Anaes SAQs · Measurement & monitoring physics

Diagnostic test performance: sensitivity, specificity, predictive values and likelihood ratios

Primary — the 2x2 contingency table (TP, FP, FN, TN); SENSITIVITY equals TP divided by (TP plus FN), the true positive rate, prevalence-independent, SnNout (rules out); SPECIFICITY equals TN divided by (TN plus FP), the true negative rate, prevalence-independent, SpPin (rules in); PPV equals TP divided by (TP plus FP) and NPV equals TN divided by (TN plus FN), both DEPEND ON PREVALENCE (high prevalence raises PPV, lowers NPV); LIKELIHOOD RATIOS are prevalence-independent — LR+ equals sensitivity divided by (1 minus specificity), LR- equals (1 minus sensitivity) divided by specificity, LR+ greater than 10 or LR- less than 0.1 very useful; Bayes' theorem — post-test odds equals pre-test odds times LR; ROC plots sensitivity against (1 minus specificity), AUC 0.5 is chance, 1.0 perfect, greater than 0.9 excellent; screening uses high sensitivity, confirmation uses high specificity.