods:

• Applied theoretical equations based on the Deming approach, extended for this study.
• Utilized Monte Carlo simulations to validate the new procedure.
• Compared the performance of the new method against OLR and SDR.

Main Results:

• Ordinary linear regression (OLR) yielded unreliable confidence intervals for bias across all tested data types.
• OLR point estimates for bias were reliable only when the correlation coefficient exceeded 0.975.
• Simple Deming regression (SDR) provided reliable bias estimates but unreliable confidence intervals when method standard deviations varied with analyte concentration.

Conclusions:

• Iteratively reweighted general Deming regression demonstrated statistically unbiased estimates of systematic bias.
• This advanced method provided reliable confidence intervals for bias in all examined cases, including varying standard deviations and coefficients of variation.
• The findings highlight the superiority of iteratively reweighted general Deming regression for robust bias assessment in method-comparison studies.