How to regress and predict in a Bland and Altman plot? Review and contribution based on tolerance intervals andcorrelated errors in variables models
(2015) , 38 pages
Files
ISBADP2015_15_govaerts_HowtoregressandpredictinaBlandandAltmanplot_web.pdf
Restricted Access - Adobe PDF
- 799.33 KB
Details
- Authors
Govaerts, Bernadette 
- Abstract
- To assess equivalence in method comparison studies, two main methodologies are presented separately in the literature. First, the well-known and widely applied Bland and Altman approach with its agreement intervals where two devices are considered interchangeable if their differences are not meaningful in practice. The second approach is based on errors-in-variables regressions in a classical (X,Y) plot and focuses on confidence intervals. Two devices are considered equivalent when providing similar measures notwithstanding the random measurement errors. This paper reconciles these two methodologies, shows their similarities and complementarity with real data and simulations. A new consistent Correlated-Errors-in-Variables (CEIV) regression is introduced to compare these two approaches. Indeed, the errors are shown to be correlated in a Bland and Altman plot. When ignoring this correlation, the coverage probabilites collapse drastically and the biases soar considerably. Novel tolerance intervals are compared to agreement intervals with or without replicated data as well as novel predictive intervals to predict single measure in a (X,Y) plot or in a Bland and Atman plot with a robust estimator of the measurement differences. It will be then concluded that (C)EIV regressions can unfortunately not be avoided in method comparison studies, although the Bland and Altman approach is, usually, applied to avert the complexity of this statistical method. Tolerance or predictive intervals are presented in this paper as better alternatives than agreement intervals. Tips for the user are provided.
- Affiliations
Citations
Francq, B. G., & Govaerts, B. (2015). How to regress and predict in a Bland and Altman plot? Review and contribution based on tolerance intervals andcorrelated errors in variables models (ISBA Discussion Paper 2015/15). https://hdl.handle.net/2078.5/190236