Use of clinical tolerance limits for assessing agreement

Area of Science:

• Biostatistics
• Measurement Error Analysis
• Statistical Methodology

Background:

• Existing methods for assessing agreement between measurement methods, such as Bland & Altman limits, have limitations.
• Previous work by Lin et al. (2002) and Stevens et al. (2017, 2018) established a foundation for coverage probability/probability of agreement.
• Strong parametric assumptions about latent trait distribution can hinder the accuracy of agreement assessments.

Purpose of the Study:

• To extend existing methodology for coverage probability/probability of agreement by relaxing parametric assumptions.
• To develop inference methods for computing pointwise and simultaneous confidence bands for agreement.
• To provide a more direct and robust assessment of measurement agreement compared to existing methods.

Main Methods:

• Relaxation of strong parametric assumptions regarding the distribution of the latent trait.
• Development of inference methods for calculating pointwise and simultaneous confidence bands.
• Accommodation of heteroscedastic measurement errors in the methodology.

Main Results:

• The enhanced methodology performs well even with limited repeated measurements (one method with one measurement, the other with at least five).
• The approach successfully computes both pointwise and simultaneous confidence bands.
• The method circumvents deficiencies associated with the Bland & Altman limits of agreement.

Conclusions:

• The extended methodology offers a more flexible and accurate approach to assessing agreement between measurement methods.
• This approach provides a more direct assessment of agreement levels, overcoming limitations of prior techniques.
• The developed methods are valuable for studies involving repeated measurements and heteroscedastic errors.