Why pooling sensitivity alone gets diagnostic tests wrong
16 October 2026 · Cohira team
Every diagnostic study reports two linked numbers, and every threshold trades one against the other. Pool sensitivity in one meta-analysis and specificity in another and both answers are wrong.
The bivariate model
It pools sensitivity and specificity jointly, respecting their correlation, and draws a summary ROC curve with a confidence region around the summary point. That is the Cochrane-recommended standard, and what journals expect.
From accuracy to the bedside
Likelihood ratios and a Fagan nomogram translate pooled accuracy into post-test probability: what a positive or negative result actually does to your suspicion in a given patient.
Quality is the battleground
QUADAS-2 risk-of-bias judgements, threshold effects, and spectrum differences explain most heterogeneity in accuracy reviews. Reviewers go straight there; the analysis should get there first.