Maternal plasma cellular fibronectin for early prediction of preeclampsia

Abstract

To the Editors: We read with interest the article by Chavarria et al.1.Chavarria M.E. Lara-Gonzalez L. Gonzalez-Gleason A. Sojo I. Reyes A. Maternal plasma cellular fibronectin concentrations in normal and preeclamptic pregnancies: a longitudinal study for early prediction of preeclampsia.Am J Obstet Gynecol. 2002; 187: 595-601Abstract Full Text Full Text PDF PubMed Scopus (43) Google Scholar We applaud the investigators' effort in collecting frequent blood samples throughout pregnancy in a large cohort. For the current study, the authors selected all preeclamptic cases, excluded women with gestational hypertension or transient hypertension or other pregnancy complications, and selected two normal pregnancies for each case, matching on maternal age, date of delivery, and gestational age at sampling. This became a classic nested matched case-control study.2.Rothman K.J. Greenland S. Matching.in: Rothman K.J. Greenland S. Modern epidemiology. 2nd. Lippincott-Raven, Philadelphia (PA)1998: 147-161Google Scholar They confirmed previous reports that maternal plasma cellular fibronectin levels were significantly elevated in women who were destined to have preeclampsia later in pregnancy. The authors then calculated sensitivity, specificity, and positive and negative predictive values on the basis of the matched case-control data. The latter practice is a common problem in general clinical literature. Although the original study was a cohort study, the authors matched the cases with the controls on some potential confounders (eg, maternal age) in the subsequent nested case-control study. The cases and the controls now became more comparable on age. However, the statistical analysis did not account for the matching. When matching design is ignored, the study efficiency is hampered and the risk estimates are attenuated.2.Rothman K.J. Greenland S. Matching.in: Rothman K.J. Greenland S. Modern epidemiology. 2nd. Lippincott-Raven, Philadelphia (PA)1998: 147-161Google Scholar Moreover, matching could artificially inflate or reduce the true sensitivity and specificity, depending on the type of relationship among exposure (cellular fibronectin), outcome (preeclampsia), and confounder (maternal age). The estimates of the area under the receiver operating curve and the optimal cut point are also likely to be biased. The degree of the above biases depends on how strong the confounding effect is by the matching variable(s). Furthermore, because the authors excluded other subjects from the original cohort and matched the controls with the cases, the selected sample does not represent the prevalence of the disease in the original cohort. Therefore, because positive and negative predictive values are partly determined by the prevalence of the disease, these estimates are not generalizable to the original cohort. The degree of bias and discrepancy can vary from study to study and from variable to variable. Although the nested matched case-control design is an efficient approach for many clinical studies, correct statistical analysis is the key to obtain valid results. Appropriate common statistical methods for matched data include paired t test, Mantel-Haenszel χ2 test and conditional logistic regression. In a study such as this where a series of samples from each subject were obtained and analyzed, a repeated measures analysis is more appropriate. To calculate positive and negative predictive values, a weighting correction should be applied.3.Kleinbaum D.G. Kupper L.L. Morgenstern H. Selection bias.in: Kleinbaum D.G. Kupper L.L. Morgenstern H. Epidemiologic research. John Wiley, New York1982: 194-216Google Scholar