Fused Lasso to Determine the Risk Factors for Dementia

• Activity Number: 256
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Mental Health Statistics Section
• Author(s): Samrachana Adhikari*+ and Fabrizio Lecci and Ryan Tibshirani and Brian Junker and James Becker and Oscar Lopez
• Companies: Carnegie Mellon University and Carnegie Mellon University and and Department of Statistics, Carnegie Mellon University and University of Pittsburgh and University of Pittsburgh
• Keywords: sparse logistic regression ; fused lasso ; cognitive study ; longitudinal study ; Alzheimer's disease

Abstract

Determination of the onset of cognitive symptoms is one of the most important aspects of any epidemiological study of dementia. In this study we investigate the longitudinal information available from 1989-90 to 2011-12, to predict the earliest manifestation of the clinical symptoms of dementia.

We use a sparse logistic regression approach, with fused lasso penalty, to identify sets of time variant and invariant predictors that are influential in the onset of dementia 10 years ahead. Our model uses clinically observable data from normal subjects at t years of age to predict onset of symptoms at t + 10 years of age. The fused lasso penalty utilizes the L1 norm of the coefficients in two ways, encouraging many of the coefficients to be exactly zero, and encouraging coefficients of the same predictor to be similar in value across neighboring age panels. Aside from reducing the complexity of our predictive model and increasing interpretability, enforcing this property reflects the continuity of time across panels, which would otherwise be unaccounted for.