When studying the effects of drug exposure in diseases with a long asymptomatic clinical course, exposure classification may be biased by the gradually developing "visibility" of the disease. Benign prostatic hyperplasia (BPH) is such a disease. We found that cardiovascular morbidity is two times more prevalent in patients starting drug treatment for BPH when compared to age-matched population controls. This resulted in a difference of cardiovascular prognostic factors between the exposed and non-exposed. This feature can jeopardize the validity of non-randomized comparisons of drug effects. Moreover, the existence of non-treatment strategies, disease under-reporting, and an elderly population with a high baseline risk of experiencing (cardiovascular) outcome events were encountered as methodological problems. When studying adverse cardiovascular effects in patients using BPH products in a non-randomized fashion, an important question is whether we can measure in the database all relevant prognostic factors and use the information for statistical adjustment. This question is an important challenge to observational research and once again stresses the need for control of possible biases in choosing an appropriate study design.