Identifying Subgroups of Enhanced Predictive Accuracy from Longitudinal Biomarker Data with Applications to Monitoring Fetal Growth

• Activity Number: 466
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Jared Foster*+ and Danping Liu and Aiyi Liu and Paul Albert
• Companies: NICHD and Eunice Kennedy Shriver National Institute of Child Health and Human Development and NICHD and Eunice Kennedy Shriver National Institute of Child Health and Human Development
• Keywords: Personalized medicine ; Tree-based methods ; Longitudinal data ; Fetal growth

Abstract

Longitudinal monitoring of biomarkers is often done to predict disease or a poor clinical outcome. Recently, new methodology has been developed to predict binary disease outcomes from longitudinal data. Typically these longitudinal predictors are evaluated across an entire population; however, it is also possible that this prediction is only accurate in one or more subsets of the population. In this work, we propose new statistical methodology for identifying subgroups of the population which show markedly improved prediction. We consider the problem of identifying such subgroups, while simultaneously attempting to control the type-I error. This work is motivated by an interest in predicting macrosomia using repeated ultrasound measurements during pregnancy, as we believe the accuracy of these predictions might be enhanced within certain subgroups of pregnant women. We propose a tree-based approach, which extends the classification and regression tree (CART) methodology to a longitudinal classification setting. To assess the performance of the proposed methods, a simulation study is undertaken. In addition, the proposed methods are applied to data from the Scandinavian Fetal G