Choosing the Wrong Patients Will Bias Estimates of the Usefulness of a Diagnostic Test
For clinicians to appropriately use diagnostic tests in clinical practice, they need to know how well the tests can distinguish between those who have the target condition and those who do not. If investigators choose clinically inappropriate populations for their study of a diagnostic test (introducing what is sometimes called spectrum bias), the results may seriously mislead clinicians (see Chapter 18, Diagnostic Tests).
In this chapter, we present a series of examples that expand on the points related to spectrum bias. Working through these examples, you will gain a deeper understanding of which characteristics of a study population are and are not likely to result in misleading results. Readers will find an elaborated version of this demonstration, intended to assist teachers in interactive sessions with small groups, in another publication.1
Target-Positive Patients with Unequivocally Severe Disease and Target-Negative Patients with no Reason to Suspect Disease are the Wrong Patients to Study
Ideally, the ability of a test to correctly identify patients with a particular disease, condition, or outcome (target-positive patients) and those without (target-negative patients) would not vary from patient to patient. A test may, however, perform better when used to evaluate patients with more severe disease than it would in patients whose disease is less obvious and/or less advanced. Moreover, clinicians do not need diagnostic tests when the disease is clinically obvious or sufficiently unlikely that they need not seriously consider it.
A study of a diagnostic test involves performing the test of interest, together with a second test or investigation (which we will call the reference standard, criterion standard, or gold standard) in patients with and without the disease or condition of interest. We accept the results of the reference standard as the criterion by which the results of the test under investigation are assessed.
In designing such a study, investigators sometimes choose patients with unequivocally far-advanced disease together with unequivocally disease-free people, such as healthy asymptomatic volunteers. This approach ensures that the criterion standard will not misclassify any patients and may be appropriate in the early stages of developing a test. Any study performed on a population that lacks diagnostic uncertainty may, however, produce a biased estimate of a test’s performance relative to a study restricted to patients for whom the test would be clinically indicated.
Distributions of Test Results Illustrate the Spectrum Problem
A crucial issue in the design of a diagnostic test study is the distribution of severity of illness or abnormality among the patients who were enrolled. We refer to this distribution as the spectrum of disease, illness, or abnormality.
For example, consider brain natriuretic peptide (BNP), which is a hormone that the ventricles of the heart secrete in ...