Reliability and validity of prostate-specific antigen.

Abstract

To the Editor: Dr Eastham and colleagues found that many men with abnormal levels of prostate-specific antigen (PSA) were found to have normal levels when retested 1 year later. The authors thus concluded that PSA lacks specificity and that a single PSA test is unreliable. They recommend confirmatory assessment before undergoing further testing. We disagree with this logic. Specificity can only be improved by more stringent test criteria, not by further testing. Because the proposed ancillary evaluation includes biopsy, the interval of a few weeks is unnecessary for establishing a diagnosis or reducing adverse effects, but the delay is likely to increase the patient’s anxiety and fear. We also believe that the evidence of Eastham et al was insufficient to make these claims. Assessment of the validity of a screening test requires knowledge of the true disease status; this was not available in the study by Eastham et al. On the basis of such data, we found a single PSA determination to have a test sensitivity of 86% at a specificity level of 94% for a 5-year follow-up period. For patients aged 65 years or younger, the values were even higher—93% and 96%, respectively. Thus, the specificity of the PSA test appears superior to the specificities for the Papanicolaou smear and mammography used in screening for cervical and breast cancer, respectively. Furthermore, the assessment of the PSA test did not rely on cross-sectional evaluation with biopsy (thus inevitably detecting indolent cancers and missing some small carcinomas), but on clinically relevant disease as it was based on serum-bank study with clinical cancer diagnosis in the pre-PSA era. The large fluctuations and low specificity reported by Eastham et al may reflect insufficient reliability, perhaps related to laboratory problems, rather than poor validity of the test itself. The effect of such fluctuations cannot be assessed only by follow-up of those patients with high levels of PSA, but also require follow-up of those with low levels. Without examining the entire distribution, the conclusions are subject to regression toward the mean. The specificity of a test can be improved by using a higher cutoff value. Because prostate biopsy cannot definitively establish or rule out clinically relevant prostate cancer, a meaningful evaluation of test validity would require a design with follow-up for cancer either retrospectively nested within the serum bank study or prospectively after PSA testing. This design also allows for random sampling of controls within the study base or, ideally, random allocation to PSA testing in the context of an experimental study. However, we are not advocating routine screening for prostate cancer because this decision requires knowledge of mortality effects. Our ongoing randomized trial 3 may yield the evidence for decisions about routine screening.