Prenatal Diethylstilbestrol Exposure: A Harbinger for Future Testicular Cancer Incidence?

Testicular cancer incidence has increased 65% during the course of the last 40 years. In 1975, the age-adjusted rate of testicular cancer was 3.73 10 y and in 2016 it rose to 6.17 10 y (1). Possibly this increase could be partially attributable to increased prenatal exposure to estrogen-like compounds. As early as 1979, the disease was believed to have prenatal origins (2). Numerous studies have since been conducted to support the hypothesis that prenatal exposure to hormone or hormone-like compounds, specifically diethylstilbestrol (DES), influence testicular cancer development. These studies, however, have been hampered by considerable statistical uncertainty because of two factors. Testicular cancer occurs relatively rarely with only a 0.4% likelihood of it developing over the course of a male’s lifetime. Furthermore, testicular cancer accounts for only 0.5% of all incident cancers in a particular year. Contrasted with prostate cancer, which is associated with a 11.6% lifetime risk and attributable to 9.9% of all cancers developing in a year, testicular cancer is a particularly infrequent occurrence (1). This makes prospective studies of testicular cancer challenging to design as evidenced by the study by Strohsnitter et al. Only seven testicular cancer cases developed during approximately 40 years of follow-up among a cohort of 1787 men exposed to DES before birth. Although the investigators observed an increase in testicular cancer risk among this cohort when compared with the national rates, the estimate of this effect was imprecise (relative risk [RR] 1⁄4 2.04, 95% confidence interval [CI] 1⁄4 0.82 to 4.20). The imprecision was more pronounced when testicular cancer rates among this cohort were compared with those among a cohort of unexposed men followed for the same length of time (RR 1⁄4 3.05, 95% CI 1⁄4 0.65 to 21.96) (3). Also, DES was not frequently used. It was administered to between two and four million women for, among other indications, threatened miscarriage between 1940 and 1971 (4). Its use was then banned on report of women prenatally exposed to the drug having increased risk of clear cell adenoma of the cervix and vagina (5). Nonetheless, during this time period, it was not frequently used, with an exposure prevalence ranging from 1.5% (6) to 7% (7). The infrequency of DES usage also made case-control studies of the effects of DES on testicular cancer prone to imprecision. To address this dual issue of rare outcome and rare exposure rendering both cohort and case-control studies statistically uncertain, Hom et al. conducted a much-needed meta-analysis of six studies to present a precise summary estimate of the effect of prenatal DES exposure on testicular cancer risk (8). Although the resultant estimate of RR 1⁄4 2.98 (95% CI 1⁄4 1.15 to 7.67) reduced the imprecision of the estimate, it was based in part on the results of three retrospective case-control studies, including one not yet published. The number of exposed cases in these studies ranged from two (9) to five in the unpublished study in the current analysis. Exposure misclassification due to erroneous recall by the mother of a case may have inflated the estimates of these studies. Shifting of one case from the exposed category to unexposed, and recalculating the summary estimate, however, did not exert much influence and the summary estimate was only slightly reduced (RR 1⁄4 2.64, 95% CI 1⁄4 1.05 to 6.66). For the sake of completeness, the possibility should also be considered that one mother of a selected control incorrectly recalled that she was unexposed when in actuality she could have been exposed. This is reasonable because two studies had a DES exposure control distribution of less than 1% (9,10). Cited prevalence of DES exposure during the time it was in use had a lower range of 1.5% (6). There were also, however, some regions where the drug was not used at all (6). Nonetheless, shifting one unexposed control to the exposed category in these two studies resulted in a reduced summary estimate (RR 1⁄4 1.96, 95% CI 1⁄4 0.83 to 4.65). It is therefore somewhat reassuring that the association withstands challenges posed by exposure misclassification scenarios that have uncertain plausibility. Furthermore, these resultant confidence interval bounds derived by Hom et al. resulted in an estimate equally consistent with a minute effect and one that is appreciable. The precision in this estimate is a vast improvement over that in the comparison of the rates