Spectrum bias

Spectrum bias arises when the patients enrolled in a diagnostic accuracy study span a narrower or different range of disease severity, stage, or competing conditions than the patients a clinician actually orders the test for. Classic case-control designs that compare people with advanced, unmistakable disease against fit, symptom-free controls inflate both sensitivity and specificity, because the easy cases are simple to classify. In real practice the test faces early, mild, or atypical presentations and patients with mimicking conditions, so accuracy falls.

When you read a diagnostic study, check who was enrolled: the setting (referral center versus primary care), the severity mix, and how the diseased and non-diseased groups were assembled. A two-gate (case-control) sampling design is a red flag; a single consecutive series of patients with the target symptom is stronger. Ask whether the reported sensitivity, specificity, and likelihood ratios would still hold in your own patients, who usually sit in the harder-to-classify middle of the spectrum.