Biomedical subjects
Publications and source records attributed to Xiaogang Su.
A tree procedure is proposed to check the adequacy of a fitted logistic regression model. The proposed method not only makes natural assessment for the logistic model, but also provides clues to amend its lack-of-fit. The resulting tree-augmented logistic model facilitates a refined model with meaningful interpretation. We demonstrate its use via simulation studies and an application to the Pima Indians diabetes data.
This paper provides explicit sample size determination formulas for planning a long-term trial in patients with chronic disease by using available results from existing short-term studies that may predict long-term disease progression patterns. The sample size calculation formulas are flexible to incorporate different nonlinear disease progression patterns. Various within-patient correlation structures are considered. By using the proposed formulas, sample size sensitivity can be easily explored for possible choices of study duration, assumed nonlinear disease progression patterns, randomization ratio, and expected clinical meaningful difference in the end of study. In addition, sample size calculation formulas are provided when the primary endpoint is change from baseline. Discussions on the relationship among required sample size, study duration, randomization ratio are also included.
We study a hybrid model that combines Cox proportional hazards regression with tree-structured modeling. The main idea is to use step functions, provided by a tree structure, to 'augment' Cox (1972) proportional hazards models. The proposed model not only provides a natural assessment of the adequacy of the Cox proportional hazards model but also improves its model fitting without loss of interpretability. Both simulations and an empirical example are provided to illustrate the use of the proposed method.
A method of constructing trees for correlated failure times is put forward. It adopts the backfitting idea of classification and regression trees (CART) (Breiman et al., 1984, in Classification and Regression Trees). The tree method is developed based on the maximized likelihoods associated with the gamma frailty model and standard likelihood-related techniques are incorporated. The proposed method is assessed through simulations conducted under a variety of model configurations and illustrated using the chronic granulomatous disease (CGD) study data.