Abstract:
|
Classification of mixed binary outcomes is commonly seen in infectious disease when the disease outcome could consist of either relapse from previous occurrence of the disease or completely random re-infection. However, while a conventional approach using the baseline information may give a naïve classification of the outcome, little literature has suggested how to use the biomarker information attached to the event for classification, especially when the biomarker information is time-varying. In this talk, we will first review the data structure using Plasmodium Vivax infections in malaria as an example, and then proposes a model-based method using biomarker information to enhance the accuracy of classification. Our approach will be tested in extensive simulation experiments when there is a true underlying outcome, with superior sensitivity and specificity for better performance of prediction. The malaria infection data will be revisited using our proposed method.
|