Effective plots to assess bias and precision in method comparison studies

Area of Science:

• Biostatistics
• Clinical Research Methodology
• Measurement Error Analysis

Background:

• Bland and Altman's limits of agreement are standard for assessing measurement agreement in clinical research.
• These plots can be misleading when measurement errors have unequal variances (heteroscedasticity).
• Existing methods may incorrectly suggest bias or trends due to heteroscedasticity.

Purpose of the Study:

• To explain the origins of bias in agreement plots with heteroscedastic measurement errors.
• To introduce novel "bias plot" and "precision plot" for visual assessment of measurement method performance.
• To provide an interpretable alternative to Bland and Altman plots when heteroscedasticity is present.

Main Methods:

• Utilized a modeling framework allowing heteroscedastic measurement errors dependent on a latent trait.
• Developed a simplified estimation procedure for the proposed models.
• Employed simulations to validate the performance of the new estimation procedure.

Main Results:

• Illustrated how heteroscedastic errors can lead to misleading trends in traditional Bland and Altman plots.
• Demonstrated the utility of the new "bias plot" and "precision plot" for clear visual appraisal.
• Showcased the effectiveness of the simplified estimation procedure through simulations.

Conclusions:

• The proposed "bias plot" and "precision plot" offer a robust and interpretable method for evaluating new measurement techniques.
• The simplified estimation procedure is practical for clinical researchers dealing with heteroscedastic measurement errors.
• These new graphical tools enhance the assessment of measurement agreement beyond traditional methods.