A0968
Title: Nonparametric estimation of repeated densities with heterogeneous sample sizes
Authors: Xiongtao Dai - University of California, Berkeley (United States) [presenting]
Jiaming Qiu - Iowa State University (United States)
Zhengyuan Zhu - Iowa State University (United States)
Abstract: Estimating the prevalence age distributions of patients with different diseases is considered . A key challenge comes from the highly varying sample sizes for different conditions, making it difficult to estimate the age profile of a rare condition. To address this challenge, we propose a fully data-driven approach to pool information across conditions and estimate each distribution efficiently, without specifying a parametric form. Our technique draws from functional data analysis, which concerns, for example, a sample of developmental trajectories. We model densities as random trajectories and obtain low-dimensional exponential families for approximation, which is theoretically justified. We will show that the proposed approach yields interpretable results and is numerically efficient for modeling data from electronic health records.
