We read with interest the paper by Di Martino et al. published in the December 2004 issue of HEPATOLOGY 1 that examines estrogen-related effects on hepatitis C fibrosis. Specifically, progression of hepatitis C fibrosis was correlated with prior pregnancies, menopausal status, past use of oral contraceptives, and hormone replacement therapy. Hepatitis C-infected women who completed a survey and had previously undergone liver biopsy were eligible for the study. The investigators calculated a rate of progression of fibrosis, dividing fibrosis stage (in Metavir units) by years of hepatitis C infection. This approach has been used previously in a selfdescribed cross-sectional study. 2 The present study is labeled a "retrospective cohort study," although it would be better characterized as a cross-sectional study. Information from a single survey was correlated with results from a single liver biopsy. Patients were not followed by repeated biopsies or repeated surveys over time.
The distinction between cohort and cross-sectional is not merely semantic in this case. Figures 3 and4 are Kaplan-Meier curves for the development of significant fibrosis in "cohorts" of hepatitis C patients. Kaplan-Meier curves depend on the precise knowledge of when a patient in a cohort develops significant fibrosis. For example, a patient with a biopsy showing stage 4 fibrosis and 20 years of disease duration would be plotted on a Kaplan-Meier curve as having developed significant fibrosis (stage 2) after 20 years' disease duration. Without a prior biopsy, it is not known at what time the patient actually developed stage 2 fibrosis. To address the impossibility of accurate censoring, the authors could have back-extrapolated to estimate time to development of stage 2 fibrosis, assuming linear progression of fibrosis. The problem with that approach is that it involves using a calculated result to generate primary data-for re-analysis.
In short, even though the investigators describe a rate of progression of fibrosis, the data do not truly describe a cohort. Statistical methods generally reserved for cohort studies cannot be applied easily to cross-sectional data.