|
Abstract:
|
In clinical diagnosis of diseases, the evaluation of biomarkers draws keen attentions recently. A cutpoint is often selected to discern high risk persons from the row risk persons by dichotomizing naturally continous biomarker levels. We present an approach to select several cutpoints based on the covariate balancing propensity scores. The approach produces sensitivity and specificity of clustered data, adjusting for the possible correlation between observations. We find that the proposed approach can be used to improve sensitivity and specificity concurrently. Extensive simulation studies and real data from a colorectal adenoma study illustrate the advantages of our approach.
|
