What is the type of my study, case control or retrospective cohort?

We performed a retrospective analysis of patients with disease X, who were underwent treatment A or B. The objective of the study was to determine the survival based on C or D (alive/death) and wanted to explore what confounders could have played a role in patients’ survival.

I received a suggesting saying that our cases are defined as patients whom successfully underwent treatment A, while controls were defined as patients whom did not receive A (i.e. B). But I see ours as a retrospective analysis of patients with disease X, who were exposed to treatment A or B, while our outcomes are C/D. I think it will be a good case control study if we have disease X and Y, then retrospectively see which had A or B

I would appreciate your input here

If I understand correctly, your data is collected conditional on having disease X. Your exposure variable can take values A or B, and the outcome variable can take values C or D.

I am assuming you are not sampling conditional on being dead (for example: obtaining data on everyone who died, and taking a random sample of those who survived). If that is correct, this is not a case control study. That is a good thing: A case control study is just a way to emulate a cohort study, it is a study design that can be used if it is too expensive to collect data on everyone. If data collection costs are not a factor, a retrospective cohort study will always be preferred.

If you can collect data on people who do not have disease X, I guess it is in principle possible to obtain a sample of data on people who have disease Y. This would be useful if you are interested in estimating whether treatment A/B causes disease X/Y. To me, this does not appear to be a meaningful research question, at least not without additional context.

Isn’t there a problem defining controls as those who did not receive A? Sounds like an immortal time bias?

@f2harrell: More context: A is combined treatment of surgery and chemotherapy, while B is chemotherapy alone. Surgery is being a controversial treatment in our field, so we thought to see what confounders previous studies have not looked at when they found surgery to be beneficial

@AndersHuitfeldt Thank you! your understanding is correct (referring to the first 2 lines). However, I don’t know why would I need to get samples for disease Y, when I want to see the effect of best management option (A/B) for X. I can’t give a treatment to a disease-free patient in my case

Start by asking yourself what the research question is. If your goal is to estimate the effect of treatment on mortality (among patients who have disease X), and if your dataset has not been constructed from a case-control sampling procedure, then it is not a case control study, and that is good news.

I also do not understand why you would want to consider Y in your analysis, I was only trying to respond to your suggestion that this might be relevant in some kind of case control setup.

1 Like


1 Like