Letters Section Editor: Robert M. Golub,
MD, Senior Editor.
In Reply: We agree with Drs Royston and Altman
that the AUROC for the logistic regression model was greater than that for
the CART model. We also recognize that there remains a debate regarding the
relative merits of CART compared with other models. Both the dichotomization
of continuous predictors and the classification of patients into a finite
number of risk categories can contribute to a reduction in the AUROC. However,
the AUROC may not be the best way of assessing the accuracy of nonlinear models
such as CART and newer, more complex statistical methods to assess this accuracy
are being explored.1 Our evaluation was limited
to the AUROC comparison because these new methods are not fully developed
or readily available.
Fonarow GC, Adams KF, Abraham WT, Yancy CW, Boscardin WJ. Risk Stratification for In-Hospital Mortality in Acutely Decompensated Heart Failure—Reply. JAMA. 2005;293(20):2467-2468. doi:10.1001/jama.293.20.2468-a