Share this article on:

The Limits of Case-Control Studies

Stang, Andreas; Jöckel, Karl-Heinz

doi: 10.1097/01. ede.0000239726.53102.79
Letters to the Editor

Clinical Epidemiology Unit, Institute of Medical Epidemiology, Biometry and Informatics, Martin-Luther-University of Halle-Wittenberg, Halle, Germany, (Stang)

Institute for Medical Informatics, Biometry and Epidemiology, University of Duisburg-Essen, University Hospital of Essen, Essen, Germany (Jöckel)

Back to Top | Article Outline

To the Editor:

We read with interest the recent commentaries by Hartge1 and Bernstein2 and would like to comment on them. First, sensitivity analyses can sometimes show that the potential biases from nonresponse are small. However, using reasonable assumptions, sensitivity analyses can also show much more uncertainty of effect measure estimation than anticipated. In these situations, the nonresponse issue has not been solved.3 We therefore believe that for reasons of validity, epidemiologists should give consideration to alternatives to population-based case-control studies, for example, nested case-control or case-cohort studies within prospective cohort studies with high baseline response. Another option (depending on the etiologic scenario) is the application of case-only and case-crossover studies.4 Researchers could also restrict the study base to population segments with anticipated higher response proportions such as people with identifiable telephone numbers.5

Second, with a word count of 3000 words in many journals, authors are not able to present detailed results of sensitivity analyses that address potential nonresponse bias. Detailed response results and analyses that give clues to selection effects could be published in separate methodologically oriented papers.5 However, which high-ranking journal might be interested in these papers? An alternative could be an Internet-based report that accompanies an analytic paper on an epidemiologic study. (Editor's Note: Epidemiology is one of a growing number of journals that can publish such supplemental material online.)

Third, editors and reviewers of epidemiology papers are too often dogmatic in their judgment of the potential for nonresponse bias in population-based studies. One cannot logically deduce the likelihood of selection bias solely from the response proportions.6 For example, in an article describing our population-based testicular cancer case-control study,7 we reported a response proportion of 76% among cases and 46% among controls. We presented a sensitivity analyses that gave insights into the potential of selection bias with assessment of the limits of the bias. Even so, one reviewer wrote: “The participation rates are so low that the study may be judged to be noninformative.”

Fourth, what do the low response proportions in the general population actually mean? They suggest a lack of interest or even distrust among the general population or at least segments of the population. That brings up the necessity of studying the reasons for refusals in epidemiologic studies. The results of such investigations may help to start a public discussion regarding the importance of participation in epidemiologic studies.

Andreas Stang

Clinical Epidemiology Unit

Institute of Medical Epidemiology, Biometry and Informatics

Martin-Luther-University of Halle-Wittenberg

Halle, Germany

Karl-Heinz Jöckel

Institute for Medical Informatics, Biometry and Epidemiology

University of Duisburg-Essen

University Hospital of Essen

Essen, Germany

Back to Top | Article Outline


1. Hartge P. Participation in population studies. Epidemiology. 2006;17:252–254.
2. Bernstein L. Control recruitment in population-based case-control studies. Epidemiology. 2006;17:255–257.
3. Stang A. Nonresponse research—an underdeveloped field in epidemiology [Editorial]. Eur J Epidemiol. 2003;18:929–931.
4. Stang A, Jöckel KH. Appending epidemiological studies to conventional case-control studies (hybride case-control studies). Eur J Epidemiol. 2004;19:527–532.
5. Stang A, Moebus S, Dragano N, et al. Baseline recruitment and analyses of nonresponse of the Heinz Nixdorf Recall Study: identifiability of phone numbers as the major determinant of response. Eur J Epidemiol. 2005;20:489–496.
6. Stang A, Jöckel KH. Low response rate studies may be less biased than high response rate studies. Am J Epidemiol. 2004;159:204–210.
7. Stang A, Ahrens W, Bromen K, et al. Undescended testis and the risk of testicular cancer: importance of source and classification of exposure information. Int J Epidemiol. 2001;30:1050–1056.
© 2006 Lippincott Williams & Wilkins, Inc.