Sensitivity | How good is the test at identifying patients who die |

| True positives (ie, tested positive and died) |

| All deaths |

| |

Specificity | How good is the test at identifying patients who do not die |

| True negatives (ie, tested negative and lived) |

| All alive |

| |

PPV | In the event of a positive test, the probability that the patient will die |

| True positives |

| All positives |

| |

NPV | In the event of a negative test, the probability that the patient will not die |

| True negatives |

| All negatives |

| |

Accuracy | The proportion of all tests that have given a correct result |

| True positives and negatives |

| All positives and negatives |

| |

ROC | A curve created by mapping sensitivity against 1 − specificity. The AUC is a marker of performance with higher values indicating better diagnostic ability. An AUC = 1 indicates perfect performance whereas an AUC = 0.5 indicates that there is a 50:50 chance that the test correctly identifies those who die |

Comment |

Sensitivity and specificity are useful when considering the performance of a test at the population level. In contrast, PPV, NPV and accuracy are better when considering the performance of a test at the patient level (ie, in the event of a positive test result, what is the probability that the patient in front of me will or will not die?). It is also important to consider how easy a test is to apply in clinical practice (see discussion). |