American Journal of Epidemiology Advance Access originally published online on June 12, 2007
American Journal of Epidemiology 2007 166(2):128-129; doi:10.1093/aje/kwm138
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Østerdal et al. Respond to "Identifying Women with Hypertension during Pregnancy"
1 Maternal Nutrition Group, Danish Epidemiology Science Centre, Statens Serum Institut, Copenhagen, Denmark
2 Department of Obstetrics and Gynecology, H:S Hvidovre Hospital, University of Copenhagen, Hvidovre, Denmark
3 Ultrasound Clinic, Juliane Marie Center, H:S Rigshospitalet, Copenhagen, Denmark
4 Department of Epidemiology, Institute of Public Health, University of Aarhus, Aarhus, Denmark
Correspondence to Marie Louise Østerdal, Maternal Nutrition Group, Danish Epidemiology Science Centre, Statens Serum Institut, Artillerivej 5, DF-2300 Copenhagen S, Denmark (e-mail: mae{at}ssi.dk).
Received for publication March 2, 2007. Accepted for publication April 4, 2007.
We appreciate the opportunity to comment on Dr. Callaghan's commentary (1) on our original paper published in this issue (2). Dr. Callaghan raises pivotal points regarding the highly complex nature of preeclampsia.
Complexity exists on at least two different levels. First, there is variation in the time of onset, severity, and type of clinical presentation. Because a particular presentation may have particular etiologic implications, this variation should be captured in the data collection, when we want to study both the etiology of preeclampsia and the etiologic role of preeclampsia for other diseases. Second, a particular category of the clinical presentation—however precisely it has been captured and identified in the data collection—may not match a particular pathophysiologic entity; in other words, we do not "see" the underlying true disease, only how it presents clinically, and each category can only be taken to be a marker for an underlying disease or disease process. Although the problem described under the second level above is a basic premise for most epidemiologic research (sometimes described as "black box" epidemiology), the problem described under the first level above can in principle be solved by intensifying the observation of each woman. However, as mentioned by Callaghan, how intensely we are able to observe is limited by both the human and the material resources available to us and by logistic constraints: It simply may not be feasible to get the desired detailed information for a large number of individuals irrespective of available resources.
Our approach in the evaluation of the discharge registry (in our case, the Danish National Patient Registry) has been that of studying the etiology of preeclampsia in the Danish National Birth Cohort, that is, the relation between various exposures and preeclampsia. We posed two questions to evaluate the degree of misclassification of the preeclampsia diagnosis: 1) How large is the attenuation of exposure effect caused by the misclassification? and 2) what is the statistical power to detect an exposure effect when there is misclassification? In table 1, under the assumption of no differential misclassification and a (true) disease prevalence of 6 percent in the exposed and 3 percent in the unexposed, the degree of attenuation is given for various combinations of sensitivity and specificity. The statistical power of detecting an exposure effect is furthermore provided under two scenarios: 5,000 exposed and 45,000 unexposed (realistic in the Danish National Birth Cohort) or 1,000 exposed and 9,000 unexposed (corresponding to a smaller study). Table 1 confirms that the answers to questions 1 and 2 are highly sensitive to reductions in specificity, whereas a moderate sensitivity should be less of a concern.
|
Dr. Callaghan discusses the strategy of using the discharge registry as a "screening test" to be followed by a detailed review of medical records of the women who "screen positive," allowing reclassification of false positives from cases to noncases and thus increasing specificity. Given the already high specificity of the discharge registry (around 99.5 percent) and moderate sensitivity (around 70 percent) that were found in our original paper (2), there seems to be little to gain from such a strategy in both of the settings described above; that is, increasing the specificity to 100 percent does not have a substantial impact on the statistical power. However, if the aim is to identify as many cases as possible (e.g., to study the effect of exposure to preeclampsia on rare outcomes such as offspring cancer risk), we agree that a high sensitivity in the discharge registry would be primarily desirable, whereas a low specificity can be "repaired" during a review of records. It should be noted, though, that the lower the specificity in the discharge registry, the greater will be the records review task. In particular, if we are stuck with discharge registry data for studying the etiologic role of preeclampsia, we may not want to increase sensitivity at the expense of specificity, because a low specificity will, even if sensitivity is high, tend to undermine statistical power.
Dr. Callaghan's point about the need to identify cases with early onset preeclampsia (or any other nuance of presentation) can also be viewed in terms of specificity. If cases with late onset contaminate the case group of interest, this would imply low specificity and a corresponding attenuation of the observed measure of association. Mandatory registration of the gestational week of diagnosis, in addition to data on dates of admittance to and discharge from the hospital (such as is currently available in the Danish National Patient Registry), would be an important improvement of the discharge registries and should be considered.
In conclusion, it is our contention that a discharge registry with a high specificity is likely to do sufficiently well in revealing the etiology and etiologic role of preeclampsia.
 |
ACKNOWLEDGMENTS |
|---|
Conflict of interest: none declared.
|
References |
|---|
 TOP References |
|---|
- Callaghan WM. Invited commentary: identifying women with hypertension during pregnancy—is high specificity sufficient? Am J Epidemiol (2007) 166:125–7.
[Abstract/Free Full Text] - Klemmensen ÅK, Olsen SF, Østerdal ML, et al. Validity of preeclampsia-related diagnoses recorded in a national hospital registry and in a postpartum interview of the women. Am J Epidemiol (2007) 166:117–24.
[Abstract/Free Full Text]
CiteULike 
Connotea 
Del.icio.us What's this?
Related articles in Am. J. Epidemiol.:
- Validity of Preeclampsia-related Diagnoses Recorded in a National Hospital Registry and in a Postpartum Interview of the Women
- Åse K. Klemmensen, Sjurdur F. Olsen, Marie Louise Østerdal, and Ann Tabor
Am. J. Epidemiol. 2007 166: 117-124.[Abstract] [FREE Full Text]
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||