TABLE 2

Descriptions of and calculations for diagnostic tests conducted

EndpointDescriptionCalculation
ConcordanceProportion of compounds where the model accurately predicted clinical measure(TP + TN)/ Total # of compounds
SensitivityProportion of +HSE or scheduled drugs accurately identified by the modelTP/(TP+FN)
SpecificityProportion of drugs that do not induce +HSE or are unscheduled accurately identified by the modelTN/(TN+FP)
Positive Predictive Value (PPV)Proportion of drugs that are TPs relative to all drugs with a positive resultTP/(TP+FP)
Negative Predictive Value (NPV)Proportion of drugs that are TNs relative to all drugs with a negative resultTN/(TN+FN)
Adjusted PPVPPV adjusted for a pretest probability (prevalence) of 0.3(sensitivity x prevalence)/ [sensitivity x prevalence + (1 – specificity) x (1 – prevalence)]
Adjusted NPVNPV adjusted for a pretest probability (prevalence) of 0.7specificity x (1 – prevalence)/ [specificity x (1 – prevalence) + (1 – sensitivity) x prevalence)]
Value added PPV (VaPPV)Quantitative measurement of value added for positive results relative to pretest probabilityAdjusted PPV – prevalence
Value added NPV (VaNPV)Quantitative measurement of value added for negative results relative to pretest probabilityAdjusted NPV – (1 – prevalence)
Proportionate reduction in uncertainty (PRU) positive (+)Proportion by which model reduces uncertainty in predicting drugs with +HSE or are scheduledVaPPV/(1 – prevalence)
Proportionate reduction in uncertainty (PRU) negative (–)Proportion by which model reduces uncertainty in predicting drugs without +HSE or are unscheduledVaNPV/prevalence
  • FN, false negative; FP, false positive; TN, true negative; TP, true positive.