To tolerate or to agree: A tutorial on tolerance intervals in method comparison studies with BivRegBLS R Package

Area of Science:

• Biostatistics
• Clinical Research Methodology

Background:

• The Bland-Altman agreement interval is widely used in method comparison studies to assess clinical measurement interchangeability.
• This interval is known to be approximate and potentially too narrow, leading to concerns about its accuracy.

Purpose of the Study:

• To demonstrate the limitations of confidence intervals applied to Bland-Altman bounds.
• To introduce and advocate for the use of exact tolerance intervals as a superior alternative.

Main Methods:

• Critique of the confidence interval approach for Bland-Altman bounds.
• Application and illustration of tolerance intervals using real data.
• Utilized the R package BivRegBLS for calculations under normal or log-normal distributions.
• Assessed tolerance interval coverage probabilities via simulations.

Main Results:

• The approach of adding confidence intervals to Bland-Altman bounds is shown to be misleading and confusing.
• Tolerance intervals provide exact calculations and are easier to interpret.
• Demonstrated practical application and simulation-based validation of tolerance intervals.

Conclusions:

• Tolerance intervals are a more accurate, interpretable, and practical method for assessing agreement in method comparison studies.
• The use of tolerance intervals is recommended over approximate methods like the modified Bland-Altman approach.