Dr Troisi et al reported the results of a reanalysis of data from published cohorts to determine whether the elevated risk of high-grade squamous neoplasia associated with in utero diethylstilbestrol (DES) exposure continues with age and concluded that cervical intraepithelial neoplasia (CIN)-2+ incidence was higher among the DES exposed. We would like to challenge this claim in the context of methodological flaws in this observational study.

First, the authors link together data from several exposure cohorts and one unexposed (control) cohort. These cohorts do not arrive from the same source population. They are different in data quality and heterogeneity and underlying population characteristics. Exposure is measured using different approaches, and so are the invariant and time variant confounders. Selection bias as well may affect the point estimates and confidence intervals. The control cohort is small compared with the exposure cohorts and does not arrive from the source population.

Second, reanalyzing a data set previously used to draw hypotheses to conclude on the veracity of these hypotheses is not a scientific approach. The right procedure would be to test this hypothesis using other data sets.

Multiple comparisons have been carried out in this paper and should accordingly be taken into account when estimating confidence intervals. Before drawing conclusions from their data set, the authors should reckon their confidence intervals with lower P values. Applying a 1% significance level, it is very likely that analyzed risks would not be significant and the hypothesis ruled out.

Although the authors found that the relation exposure-disease varied according to the 4 analyzed cohorts (only the hazard ratios estimated from the Horne cohort, without adjustment for confounding factors but the birth year, was statistically significant), they did not test for a first-order qualitative interaction between exposure, disease, and cohorts and kept on drawing conclusions on aggregate data in spite of analyzing them separately.

The authors identified statistically significant differences for several confounding factors (CFs) at baseline between exposed and nonexposed women. However, the authors did not adjust their comparisons accordingly. As in the analysis already carried out in 2001 and not discussed in the present paper, they estimated hazard ratios with an adjustment for date of birth and cohort only and did not adjust for other CFs of CIN2+ in each cohort, especially when they analyzed the age effect: tobacco, number of sexual partners, number of pregnancies, age at birth of first child, and the number of Papanicolaou smears in the previous 5 years. Similarly, when the investigators analyzed the relation (1) DES exposure dose-CIN and (2) exposure-vaginal epithelial changes, they adjusted only for birth year and no other CFs. Furthermore, a major CF (human papillomavirus) was neither included nor properly discussed.

Observational studies are sensitive to biases and confounding, and although the investigators are highly qualified, they have not made a convincing point that these are adequately addressed in the presented study. We do believe that the study conclusions are erroneous and not likely supported in the analysis.