I am analysing repeated measurement of ratio-scaled data. For doing that I rely on Bland Altman analysis. The common case where Limit of Agreement is horizontal can be adressed by most of the software, but how to compute LoA when random error vary with mean data value? I am looking for analytical solution of free software package.
Maybe the R-package "MethComp" by Bendix Carstensen might be of interest to you. His approach puts the standard (Bland-Altman) procedure in a proper modeling framework. Further, more elaborate designs of method comparison studies are covered, in particular: studies with replicate measurements by each method, studies with more than 2 methods, dealing with non-constant bias (scaling between the methods), and even non-linear relationships between methods, etc.
If the data do not exhibit a linear relationship between methods, and/or not a constant variation across the range of measurements, a transformation might be helpful. Then do Bland-Altman analysis on the transformed scale.
----- For further details, e.g.:
http://bendixcarstensen.com/MethComp/ (=> see the course notes)
B. Carstensen. Comparing Clinical Measurement Methods: A practical guide.
Wiley, 2010.
Carstensen B. Comparing methods of measurement: Extending the LoA by regression.Stat Med. 2010 Feb 10;29(3):401-10.
http://bendixcarstensen.com/MethComp/
Thank you Christine. Actually, as you said, MethComp offers many advanced tools for regular (or almost) Bland Altman analysis. That's a good idea to transform the data instead, I didn't think about this approach. Thanks again
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Software