pre-post analysis in two groups with missing post in control group

Question

A bit background of the study:

Case group: tumor patients underwent a treatment Control group: a healthy subjects with matched gender, age and some other variables

Target parameter: a continuous variable x, was measured both before and after the treatment in cases but ONLY once in conrols.

x is normally a lot higher in tumor patients than in healthies, and the treatment might bring it down, but still higher than normal.

So my purpose is to have a pre-post comparison of x within the tumor patients, also between the two groups...

I see the data as a pro-post with a missing post in controls.

Who he has experience in dealing with such data please help me with some clues... Thank you very much in advance

Answer 1

The comparison of pre-treatment tumor patients and control patients doesn't seem to make much sense in this case. I suspect it would have been unethical to provide the treatment to the controls, so there was no way to test the effect of treatment on x in controls.

You certainly can examine whether the treatment lowered x among the tumor patients, and then see whether post-treatment x in the cancer patients differs from that of the control patients.

Since you have some covariates which you tried to match (but were unlikely to have matched exactly) you might consider including them as additional explanatory variables, both in the post-treatment cancer versus control comparison and in the pre- versus post-treatment cancer comparison.

Answer 2

I don't think a case-control study is what you are looking for. A case-control study is by definition retrospective: you do it when you want to know what factors increase or decrease the likelihood being a "case" vs. a control. Here, your study appears to be prospective. That said, I don't think it made sense to administer treatment to healthy subjects (your control group) without knowing much more about what the treatment was supposed to do. If, as you say, $X$ is higher in tumour patients, then why have the need for comparisons with healthy patients? Is the treatment supposed to affect $X$ across the board regardless of whether one is a tumour patient? What is the hypothesized relationship between the treatment and $X$ in healthy patients, and again: why the need to study how $X$ changes in healthy patients? Is there a level of $X$ below which the person's health would be at risk?

If this study was aimed at evaluating the effect of a novel treatment on $X$--and I think this is what you are trying to do--then the ideal design would be to have tumour patients randomised into treatment or control (placebo), then taking baseline and endline measures for all study subjects. The control group should have been other tumour patients, not healthy patients.

Since you do not have a control group, one way you can still evaluate how the treatment affects $X$ is by fitting a fixed effects model just on the "cases" group (since we don't have post-data on the "control" group, it would have been hard to do the standard comparison, e.g. difference-in-differences, anyway).

Note that with fixed effects models, anything that doesn't change between time1 and time2 are differenced out of the model (excluded from the model), so be mindful of that when you build your model.