In Reply Drs Leening and Pencina discourage the use of absolute NRI when comparing the performance of 2 predictive models by stating that “the absolute NRI is mathematically improper and has no clinical interpretation.” They also argue that the absolute NRI does not consider the prevalence of the event of interest. These arguments are incorrect and the refutations are presented in our Users’ Guide.
Briefly, the absolute NRI mathematically represents a net proportion of patients reclassified by a new model. Because it is a proportion, its clinical interpretation is easier than the additive NRI. For example, an absolute NRI of 6% means that 6% of the patients were better classified by the new model. An additive NRI may mean almost anything in terms of the absolute percentage of patients better classified, depending on the proportion of patients with and without events. The key difference is that the absolute NRI considers the prevalence of the event by computing the net benefit of the new model before applying it to the overall population.
Alba AC, Devereaux PJ, Guyatt G. Absolute vs Additive Net Reclassification Index—Reply. JAMA. 2018;319(6):617. doi:10.1001/jama.2017.20548
Customize your JAMA Network experience by selecting one or more topics from the list below.
Create a personal account or sign in to: