Activity Number:
|
594
- Recent Advances in Statistical Modeling for Multivariate/Correlated/Time-Varying Longitudinal Data
|
Type:
|
Invited
|
Date/Time:
|
Thursday, August 1, 2019 : 8:30 AM to 10:20 AM
|
Sponsor:
|
WNAR
|
Abstract #300470
|
|
Title:
|
Longitudinal Predictive Risk Modeling
|
Author(s):
|
Seonjin Kim* and Hyunkeun Cho and Mi-Ok Kim and Zhuangzhuang Liu
|
Companies:
|
Miami University and University of Iowa, College of Public Health and University of California San Francisco and University of Iowa
|
Keywords:
|
Bivariate time varying coefficients;
longitudinal studies;
generalized linear regression
|
Abstract:
|
Many longitudinal studies are concerned with evaluating relationships between potential risk factors and outcomes that were not concurrently measured. For example, BMI and blood pressure in early adulthood may influence the development of cardiovascular disease in middle age. In this paper, we propose a novel generalized bivariate time-varying coefficient model that enables investigating the simultaneous association between the response variable at later fixed time points and the predictors at an earlier time point. A strong association between early worsening risk factors and poor late outcomes may call for early preventive interventions. We propose modeling the association as a bivariate time-varying coefficient and a new non-parametric estimation method that combines features of the kernel method and the spline method. The Framingham Heart Study is analyzed to study how risk factors including BMI and blood pressure in early adulthood have an influence on the development of hypertension and predict the likelihood trajectory of occurrence of hypertension given a subject's health condition in early adulthood.
|
Authors who are presenting talks have a * after their name.