Nonparametric Effect Decomposition with an Application to Trend Decomposition for HIV Incidence Rate in Rakai Teenagers, Uganda

• Activity Number: 492
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Xiaoyu Song*+ and Ying Wei and John Santelli
• Companies: Columbia University and Columbia University and Columbia University
• Keywords: Nonparametric regression ; attributable risk models

Abstract

The HIV incidence rate has declined over past 10 years among the teens in Rakai, Uganda. Meanwhile, social environment, individual sexual behaviors and demographic composition also changed over time. It is desirable to understand the dynamic contributions of those factors towards the decline of HIV incidence. Most of existing methods for effect decomposition and relative contributions are focused on a fixed time point. We propose to quantitatively decompose the declining trend to various contributors based a set of nonparametric models. Specifically, we model the association between the main outcome (HIV incidence) and potential contributors using nonparametric varying-coefficient poison regression; and then model the changes of the contributors over time using separate nonparametric regressions. Combining those models, we develop a new index that visualizes the relative contributions, and allow the decomposition to vary over time. We applied the methods to the Rakai Community Cohort Study from 1999-2011 for adolescent of 15 to 19 years old. The results provide new insights to the HIV cohort study.