Can a cross-sectional study use incident cases?

Question

I am struggling with the epidemiological definition of my clinical study design.

Looking at the literature I take it that case-control studies are retrospective (i.e. you must look back on the exposure, simply having cases and controls does not allow you to call the study case-control) while in cross-sectional studies the exposure and outcome are simultaneously assessed.

I have 2 different groups of patients (dementia type A and dementia type B) and one group of controls (no dementia). The studies investigates the association between disease status and some cardiological, neuropsychological and imaging measures, which are taken at enrolment. So my study must be cross-sectional. However, I am using incident cases of dementia (i.e new diagnoses) because we wanted the diagnosis to be homogeneous and thus made at our Centre. Can I call it a cross-sectional study with incident cases and controls?

Can someone help? Thanks!

---

thank you very much for your detailed reply. It took me a while to get back to you because I had to do a little more research to understand the points you make and see where my problem is. I think I may now have focused it:

1) I am fine with the fact that in cohort studies study groups are defined by exposure and that data collection can be either prospective or retrospective

2) However, in most of the literature I looked up, case-control studies have two characteristics: study groups are selected based on the outcome and data collection is retrospective (e.g doi: 10.1097%2FPRS.0b013e3181f44abc; doi: 10.1089%2Fjpm.2008.9690). In fact for some authors the retrospective nature of the study is fundamental to define it as a case control study and they object that no such thing as a cross-sectional case control study can ever exist (e.g. doi.org/10.1161/CIRCGENETICS.111.959783; by the way this was actually the term I was thinking I would use for my study...); they also maintain that because of its retrospective nature, a case-control study, unlike a cross-sectional study, always allows for (some) causal inference.

3) I do not understand that cross sectional studies are defined based on the outcome. It seems to me that they are simply defined by the simultaneous assessment of both outcome and exposure, precluding any causal inference (e.g. doi:10.1089%2Fjpm.2008.9690, doi:10.4103/0019-5154.182410). However, I am perfectly clear and absolutely agree with what you say about a cross-sectional study with incident (instead of prevalent) cases: it will have the limitation of cross-sectional assessment (direction of causality cannot be determined) but not the bias related to survival.

4) Back to my specific study, I have the impression I am selecting cases and controls based on the outcome (incident dementia or not) and assessing exposure at the same time. So what I call my study appears to depend on the epidemiological definition I use . If case control studies just require selection based on outcome then my study would qualify as a cross sectional case-control study ( which I understand would be epidemiological nonsense..). If, instead, case control studies require both selection based on outcome and retrospective data collection I can only call it a cross-sectional study (data assessment is practically simultaneous).

I would really appreciate if you could follow-up on your response and apologize beforehand if I have made mistakes (PS: as you may have noticed, I am a clinician not an epidemiologist...). Also, I didn't know how post a detailed reply to your comments other than this way, if you or anyone else feels I should instead edit the original post, I will be happy to do so.

Paola

Answer 1

I would be careful with the terminology "case" and "controls", and avoid it in your described situation. Case-control studies are much more than retrospective studies as described in your definition. The defining feature of a case-control study is that the two comparison groups are defined in terms of their outcome, rather than the treatment/exposure (e.g. cross-sectional, cohort). Depending on how the case-control selection strategy is designed, you select cases and corresponding controls defined by their disease status at some time point. Case-control studies are also not required to be retrospective (see case-cohort studies as a counterexample).

Your study is a cross-sectional study, not a case-control study. The groups you are comparing are defined by their exposure (incident dementia) and not the outcome (cardiological, neuropsychological and imaging measures). The fact that you are using incident dementia cases does not change the study type, but is only part of the exposure definition (it is also a good exposure definition because it minimizes survivor bias compared to incident and prevalent dementia cases). Rather than using any terminology like "case" and "controls", I would explicitly write-out the 3 groups being compared (type A dementia, type B dementia, no dementia).

Update:

The following is a brief update to the clarifications provided.

2) I would generally agree that no such thing as a cross-sectional case-control design can occur. If you are taking a "snapshot" of the population at one particular moment in time, then that would be considered a cross-sectional study. However, if you are deliberately selecting cases of dementia then selecting controls to represent the population that would constitute a case-control study. The exposure would precede the outcome.

3) Cross-sectional are not necessarily defined by the outcome because they are a single time-point snapshot of some population. However, a case-control study is explicitly defined in terms of the outcome (the definition of "cases" and "controls" depends on the outcome).

4) It may help if you clarify the question of interest. Are you saying that dementia causes changes in cardiological measures or that cardiological measures lead to dementia? This may help to clarify (at least for me). Regardless, since your data is all measured and corresponds to a single moment in time (i.e. cardiological measures taken at enrollment when dementia is assessed) I would recommend referring to your study as cross-sectional. Similarly, I would avoid the use of case/controls and instead refer to as dementia diagnosis/ no dementia diagnosis.

Hope these clarifications help!

Answer 2

Your cases are not incident, that is not correct. If cases were incident, you could estimate an incidence rate. That is precisely what's missing in a case control study. However, you did not sample them prior to their onset of dementia. The time since diagnosis is an important feature of the cohort and the study in general, and the fact that it is generally short is a critical aspect for consideration by the general audience: what biases may be present? to whom might these results generalize? (outpatient care? patients in hospice? etc).

The point of a case-control study is precisely that you sample patients based on their disease status and retrospectively evaluate exposure. "Retrospective" simply means the ordering of data collection, not the events per se. They may be coincident such as helmet-wearing status of fatal/non-fatal motorcycle collisions reported by the state transportation department.

We assume that studies performed in the clinical setting identify patients not only with disease, but a unique presentation thereof. That's why we are careful to select a control group so as to avoid Berkson's bias. In fact, if the non-dementia patients presented at your clinic, presumably one which performs neurologic evaluation, I might be highly concerned that this design needs further evaluation/correction to avoid bias.