What is a case–control study? Comment on ‘Oral midodrine does not expedite liberation from protracted vasopressor infusions: A case–control study’

I was interested to read the recent report from Wood and colleagues, ¹ who described a retrospective observational study of patients receiving intravenous vasopressor support in an intensive care or high dependency unit. Their primary aim was to determine whether or not midodrine facilitated more rapid weaning of vasopressor support. They compared a group who had received midodrine with a contemporaneous group who had not and found that there was no evidence that midodrine was associated with a reduction in duration or dose of vasopressors. Whilst the findings are supported by the data, I would like to correct their mistaken description of this as a case–control study.^2,3

Wood and colleagues are not alone in their misunderstanding of case–control studies. ⁴ For a case–control study, a ‘case’ is defined by the outcome of interest and a ‘control’ is defined by absence of such an outcome. Selection is not based on exposure status. Ideally, both cases and controls should come from a similar population of interest (in this case, patients receiving intravenous vasopressor support). The researcher(s) then looks backward in time to identify potential exposures or risk factors associated with that outcome. The strength of any associations is typically reported as an odds ratio. For Wood and colleagues, this design would necessitate defining, for example, cases as those requiring protracted vasopressor support versus control patients who did not. In their circumstance such an option would not be ideal given the likely loss of statistical power – case–control studies are best reserved for studies of uncommon/rare outcomes.

Wood and colleagues report a cohort study. A cohort study begins by first defining the groups according to one or more exposures of interest (in this case, it was treatment with midodrine) and then looks forward in time to measure outcome – that is, selection is based on exposure status. Unlike a case–control study, incidence can be estimated.

All observational studies are exposed to bias and confounding, and these can be mitigated by matching and/or multivariate adjustment. However, case–control studies are particularly exposed to ascertainment bias and this structural deficiency can be difficult to detect so must be dealt with in the design phase. ²