When examining a possilbe confounding factor (Z) of the relationship of a 3-group categorical variable (X) (high-intermediate-low) with survival/relapse (Y) using the three classical criteria (KJ. Rothman et al., Modern Epidemiology p.260), which of the three groups would you take as the "unexposed" one for the first criterion assessment (a confounding factor must be a risk factor among the unexposed): just the Low (or High) group or better include also the intermediate category?
There is no real rule to this situation. You should try, first, the "high" or "low" group and then adjust it according to the results (if you do not have any clue in advance). This adjustment is just to put, in reference level, all categories with the higher (lower) risk. It will makes your interpretation easier.
But categories with very limited sample size should be avoided as reference level, because they will compromise your result.
I asked a similar question some time ago and can be helpfull to you:
https://www.researchgate.net/post/Reference_level_in_a_multiple_regression
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis