Re: "increased concentrations of polychlorinated biphenyls, hexachlorobenzenes, and chlordanes in mothers of men with testicular cancer".

The paper by Hardell et al. (2003) is a potentially important contribution to the literature on endocrine disruption and testicular cancer. It may also have implications for other disorders of the male reproductive tract, collectively known as the testicular dys-genesis syndrome (Skakkebaek et al. 2001). It is therefore vital to establish the degree of confidence that can be placed in its findings. Issues arise in relation to the selection of cases and controls and to the selection of chemical and statistical analyses.

In a case-control study, the conventional practice is to recruit sequential cases (or a random sample) and not re select among them, other than for reasons of ability to participate, such as severity of illness or language problems. Hardell et al. (2003) stated that "... these patients did not represent all cases with testicular cancer admitted to these hospitals during this time," bur they did not explain why the other cases were not included; it was not because of refusal.

To assess the extent to which the apparently nonrandom method of case selection has led to bias, the following questions need to be answered. What proportion of the total incident cases in the various participating hospitals were deemed eligible? How was eligibility determined, and by whom? From the viewpoint of the quality of the study, the important possibility is that this decision was influenced by the hypothesis under study. If, for example, cases were more likely to be included if they were particularly "interesting" for the study (e.g., because of a family background involving agriculture), the study would have been seriously biased. Hardell et al.'s (2003) statement that "no selection bias occurred because the physicians treated patients regardless of tumor type" does not address this issue.

In choosing controls, the obvious course of action is to select the control group from the same hospital(s) as the cases (sometimes with attention needed for tertiary referral processes in the case of rare diseases or those requiring specialist treatment) to try to ensure that the two groups are drawn from the same underlying population. In this study, Hardell et at. (2003) used a random sample of the Swedish population instead. This led to a geographically biased sample, unless the population density of males approximately 20-40 years of age happened to correspond with the catchment areas for testicular cancer of the hospitals used for the recruitment of cases, weighted by the number of cases for each one. This geographical bias in the match between the two samples is important for spatially varying exposures, which are likely to include at least some of the chemicals studied.

It would be possible to explore the possible geographical bias introduced by the method of selection of controls if there is information on the spatial distribution of the case and control groups. The main possibilities to compare are a) the proportion of the two groups who are living in rural, and especially in agricultural, areas; and b) their distribution across the different regions of Sweden. At least some of the chemicals studied--although possibly not dichlorodiphenyldichloroethylene (DDE) (Ekbom et al 1996)--are likely to have considerable spatial variation and to have higher concentrations in farming areas in the case of hexachlorobenzene (HCB) and the chlordanes [U.S. Environmental Protection Agency (EPA) 1997]. Origin in a Farming family would be another way re examine the same thing, if such an analysis were to show that the case population is more agricultural, then it would be impossible to distinguish a real biological effect (that testicular cancer itself has "selected" farming or rural families because of exposure to these persistent organochlorine pesticides) from a selection effect created by the study design. However, if this is not the case, the findings cannot be attributed to this particular bias.

Hardell et al. (2003) gave no clear justification for their choice of these chemical and statistical analyses from among all possible endocrine disruptors (agents with estrogenic, antiestrogenic, and antiandrogenic effects) that persist in the body. HCB and p,p'-DDE do not seem to be controversial. Polychlorinated biphenyls (PCBs) are heterogeneous; as the authors state, some (but not others) are thought to be endocrine disruptors--but Hardell et al. (2003) provided no separate results for this group, which would be important in evaluating whether their observed effects are hormonally mediated. It would also be important to have separately-presented findings for "dioxin-like" PCBs (De Rosa et al 1997). The chlordanes are a less obvious choice, if only because they are considered to be probable human carcinogens for nonendocrine cancers (U.S. EPA 1997); therefore, it is difficult to interpret whether any observed effect on testicular cancer is or is not hormonally mediated.

The author declares he has no competing financial interests.

REFERENCES

De Rosa CT, Brown D, Ohara R, Garrett W, Hansen H, Holler J, et al. 1997. Dioxin and dioxin-Like compounds in soil, Part 1: ATSDR interim policy guideline. Agency for Toxic Substances and Disease Registry Toxicol Ind Health 13(6):759-768, Available: http://www.atsdr.cdc.gov/dioxindt.html [accessed 15 August 2003]

Ekbom A, Wicklund-Glynn A, Adami H-O. 1998. DDT and testicular cancer. Lancet 347:553-554

Hardell L, van Bavel B, Lindstrom G, Carlberg M, Dreifaldt AC, Wijkstrom H, et al. 2003, Increased concentrations of polychlorinated biphenyls, hexachlorabenzene, and chlordanes in mothers of men with testicular cancer, Environ Health Perspect 111:930 934.

Skakkebaek NE, Raypert-De Meyts E, Malm KE. 2001. Testicular dysgenesis syndrome: an increasingly common developmental disorder with environmental aspects Hum Reprod 16(5):97-970.

U.S. EPA. 1997 Toxicological Review of Chlordane (Technical) (CAS No. 12789-03-6). Washington, DC:U S. Environmental Protection Agency

Michael Joffe

Department of Epidemiology &

Public Health

Imperial College Faculty of Medicine

London, United Kingdom

E-mail: [e-mail]