|
Abstract:
|
In biomedical practices, multiple biomarkers are often combined using a classification rule with tree structure to make diagnostic decisions. These tree structures, and cutoff points at each node of a tree, are commonly chosen based on experience of practitioners, and there is a lacking in statistical methods that can be adopted to evaluate the optimal prediction performance of a classification tree, or to guide the choice of optimal cutoff points. We propose to search for the optimal decision rule through a generalized semi-parametric regression model using the approach of rank correlation maximization. We show that the proposed estimator is consistent and asymptotically normal under some regularity conditions. The proposed method is flexible in modeling, and computationally feasible when there are many biomarkers available for classification or prediction. Using this method, we are able to guide the choice of optimal cutoff at tree nodes, as well as estimate optimal prediction performance of multiple biomarkers combined by a tree-based classification rule.
|
