Verification and incorporation biases in studies assessing screening tests: prostate-specific antigen as an example

Prostate-specific antigen (PSA) measurement is a widely used screening test for prostate cancer. However, studies assessing the performance of PSA suffer from verification bias and incorporation bias, because men with PSA levels less than 4 ng/mL do not undergo biopsy unless the digital rectal examination findings are suspicious for prostate cancer. We have reviewed how these biases have influenced such studies.

We compared the results of three studies that reported on the test characteristics of PSA, taking into account the types of bias present. We applied approximate verification bias corrections to previously reported age-specific reference ranges, based on the reported 22% cancer detection rate in the PSA range 2.5 to 4 ng/mL. To adjust for incorporation bias, we removed PSA from the criteria establishing the absence of prostate cancer and recalculated the test characteristics of PSA.

Verification bias led to an elevation in the sensitivity estimates of PSA. The specificity increased or decreased depending on the type of bias present. After accounting for incorporation bias, the sensitivity of PSA testing at a 4-ng/mL cutoff level was less than 67%, and the specificity was more than 80.6%.

Studies evaluating PSA testing have been biased because of the presence of verification and incorporation biases. Future studies must consider the possibility of these biases to obtain valid results.