variation between subjects vs within subjects

Hello all. This is my first post.

I am interested in making a comparison between outcomes of treatment A, and standard of care, treatment B.

I have 70 cases of treatment A, and 400 cases of treatment B. However, there are only 40 subjects who received treatment A (some received multiple treatments occurring at different times), and about 200 Subjects who received treatment B. Maybe 25 subjects received at least one of each.

This is a retrospective review.

How should I compare the outcomes?
Record only subjects with self comparison?
Record all treatment A and "match" to treatment B. And, if done this way can I count each treatment as a separate "subject"?


TS Contributor
What are these outcomes? When were outcomes measured? Are there baseline measurements?
Maybe you could describe the actual context and the actual study in some more detail?

With kind regards

These are patients undergoing a surgical procedure. Outcomes are pain score average for 24 hours and opioid consumption within this period.
Basically, we are using a new method of pain control for management of postsurgical pain. The procedures are consistent from person to person and within subjects. Some subjects have a historical control when they were given the old pain control treatment. Many have several data points with both new and old pain control methods. Some only have the new method of pain control, and quite a lot have only the old method.

Outcomes are recorded as follows: pain score: 0-10 recorded often for first 24 hours. Will average these scores for each 24-hour post procedure window. The next outcome measure is milligrams of morphine equivalent taken within that 24 hour period. Only interested in the 24 hour period.


TS Contributor
This sounds like some multilevel model, where treatement episodes are nested within subjects.
Predictors could be type of treatment for each episode, and the number of the treatement
episode (within a patient).

With kind regards

I previously used a mixed model analysis in SPSS where I had 8 patients with 1 to 3 treatments with each treatment type.
Could I perform the same set-up with the new data, knowing that some patients only had exposure to one treatment type?

To be honest this was pretty uncomfortable for me because there are a lot of different models making different assumptions that I'm unsure how to deal with.

Does anyone have experience using mixed method analysis for this type of project?