Pooled sensitivity of a diagnostic test

Method

Sensitivity of an index diagnostic test against a stated reference standard (histology, or histology plus colposcopy), pooled across studies in a bivariate or hierarchical summary-ROC random-effects model. The index test, the reference standard and the positivity threshold are recorded on the edge.

The method is part of the endpoint identity. Studies measuring the same quantity by a different method are held as separate endpoints.

Notes

Pooled specificity is computed from the same 2x2 tables in the same model and also improves by rising, which is exactly why the two get merged — and the merge is wrong because they trade off along the ROC curve: any threshold shift that raises one lowers the other. Recorded as one endpoint, a threshold change reads as a null. Also distinct from positive and negative likelihood ratios and from the diagnostic odds ratio, which are derived combinations of both and are dimensionless; and from positive predictive value, which additionally depends on disease prevalence and therefore is not a property of the test.