Raoul,
You bring up an excellent issue about how studies are named depending on how the data set was obtained. I share your observation that “case” and “control” is often used in other than the most stringent definitions in epidemiology and medical research. Reading your description also raises some other questions. You may (or may not…) have a case-cohort study, see my elaboration below.
In your case, the data including covariates and outcome information already exists and, as you point out, it is clearly a retrospective study. It is not entirely clear to me what you mean by “controls” that are matched to “cases” that are similar on important variables, though. All these studies are concerned about relating outcome to exposures in one way or another. When identifying those who developed the condition (“cases”) to those who did not (“controls”) one matches (or otherwise builds two groups) based on outcome. One can then investigate whether certain exposures (covariates) seem to be systematically different for cases as opposed to controls. On the other hand, one can match on “important covariates” and investigate whether a certain exposure to something of specific interest seems to be associated with the different outcomes of interest in otherwise comparable individuals. Example: You do procedure X in some patients that you see in your hospital for a special condition and want to see whether that procedure seems beneficial with respect to some outcome Y. Since this is observational data, it is likely that those who receive procedure X are systematically different from those who do not. Procedure X is the exposure of interest and you can find comparable patients by matching on the exposure (procedure X), often done by identifying all those who received procedure X and finding patients who did not but are similar with respect to age, race, gender, important preexisting conditions, time since onset of symptoms, whether transferred from some other hospital, insurance status, etc. (often propensity score matching is used but other techniques are available as well). One can then see whether exposure X is associated with outcome Y. Note that this is fundamentally different from matching on the outcome Y and that it does *not* control for outcome Y but for exposure X instead.
Your description could be read as implying that matching was both, on outcome Y *and* on important covariates X and that has the potential to destroy the information you are looking for. Consider this: You pick “cases” and “controls” and make sure that they have comparable traits like age, gender, etc. Now: What can you learn about the relationship of gender with outcome Y? Really not much, because you manufactured the data such that traits follow a specific distribution that you decided on.
Let’s assume that “cases” and “controls” where *not* also matched on covariates that are to be investigated in this study. There is the approach of case-cohort study that uses all cases and combines them with a sub-cohort of controls. A good motivation for doing so would be that covariate information is costly to obtain and therefore all controls cannot be assessed (e.g., additional lab-work using stored specimen/samples from the participants/patients). (If all covariates are known for all controls anyhow, one would need specific reasons why one would not adjust for differences in known covariates by other means like regression or inverse probability weighting based on propensity scores instead of disregarding some participants completely – namely those who are not matched, which is regularly are very large group.)
Reference to case-cohort studies:
Prentice RL. A case-cohort design for epidemiologic cohort studies and disease prevention trials. Biometrika. 1986;73(1):1-11. doi: 10.1093/biomet/73.1.1.
Kalbfleisch JD, Prentice RL. The Statistical Analysis of Failure Time Data. Second ed. Hoboken, New Jersey: Wiley; 2002.
Reference to epidemiological study design (overview):
Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. Third ed. Philadelphia: Lippincott Williams & Wilkins; 2008. (Chapters 6, 7, 8; note especially the remarks on p. 112 regarding “controls” – for a correct interpretation of what you are estimating you will have to pay attention to whether “controls” were “disease-free” but at risk at the beginning of the study period, whether they were “disease-free” at the point in time when the respective “case” occurred (e.g., did not have colorectal cancer by the case’s age 54, but possibly could developed the condition later as well), or whether they were required to be “disease-free” throughout the entire study period irrespective).
------------------------------
Fridtjof Thomas, Ph.D.
Associate Professor, Division of Biostatistics
College of Medicine/Department of Preventive Medicine
University of Tennessee Health Science Center
------------------------------
Original Message:
Sent: 02-15-2017 02:10
From: Raoul Burchette
Subject: proper name of a retrospective matched cohort study?
A common method used in clinical research involving retrospective data is a
retrospective cohort study with a "control" cohort matched to a "case"
cohort so the cohorts are "similar" on important (potentially confounding)
variables. This is often mislabeled as a "case-control study" in the
medical community, but it is not in the epidemiological sense. Does
someone know an accurate name for these kinds of studies that is true to
the epidemiology, and reflects the observational retrospective nature of
these artificially-matched cohort studies? Even a nice catchy name that
distinguishes it from a true case-control study would be nice.
Raoul J. Burchette, MA, MS
Biostatistician III
Biostatistics, Programming & Research Database Services
Kaiser Permanente, Regional Offices
Department of Research and Evaluation
100 S. Los Robles Ave., 3rd Floor (32W07)
Pasadena, CA 91101
(626) 564-3471 (message)
8-338-3471 (tie-line)
(626) 564-5855 (fax)
Raoul.J.Burchette@kp.org (email)
---------
kp.org/thrive
NOTICE TO RECIPIENT: If you are not the intended recipient of this e-mail,
you are prohibited from sharing, copying, or otherwise using or disclosing
its contents. If you have received this e-mail in error, please notify the
sender immediately by reply e-mail and permanently delete this e-mail and
any attachments without reading, forwarding or saving them. Thank you.