[Maryam Bidgoli]

Thank you for the explanation. Again I am no expert on this type of method, so I will defer to you, Mike, and Paul.

I only mentioned matching techniques because I am somewhat familiar with these, and I thought certain designs might be applicable to your study. For instance, there is a paper by Kautonen et al. (2017) that examines whether job switching has an effect on the quality of life of entrepreneurs. They use propensity score matching to compare three groups: (1) Switching to Entrepreneurship vs. Staying in the Same Job, (2) Switching to a New Job vs. Staying in the Same Job, and (3) Switching to Entrepreneurship vs. Switching to a New Job. Their idea is that job switchers might have some unobserved characteristic that would confound any estimates of the effect of switching to entrepreneurship. But, if you compare switching to entrepreneurship vs. switching to a new job, you are more likely to eliminate this unobserved trait.

I'm not sure if this is useful to you, but when I saw your problem I thought you might be able to design a similar analysis. You could compare (1) those who had a mastectomy to those who did not, (2) those who had a breast reconstruction to those who did not, and (3) those who had a breast reconstruction against those who only had a mastectomy.

This may or may not be helpful to you, but I thought I would share.

Kautonen, T., Kibler, E., & Minniti, M. (2017). Late-career entrepreneurship, income and quality of life. Journal of Business Venturing, 32(3), 318-333.

Chris Boudreaux, Thank you so much. Very relevant and helpful. I read the

Kautonen, T., Kibler, E., & Minniti, M. (2017). Late-career entrepreneurship, income and quality of life. Journal of Business Venturing, 32(3), 318-333

, I can do the similar study. Thank you for sharing.

[Maryam Bidgoli]

Yes, you definitely can use -clogit-. The conditional logit model, from a practical point of view, applies whenever you want to control for some variable but do not need to estimate the effect of that variable. In a "panel" setting, that variable would commonly be the ID of the individual, with multiple observations having the same value of the ID variable. Here, the matching variable defines sets of individuals that share the same set of background characteristics. You want to control for those factors but not estimate their effects. Conditional logit thus enables a finely stratified analysis, but without estimating the effect of the stratifying variable. Note that the -xtlogit- command, designed for panels, includes -clogit- as a special case (fixed effects), and you could also use it.

Thank you so much Mike Lacy. Two things:

1. I looked at the -clogit- example 2 Stata Example (same as the one you shared here). That example presents data on matched pairs of infants, each pair having one with low birthweight and another with regular birthweight. What I'm saying is that, in that example, we have the pair matched for the outcome (low). However, I want to see the impact of either mastectomy or breast reconstruction on depression, and I paired the cases and controls based on the independent variables (mastectomy or breast reconstruction). My question is that "does it matter to pair cases and controls using any of the independent variables or it has to be the outcome variable (in my study, depression)?

2. In the study of the impact of mastectomy vs. breast reconstruction (two different observations), does it make sense to generate a binary variable that "1=those who had mastectomy and not breast construction, and 0=those who had breast reconstruction and not mastectomy?

Many thanks!

[Mike Lacy]

Maryam, regarding 1), no, I'd distinctly disagree about your description of that example. The pairs in that study *differ* on outcome (low birthweight vs. regular birthweight), but are *matched* on age of the mother. In -clogit-, pairs (or more generally, matched sets) are not selected to be matched on the outcome. To the extent that matched sets *are* the same on the outcome, they lack variation in the outcome variable, and can't contribute to the analysis. Again, your study is distinctly not a case-control study. as you are not sampling individuals differentially based on a binary outcome, so I'd avoid that terminology, as I think it is still causing you problems.

Here's how I'd think of your study: You want to compare subjects who received different treatments. One way to help control for various confounders in analyzing the effect of treatment is to select subjects from each of those two treatment groups so as to create sets of subjects who are the same on the confounders, but who differ on treatment, and then use -clogit- to analytically adjust for the matching while analyzing a binary outcome (depression or not).

Note that matching with -clogit- may or may not be the best way to improve the precision and reduce the bias of your estimateof the effect of treatment differences on depression. And the odds ratio (as opposed to a risk difference) is probably not the best effect measure. Modern treatment effects methods, as embodied in Stata's -teffects- commands, are probably better than -clogit- for this purpose. I'd encourage you to consult someone local epi or biostat person who can help you work through that choice in the context of your particular data set. Note in this regard that because you do *not* have an outcome-selective study design, you are not limited to using odds-based statistical methods as you would have been had you really had a case-control design. I may have misdirected you toward -clogit-, which is a possibility but not necessarily the best choice.

Regarding 2: Perhaps I'm misunderstanding something, but I would presume that you would *have* to have a treatment variable such as you describe, which yes, you'd want to code numerically, and for which 0/1 is a reasonable coding choice

[Paul Dickman]

My question is that "does it matter to pair cases and controls using any of the independent variables or it has to be the outcome variable (in my study, depression)?

I'm struggling to understand your question. From what I understand this is not a case control study. If that's so then I advise avoiding the words case/control as they cause confusion. My understanding is that you are interested in studying the association between type of surgery (the exposure) and depression (the outcome). Standard terminology in epidemiology for cross-sectional and cohort studies is to refer to exposure and outcome. It appears that you are considering women who underwent mastectomy as exposed and the women who underwent reconstruction as unexposed.

It's not clear what you mean by "pair". Do you mean match? In your original post you mentioned that you wanted to "match on age, length of stay (LOS) in hospital and Carlson comorbidity index". Can you motivate why you wish to match? You have identified 9000 women who underwent mastectomy and 3000 women who underwent reconstruction. I don't see a reason to match. It would seem to me that unconditional logistic regression where you adjust for age, length of stay (LOS) in hospital and Charlson comorbidity index.

Having said that, I have doubts that your data are sufficiently strong to study if type of treatment is causally associated with depression. I think you have serious issues with

1. The temporal ordering of exposure and outcome. This is a cross-sectional study where depression was assessed concurrently with the treatment. It's not unlikely that women with depression were diagnosed before surgery.
2. Uncontrolled confounding / confounding by indication / reverse causation.
3. Appropriateness of the positivity assumption.