Effect of verification bias on screening for prostate cancer by measurement of prostate-specific antigen

The sensitivity and specificity of a screening test are biased when disease status is not verified in all subjects and when the likelihood of confirmation depends on the test result itself. We assessed the screening characteristics of the prostate-specific antigen (PSA) measurement after correction for verification bias.

Between 1995 and 2001, 6691 men underwent PSA-based screening for prostate cancer. Of these men, 705 (11 percent) subsequently underwent biopsy of the prostate. Under the assumption that the chance of undergoing a biopsy depends only on the PSA-test result and other observed clinical variables, we used a mathematical model to estimate adjusted receiver-operating-characteristic (ROC) curves.

Adjusting for verification bias significantly increased the area under the ROC curve (i.e., the overall diagnostic performance) of the PSA test, as compared with an unadjusted analysis (0.86 vs. 0.69, P<0.001, for men less than 60 years of age; 0.72 vs. 0.62, P=0.008, for men 60 years of age or older). If the threshold PSA value for undergoing biopsy were set at 4.1 ng per milliliter, 82 percent of cancers in younger men and 65 percent of cancers in older men would be missed. A digital rectal examination that is abnormal but not suspicious for cancer does not affect the overall performance characteristics of the test.

A lower threshold level of PSA for recommending prostate biopsy, particularly in younger men, may improve the clinical value of the PSA test.