Ninety percent (90%) of articles evaluated diagnostic tests in indicated rather than convenience populations of patients. Unfortunately, only 54% of the articles described the techniques used to enter patients into the studies, which would allow an assessment of the degree to which sampling bias affected the study results. Also, only 61% of articles adequately described the spectrum composition (“case mix”) of their study sample. This omission is important because new diagnostic tests display an inflated diagnostic accuracy when evaluated in a sample with a limited spectrum of disease. Patients with rapidly progressive or end-stage disease are more easily separated from nondiseased patients than patients with mild or early-stage disease. The presence of other covariates that affect the new diagnostic test and exist in some but not other patient subgroups within a case mix also alter the measured diagnostic accuracy of a test. The importance of providing information on the spectrum of disease is further emphasized by the observation that most of the variation in the measured diagnostic accuracies of new tests between different evaluative studies results from differences in the case mix of the study population. Spectrum bias cannot be corrected post hoc.
Readers of evaluative studies are assisted in their decision to apply a new diagnostic test to their patient population if the investigators analyzed the diagnostic properties of the test in pertinent patient subgroups. Subgroup analysis is important in diagnostic test research because the reported estimates of diagnostic accuracy represent an average value for the entire sample of study patients. Tests may work better, however, in different patient subgroups. The importance of subgroup analysis was first recognized in the evaluation of exercise ECG wherein a higher diagnostic accuracy was observed in patients with typical compared with atypical angina Our review of the study articles indicated that only 29% evaluated diagnostic accuracies of the tests in pertinent patient subgroups.