Principal Trend Ratio Regression: Comparing Nonlinear Longitudinal Trajectories Between Groups via a Ratio of Inter-Period Changes When a Linear Trend Fits Poorly

• Activity Number: 684
• Type: Contributed
• Date/Time: Thursday, August 5, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Robert M Boudreau*+
• Companies: University of Pittsburgh
• Keywords: Growth Curves ; Random effects ; Longitudinal data analysis ; Exercise ; Observational cohort studies

Abstract

Random slopes & intercepts mixed models with a linear trend are widely used for longitudinal data. Yet, in practice the growth curve often looks more curved than linear or has eras of timepoints with steeper or diminishing changes. Polynomial terms are difficult to interpret. With the goal of one parameter that compares groups, a new growth curve model identifies a "principal trend" with the group comparison as a multiplier of the changes by the referent. Covariates and random effects apply. Gait speed (GS) over 20m was measured in an elderly cohort (n=1462,age 69-80) at years 1,3,5,8. Men at baseline had 0.24 m/sec higher GS than women, but their GS declined 21% more (beta=1.207,p=0.002) than women over time. Regular exercisers had higher baseline GS. The confidence interval for the multiplier vs non-exercisers covers 1. Still slowing, they maintain a higher parallel GS trend over time.