I have two groups of patients, responders (~75%) and non-responders (~25%) to a treatment. I want to look for a difference between them in a single biomarker (which I have not yet measured). The biomarker is non-parametric, although it has not been measured in this disease group before.
I am aiming to look at this in two different ways:
(1) Taking a single measurement from patients currently on treatment (mix of responders & non-responders). From a previous study of the same biomarker in a different disease group, I can estimate the median values for each group. I hypothesize that the biomarker is higher in responders. I would then perform a Mann Whitney test on my unpaired, non-parametric data.
(2) Prospectively sampling a group of treatment-naive subjects pre- and post-treatment. I hypothesize that responders have higher baseline levels of the biomarker and a greater reduction with treatment than non-responders. I would then perform a Wilcoxon test on my paired, non-parametric data.
How do I calculate the sample size for (1) and (2)?
Only tests are nonparametric not anything else. In your proposal, the best you can do is find a correlation. I just finished a similar study. I am attaching a copy in case you would find it helpful. Best, D. Booth
Further to David Eugene Booth's comment that only tests are nonparametric, I quite like this comment from the Vassar Stats website:
- http://vassarstats.net/textbook/parametric.html
See also this nice paper by Ronán Michael Conroy:
- https://www.stata-journal.com/article.html?article=st0253
Among other things, Dr. Conroy shows that the Mann-Whitney U test for two independent groups (aka., the Wilcoxon rank sum test) does indeed estimate a parameter (i.e., the proportion of Xs that are less than Ys).
Back to your biomarker: What descriptive statistics do people typically use for location and dispersion? Also, what are your sample sizes? Thanks for clarifying.
HTH.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size