StATS: Spectrum Bias (January 4, 2005)

I tried to start a page on diagnostic tests a while back, but have not had the time to fully develop it. One of the important issues for diagnostic tests is spectrum bias.

The sensitivity and specificity of a diagnostic test can depend on who exactly is being tested. Think of disease as a range of possibilities from slight to moderate to extreme. If only a portion of the disease range is included, you may get an incorrect impression of how well a diagnostic test works. This is known as spectrum bias.

The most obvious manifestation of spectrum bias is in the use of a case-control design in studying diagnostic tests. In the case control design, a group of patients with disease and a group of patients without disease are asked to undergo a diagnostic test. They already know that they have the disease, but they are asked to take the test to see how well it performs. In a case-control design, though, you get the black (overtly diseased) and the white (overtly healthy) but often miss out on the gray (subtly diseased). As a result, the difference in the diagnostic test between healthy and diseased patients is often overstated leading to overly optimistic values for sensitivity and/or specificity.

Much better would be to evaluate sensitivity and specificity in a cohort design. With this design, people who have certain symptoms and who come to a health care professional for evaluation are given a diagnostic test and that test is compared to a gold standard (which is often just waiting to see if the disease fully manifests itself over  time). With such a design, you are likely to get a wide range of disease states and (hopefully) a fairer depiction of the performance of a diagnostic test.

Spectrum bias is a potential issue whenever there is heterogeneity in the response of patients to a diagnostic test. For example, in Muhlerin 2002, the performance of an enzyme immunoassy for Chlamydia trachomatis was shown to vary by age group. The sensitivity was much better in the patients older than 24 years (76% vs 58%). While the specificity showed less dramatic changes, it was also significantly higher  in the older group (99.5% versus 99.2%). With such a finding, you should be cautious about extrapolating results of the study.

Spectrum bias can also influence other measures in a diagnostic test, such as the ROC curve and the likelihood ratios. The way to avoid spectrum bias is to include a broad range of disease severity in your sample and to try to identify sources of heterogeneity in that sample.

Spectrum bias is just one of several issues that you face when you make a critical evaluation of a research article on diagnosis. The STARD initiative promotes good quality reporting for these type of articles and will help you evaluate diagnostic tests properly.

The STARD Initiative -- Towards Complete and Accurate Reporting of Studies on Diagnostic Accuracy. STARD Group. Accessed on 2003-07-29.

Further reading

This page was written by Steve Simon while working at Children's Mercy Hospital. Although I do not hold the copyright for this material, I am reproducing it here as a service, as it is no longer available on the Children's Mercy Hospital website. Need more information? I have a page with general help resources. You can also browse for pages similar to this one at Category: Diagnostic testing or Category: Observational studies.