4
$\begingroup$

We are running an observational clinical study on sedation during painful procedures and planning a future randomized trial. The primary outcome variable is a 6-level ordinal scale characterizing adequacy of sedation. 0-1 represent over-sedation, 2 optimal sedation and 3-5 under-sedation. Here is the original article describing the scale

While this scale reflects clinical reality very well, having the optimal outcome in the middle of the scale is statistically challenging. There are only a handful of trials that are using this scale as a primary outcome and they all dichotomize the score into Optimal vs Non-optimal which entails a massive loss of information. I am looking for approaches to transform or analyze this scale that maintain as much power as possible.

One simple approach would be to use "distance from optimal", ie |2-score|. This could also be weighed to reflect clinicians' preferences towards under- vs over-sedation. This could then be analyzed using ordinal logistic regression.

I have also seen some papers describing something called an isometric log ratio (ilr) transformation that can be used for bipolar scales in psychology research, but I am unsure whether that would be applicable.

Do you guys have any tips or ideas?

Improve this question
$\endgroup$
14
  • 5
    $\begingroup$ Welcome. I see no reason to do any transformation. In fact, many clinical scales don't have the optimal value in the middle and end up with ceiling or floor effects. This is especially true if you're planning an RCT where sedation scores will be compared across treatments. $\endgroup$ Commented Aug 12 at 13:46
  • 3
    $\begingroup$ I agree with @RickHass. If you are in one of the p value fetishization fields, you can look at whether outcomes are significantly different from 2. $\endgroup$ Commented Aug 12 at 13:52
  • 3
    $\begingroup$ Just adding to the points already made, taking ]2 - optimal] a) Implies that the scale is interval level and b) throws away information on the direction of the problem. There may be cases where the direction is irrelevant, but I think that they are rare. $\endgroup$ Commented Aug 12 at 14:17
  • 3
    $\begingroup$ When you come to run your randomised trial presumably you will have a scientific hypothesis in mind. That hypothesis will presumably affect a suitable way of looking at the data. Do you currently have such a hypothesis? $\endgroup$ Commented Aug 12 at 15:17
  • 3
    $\begingroup$ What is your goal with this variable? You assert that "having the optimal outcome in the middle of the scale is statistically challenging", but you don't mention why; I don't see any reason why this should be the case by default, or at least it's non-obvious. There are all sorts of analyses that can deal with symmetric deviations from a central tendency; standard two-sided tests of location being perhaps the most trivial example. $\endgroup$ Commented Aug 12 at 15:23

0

Reset to default

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.