Dealing with Methodological Issues in the Functional Data Analysis of Actigraphy Data

• Activity Number: 176 - Modeling
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Stephen W. Looney* and William Vaughn McCall and Jordan S. Lundeen
• Companies: Augusta University and Augusta University and BlueChoice HealthPlan of South Carolina
• Keywords: basis functions; missing data; imputation; biomarker; validation; rest-activity data

Abstract

This presentation examines several methodological issues we have encountered when using functional data analysis (FDA) to analyze actigraphy data. For example, we discuss and compare methods used for handling missing actigraphy data, and how to determine the optimal number of basis functions to use when applying FDA. Curves fit to actigraphy data must take on non-negative values, so we also discuss how to restrict FDA curves so that they have no negative values. The methods and issues we discuss are illustrated using actigraphy data from our study of the utility of a rest-activity biomarker to predict responsiveness to antidepressants.