Keywords: Tolerance Intervals, Method Comparison Studies, Agreement, Errors-in-Variables Regression, Bivariate Least Square
Webpages:
https://CRAN.R-project.org/package=BivRegBLS The need of laboratories to quickly assess the quality of samples leads to the development of new measurement methods. These methods should lead to results comparable with those obtained by a standard method.
Two main methodologies are presented in the literature. The first one is the Bland-Altman approach with its agreement intervals (AIs) in a (M=(X+Y)/2,D=Y-X) space, where two methods (X and Y) are interchangeable if their differences are not clinically significant. The second approach is based on errors-in-variables regression in a classical (X,Y) plot, whereby two methods are considered equivalent when providing similar measures notwithstanding the random measurement errors. These methodologies can be used in many other domains than clinical.
During this talk, novel tolerance intervals (TIs) (based on unreplicated or replicated designs) will be shown to be better than AIs as TIs are easier to calculate, easier to interpret, and are robust to outliers. Furthermore, it has been shown recently that the errors are correlated in the Bland-Altman plot. The coverage probabilities collapse drastically and the biases soar when this correlation is ignored. A novel consistent regression, CBLS (Correlated Bivariate Least Square), is then introduced. Novel predictive intervals in the (X,Y) plot and in the (M,D) plot are also presented with excellent coverage probabilities.
Guidelines for practitioners will be discussed and illustrated with the new and promising
R package
BivRegBLS. It will be explained how to model and plot the data in the (X,Y) space with the BLS regression (Bivariate Least Square) or in the (M,D) space with the CBLS regression by using
BivRegBLS. The main functions will be explored with an emphasis on the output and how to plot the results.
References BG Francq, B Govaerts (2016). How to regress and predict in a Bland-Altman plot? Review and contribution based on tolerance intervals and correlated-errors-in-variables models. Statistics in Medicine, 35:2328-2358.
