Martina Castellucci · Lab

Lab · Healthcare data

Diagnostic test

How far to trust a positive result, and where to set the cut-off of a biomarker.

Positive predictive value
Negative predictive value
LR+
LR−

500 people tested

Sick, positiveSick, negativeHealthy, positiveHealthy, negative

Predictive values and prevalence

PPVNPV
Method and data

PPV = Se·P / (Se·P + (1−Sp)(1−P)); NPV = Sp(1−P) / (Sp(1−P) + (1−Se)P); LR+ = Se/(1−Sp); LR− = (1−Se)/Sp. The biomarker values are simulated from two log-normal distributions. AUC by the trapezoidal rule; Youden J = sensitivity + specificity − 1. For learning only, not for clinical decisions.