Table 1 Some of the metrics typically used to report the performance of a risk prediction tool applied to unplanned hospital admission
True outcome | |||
|---|---|---|---|
Admitted to hospital | Not admitted to hospital | ||
Predicted outcome | High-risk | True positive (TP) | False positive (FP) |
Low-risk | False negative (FN) | True negative (TN) | |
Performance metric formula | Description |
|---|---|
Accuracy = (TP + TN)/(TP + FP + FN + TN) | Accuracy measures how well the risk prediction tool identifies people who were and were not admitted to hospital |
Sensitivity (aka recall) = TP/(TP + FN) | The proportion of high-risk people who were admitted to hospital |
Specificity = TN/(TN + FP) | The proportion of low-risk people who were not admitted to hospital |
Positive predictive value = TP/(TP + FP) | The proportion of high-risk people who were admitted to hospital |
Negative predictive value = TN/(TN + FN) | The proportion of low-risk people who were not admitted to hospital |
Concordance statistic (aka c-statistics or area under receiver operating characteristic): | The probability that a randomly selected person who was admitted to hospital will have a higher modelled probability of admission than a randomly selected person who was not admitted to hospital |