A new office test for laryngeal malignancy is evaluated in 100 patients with obvious advanced cancers and 100 healthy volunteers. It reports 98% sensitivity and 97% specificity. Why may these figures overestimate performance in a rapid-access clinic?
A diagnostic-accuracy study must include the spectrum in which the test will be used; extreme phenotypes exaggerate separation and transport poorly.
A. Verification bias, because every participant necessarily lacked a reference standard: The stem does not state whether reference verification differed; the explicit flaw is the extreme case-control spectrum.
B. Attrition bias, because diagnostic studies cannot retain participants: No loss to follow-up is described.
C. Spectrum bias from comparing unequivocal disease with healthy controls: Real clinics include dysplasia, inflammation, early tumours and comorbidity; separating extremes makes sensitivity and specificity look artificially high.
D. Lead-time bias, because earlier diagnosis always prolongs survival: This concerns survival comparisons, not test accuracy.
E. The prevalence is 50%, so sensitivity is mathematically invalid: Sensitivity and specificity can be calculated at any prevalence, although predictive values change.
If consecutive rapid-access patients all received the same blinded reference standard, spectrum and verification bias would be substantially reduced.
Independent Clinora educational preparation material; not official JCIE content, an accredited programme, or a substitute for local policy and specialist clinical judgement.
Back to the Section 1 bank