Serum Levels of Testosterone Do Not Provide Evidence of Selection Bias in Studies of Male Reproductive Health

To the Editor: Studies of semen quality are especially vulnerable to selection bias from low participation rates, as these studies include groups such as volunteers enrolled after advertisement, 1 candidates for vasectomy, 2 semen donor candidates, 3 or infertility patients. 4 Recently, Mullard et al5 suggested that serum levels of testosterone could be used to demonstrate selection bias in population studies of male fertility. They reported results from blood samples in 23 men and semen samples in 8 of these men, who were all recruited in a study of polychlorinated biphenyl (PCB) exposure and indicators of fecundity among 626 men. 6 The method of selecting the 23 men from the larger population was not reported, so that this selection may itself have introduced bias. Of the 23 participants, none that had a testosterone level below 400 ng/dl donated semen. The mean testosterone of the semen donors was 626 ng/dl, while that of the non-donors was 496 ng/dl. On the basis of this sample, Mullard et al5 concluded that men with lower levels of testosterone might be less likely to donate semen. Furthermore, they found testosterone levels and sperm counts to be inversely correlated in the eight men who donated a semen sample, and concluded that men with lower sperm counts may be over-represented in samples of semen donors, relative to their frequency in the general population. Due to the military drafting system in Denmark, all 18-year-old men are required to attend a compulsory medical examination to determine their fitness for military service. By collaborating with the military health board, we had a unique opportunity to study male reproductive health in an unselected population. We obtained semen samples and blood samples from 708 young men, which represented a participation rate of 18%. Due to the low participation rate, we performed a study to assess selection bias. In this study the same protocol was used, but at recruitment men were asked to give only a blood sample and not a semen sample. In this sample, 79% of the men (N = 183) agreed to participate. To test the hypothesis of Mullard et al, 5 we made similar comparisons of testosterone levels in the two subgroups of men: (1) men who agreed to deliver a semen sample, and (2) men who gave only a blood sample (Table 1). In both groups of men, only a very small percentage (4% and 7%, respectively) had testosterone levels below 400 ng/dl (≈14 nmol/Liter), and the mean testosterone levels for donors and non-donors were similar. For comparison, the data from Mullard et al5 are given in Table 2. Table 1: Comparison of Serum Testosterone in Semen Donors and Non-DonorsTable 2: Comparison of Serum Testosterone in Semen Donors and Non-Donors (Data from Mullard et al5 )We also examined the relation between testosterone level and sperm concentration among donors. Using an analysis of covariance to control for duration of abstinence, season, and time of sampling, we found no evidence that testosterone levels and sperm counts were negatively related, as reported by Mullard et al.5 If anything, these parameters tended to be positively related (r = 0.046) (Fig. 1). FIGURE 1: Relation between serum testosterone and sperm concentration (N = 708).Thus, in this large group of unselected men, serum testosterone did not appear to be elevated among semen donors, and could not be used as evidence of selection bias. Anne-Grethe Andersen Niels Jørgensen Anna-Maria Andersson Elisabeth Carlsen Niels E. Skakkebæk Tina Kold Jensen Niels Keiding Shanna H. Swan