Table 6 Illustration of NRI and IDI computations.
Truth | ||
|---|---|---|
Occasions | case | control |
Po < Pu | Nss | Nsn |
Po >= Pu | Nns | Nnn |
Total | Ns | N − Ns |
- The survival probabilities using the original model and the updated model are denoted as Po and Pu, respectively. The numbers of survivors and non-survivors were shown in the table according to their Po and Pu values. Defining p1 = Nss/Ns (increasing rate of successfully predicting survivors), p2 = Nnn/(N − Ns) (increasing rate of successfully predicting non-survivors), p3 = 1 − p1, and p4 = 1 − p2. Then NRI = p1 + p2 − p3 − p4 = 2(p1 + p2 − 1). The IDI was defined as the average improvement in survival probability for all patients after adopting the updated model:
where Poi and Pui were the survival probabilities using the original model and the updated model for the ith individual; i = 1, …, N.
