Join Dr. Ana Diez-Roux, for SERforum Live!
May 1, 12pm EST


Do you ever find yourself struggling to figure out a question about epidemiologic methods, or other topics in epidemiology, and don’t know who to ask? The SERforum allows for individuals to answer questions that come up in our daily work around substantive and methodological topics in epidemiology.
All topics may be viewed, but to read and post comments, SER membership is required. If you are member, login! Not a member, join us!

You need to log in to create posts and topics.

Classification of impure retrospective case series

Dear colleagues,

I am a veterinarian and epidemiologist who works full time as an editor at 2 veterinary journals. Part of my role is to ensure study designs have been appropriately classified by authors in the abstracts and main text. However, there is one situation I and the other editors (only a third of whom have an epidemiology background) find challenging and I’m hoping to get some guidance on so I and the others can feel more confident in our decisions.

Approximately a quarter of all submissions to one journal are classified by authors as “retrospective case series.” However, the following are what those studies could actually involve:

1)      1) Purely descriptive summary of the characteristics, diagnostic testing, treatments, and outcomes of all qualifying cases encountered over a particular period (which I will refer to as “case series,” per my understanding of the definition).

2)      2) Case series, followed by statistical comparisons (ranging from chi square tests to multivariate Cox proportional hazards analysis) of outcome (ranging from “survival to hospital discharge” to “overall survival time” or “time to relapse”) between certain exposure groups within the cases.

3)    3) Case series, followed by statistical comparisons between characteristics-based groupings (eg, male dogs were more likely to have severe clinical signs than female dogs, or Boxers were more likely to have nodules detected on radiography than other breeds) and outcome-based groupings (eg, male dogs were more likely to die before discharge than female dogs). Follow-up after discharge may or may not have been performed.

4)     4) Other variations on observational- (analytic-) style study designs “nested” within a case series. Follow-up after discharge may or may not have been performed.

5) Cross-sectional-type studies, with included subjects selected on the basis of their having a certain disease or treatment (no regard to other factors) and comparisons made of existing medical record data (no follow-up) between certain groups (eg, Boxers treated for tracheal collapse were more likely to have tachypnea than other breeds treated for tracheal collapse, and males treated for tracheal collapse were more likely to also have radiographically detected cardiomegaly than females treated for tracheal collapse). (I think this study design is self-evident [ie, cross-sectional], but please correct me if I'm wrong.)

I realize that the validity of some of these “within case series” comparisons is questionable; however, this is what we get and, unfortunately, the veterinary medical field has few epidemiology experts that we can call upon to review these things. So almost all of them make it through peer review, and the editors are left to clean them up and ensure the authors discuss the limitations.

My question is this: how would you generally approach classifying the above scenarios? My approach has been to classify them on the basis of the stated aims (eg, The aim of the study was to describe dogs treated for tracheal collapse with regard to clinical signs, treatment, and outcome and determine whether certain characteristics were associated with survival to hospital discharge), but I can get lost even then because the subjects weren’t initially selected on any factor other than their case status, and classification of the analytic component is rarely clear cut.

Any guidance you might provide in this regard would be greatly appreciated!! Thank you!

My thought is that these are "descriptive studies". It seems appropriate to assess differences within a case series if "cases" are selected in a systematic way that represent most if not all who have the condition of interest.  Certainly describing them based on their stated aims is fine as well. My 2 cents....Bernie Harlow

You are correct to consider how study subjects were selected into the study.  These scenarios are all case series and descriptive in nature.  However, the purpose for classifying them is not clear.  Would it suffice to classify them all as case-series, which you already have done and which appears to be true?  Is the purpose to guide evaluation of analyses? If there is a comparison being made, an evaluation of the precision of the contrast (e.g., risk difference with confidence interval) should be done.  If purely descriptive, then presenting numbers and percent values should suffice.  Your assessments for the scenarios seem correct to me. Number 2 could be considered a cohort study that compares outcomes between exposure groups among a cohort of those affected with the case disease.  Number 3 could be considered cross-sectional because a comparison is being made. 


Thank you both for your responses! The reason we editors are concerned about classifying the studies is because we have a section in our structured abstract for "Study design," and given a fairly new focus in veterinary medicine on the hierarchy of evidence, we want to help readers understand where each study might fall in that regard and be as accurate as possible in the labeling. If all of the above scenarios were of no more value in that hierarchy than retrospective case series, then we wouldn't concern ourselves. However, I'm not convinced of that, and one of my admitted pet peeves as an epidemiologist is the epidemic of misclassified studies in the scientific literature in general. Author summaries are notoriously unreliable.

I agree that, for the most part, these are all descriptive studies. Where I struggle is the statistical comparisons aimed at identifying risk or protective factors, which as I was trained in graduate school (see Veterinary Epidemiologic Research by Doohoo, Martin, and Stryhn) add an analytic component (or, as Rothman, Greenland, and Lash would refer to it in Modern Epidemiology, a "nonexperimental epidemiologic" component). To me, if you're performing multivariate Cox proportional hazards analysis to identify factors associated with an outcome, that's not descriptive. Martha, your responses are quite helpful in that regard.

Thanks again!

Sandi Lefebvre (sorry, I forgot to post my personal details before)

Deputy Editor Journal of the American Veterinary Medical Association and American Journal of Veterinary Research