Integrated discrimination index

As described previously, AUROC can be useful to quantify a marker s ability to discriminate between samples from animals with and without injury. However, it does not make clear the extent to which the information provided by the candidate marker overlaps with information provided by the standard markers. One way to address this question is to compare the performance of a model that contains the standard markers together with the new candidate marker to a model that contains the standard markers without the new marker. The net reclassification index (NRI) and integrated discrimination index (IDI) (Pencina et al., 2(X)8) are metrics that can be used to quantify and test the difference in performance between binary logistic regression models and thus assess the amount of information added by the new marker. 

Objective measures of model evaluation are desirable in model building. Ideally, model performance techniques should be indexed to the purpose for which the model is intended to be used and as such should be considered an integral part of all general analysis plans. This brief overview concentrates on methods to select or reject models. By far the majority of methods used for model selection fall within the framework of model discrimination. 

---