If the gold standard is imperfect (e.g. expert consensus instead of biopsy), sensitivity and specificity are biased. Worse: if the reference standard incorporates the index test result, agreement is inflated (incorporation bias). Ask which standard was used and whether the index test fed into it.

Performance estimated in obvious cases ("sick" vs healthy volunteers) almost always overestimates real-world accuracy. Check whether enrolment was consecutive or random, and whether it spanned the clinical spectrum you'll apply the test to (mild, atypical, comorbid).

Partial verification bias arises when positive-index-test patients are more likely to receive the gold standard than negative ones — inflates sensitivity, deflates specificity. Look for differential or post-hoc verification protocols.

If the person interpreting the index test also knew the reference result (or vice versa), expect review bias: estimates drift toward perfect agreement. Imaging and pathology studies should explicitly report blinding between readers and from clinical context.

If indeterminate, missing, or excluded results were silently dropped, the 2×2 misrepresents real practice. Ask how many results fell outside the binary scheme and how they were handled (re-tested, excluded, counted as negative…).

Sensitivity / specificity often vary by age, severity, prior testing, or setting. A single 2×2 hides those differences. Check whether the authors pre-specified clinically meaningful subgroups and reported them separately.