Prostate cancer association studies: pitfalls and solutions to cancer misclassification in the PSA era.

Widespread screening of American men for elevated PSA has changed the characteristics of prostate cancer cases in the U.S. The influence of the changed nature of prostate cancer cases in the PSA era and the need for careful consideration of who is a "case" and who is a "control" on the ability to detect associations of risk factors with prostate cancer in etiologic epidemiologic studies merits discussion. Issue 1: prostate cancer cases diagnosed in the PSA era are enriched with a pool of early lesions, which may differ in etiology, and are deficient in advanced lesions, which are the most likely to be the product of promotion and progression events. By admixing the two types of cases (i.e., imperfect specificity), the associations previously detected using epidemiologic designs when the majority of cases were clinically detected may no longer be apparent in the PSA era when the majority of cases are now detected in the pre-clinical phase. Researchers must now tailor hypotheses such that they are testable using early stage cases or specifically augment the number of advanced cases when testing hypotheses related to extraprostatic growth and progression. Issue 2: even when controls are screened for elevated PSA to rule out the presence of prostate cancer, some proportion of those controls currently will have one or more foci of prostate cancer. The imperfect sensitivity of the PSA test coupled with diagnostic work-up may in part result from (a) lack of PSA elevation in some men with prostate cancer or (b) failure of biopsy to sample the tumor focus in men with elevated PSA. Misclassification of men with undetected prostate cancer as controls usually produces a bias that tends to deflate associations. Given this type of disease misclassification, whether an association still can be statistically detected depends on the extent of misclassification, the magnitude of the true association, the prevalence of the exposure in the true controls, and the sample size, although in general moderate nondifferential misclassification does not lead to profound attenuation. However, under the same scenario attenuation does not occur in cohort or case-cohort studies in which the rate or risk ratio (RR) is calculated. That prostate cancer cases diagnosed in the PSA era are enriched with early stage, minimally invasive disease in our opinion is likely to pose a far more serious obstacle to epidemiologic research on the etiology of clinically important prostate cancer than the issue of inclusion as controls some men who have undiagnosed prostate cancer because of imperfect sensitivity of PSA screening and biopsy sampling error. Copyright 2003 Wiley-Liss, Inc.