|
Abstract:
|
The two-phase study design is a cost effective option for assessing the predictiveness of emerging risk predictors. In this design, data on the outcome and inexpensive predictors is available for all study subjects (Phase I), but costly predictors are only measured on a judiciously selected subset (Phase II). We develop general statistical methods under the two-phase study design for assessing the performance of a statistical model in predicting binary outcomes. We also explore different sampling strategies for selecting Phase II subjects to improve the study efficiency. We develop inference procedures for our proposed methods, and demonstrate their performance in finite samples via simulation studies and analysis of data from a two-phase study of breast cancer.
|
