Using a prospective case-control study design, baseline levels of plasma selenium, retinol, and retinol-binding protein, and baseline blood uric acid levels were compared in 136 case patients who subsequently died from cancer and 238 matched control subjects. Subjects were followed for an average of 8 1/2 years. In matched analyses, selenium levels were lower in case patients with gastrointestinal or prostate cancer; retinol levels, lower in those with gastrointestinal or breast cancer; retinol-binding protein levels, lower in case patients with gastrointestinal cancer; and uric acid levels, lower in a group with "other" cancers. However, only the uric acid association with "other" cancers and the retinol-binding protein association with gastrointestinal cancer were statistically significant (P < or = .02) in conditional logistic regression analyses controlling for multiple potential covariates. Relationships for each of the substances varied by cancer site, and although some relationships were suggestive, our results point to the need for larger studies with adequate numbers for site-specific analyses.