“Average” plots [NCA / SHAM]
❝ So which kind of average plot would allow a significant data reduction (one line per treatment) instead of one for every subject? Ideally it should convey unbiased information.
I am not using these plots for anything, and I do not know of anyone taken decisions on basis of them regardless of whether they involve means, medians, geomeans, with or without imputation, LLOQ/2 etc . For all I care, one flavour is as good as another and is of little importance.
I like simplifications, but reducing human PK profiles from e.g. 56 subjects to a single one in the hope that the result is particularly informative and/or allows decision-making is to me too ambitious.
Spaghetti plots can get confusing but all info is (potentially) in them. I don't like them particularly either.
One thing I'd like to try out, but I can't conjure up the code now:
Create a 95% (or 90% etc) shaded cloud around each mean or median at all time points. Now this results in two average/median/whatever profiles surrounded by its cloud, and the clouds will most often be overlapping (color shades can overlap in R). These sort of plots are sometimes used in ecological sciences and they have a fancy name over there. Like MacMillian-Hemorrhoid Gymnoplasticity Wexes or some such improbable term. My memory might not be entirely accurate here as to the nomenclature.
To me, this might be an improvement, but I can't say I'd use it for anything concrete until I get familiar with it.
Pass or fail!
- “Average” plots Helmut 2019-04-25 19:06 [NCA / SHAM]