Recent adverse trends in semen quality and testis cancer incidence of Finnish men: reply to Bonde et al., IJA 2012

Abstract

Dear Editor,

Bonde et al. have recently shown great interest both in our published and unpublished data (Toppari et al., 2010; Bonde et al., 2011) by writing comments and letters (Bonde et al., 2011; Jensen et al., 2011). We have responded (Skakkebæk et al., 2011; Toppari et al., 2011) and others have commented the discussion (Sharpe, 2011; Spira, 2011; Vogel, 2011; Wilcox, 2011). This time Bonde et al. (2012) request more information and additional analyses on our data of declining semen quality in Finland (Jørgensen et al., 2011). Most of the requested information was there already in our article, but we reiterate some of it also herein. The participation rate varied between 13 and 17% at different time periods and it is very unlikely that it would have caused a differential selection bias. We also indicated that the technician who made all the semen analyses scored systematically 11% higher sperm counts than the reference in the external quality control programme; i.e., this did not vary either at different time periods. In studies where the Finnish data were compared with those from other countries this was taken into account (e.g., Jørgensen et al., 2002), but here we were asked by the reviewers to present the data as such, although we originally included also adjusted values according to the quality control scheme. Bonde et al. suggest that our results might be artefacts owing to chosen statistical model or violation of model assumptions. For the regression analyses semen volume, sperm concentration and total sperm counts were transformed, and the significance levels presented for differences between investigation periods and birth years were all based on cubic root transformations, which gave the best model. We compared the back-transformed data and tested the model assumptions, which verified the validity of the results, i.e. they are not artefacts. The commentators also ask what changed the estimates from crude values of 60, 54, and 50 million/mL to adjusted values of 67, 60 and 48 million/mL. The answer is very simple: differences in the abstinence time which influences the values most when the abstinence is below 48 h. Regression models using natural logarithmic transformations of semen variables tend to give slightly reduced estimates compared with cubic root transformed models. However, the period differences and effect of covariates are more easily interpretable from models using natural logarithmic transformation, which we used when reporting adjusted semen volume, sperm concentration, total sperm counts and total number of morphologically normal spermatozoa. This was also stated in our original publication. Thus, irrespective which of the two transformations we used the adverse trend in semen quality was apparent as it also was from the crude means and medians. Finally, Bonde et al. encourage us to make the raw data available for others. We appreciate openness in science, but this is not the right way to advance it, because the raw data alone is not enough to allow reliable analysis. In semen studies one has to know exactly how the samples were collected and what were the background variables, such as abstinence time, previous medical records etc. We doubt whether any data protection authority would allow open access to such data even in coded form. Furthermore, we continue analysis of these data ourselves in longitudinal follow-up of the same men, and we do not want a competing group to exploit our work in the way that recently occurred to us (Bonde et al., 2011; Skakkebæk et al., 2011).