Using Two-Fold Fully Conditional Specification to Impute Longitudinal Healthy Aging Index Scores

• Activity Number: 428
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Elizabeth L. McCabe* and Joanne M. Murabito and Kathryn L. Lunetta and Susan Cheng and Martin G. Larson
• Companies: Boston University School of Public Health and National Heart, Lung, and Blood Institute/Boston University Framingham Heart Study and Boston University School of Public Health and Brigham and Women's Hospital and Boston University
• Keywords: two-fold fully conditional specification ; longitudinal ; healthy aging index ; multiple imputation

Abstract

The healthy aging index (HAI) comprises 5 components designed to identify individuals who age successfully (Newman 2008). HAI scores (integers 0=healthiest to 10=least healthy) can predict mortality in older adults. We assessed how the HAI tracks longitudinally in the Framingham Heart Study offspring cohort, specifically among N=1348 attendees who were aged ?60 years at exam 5. We constructed the HAI at exams 5-8: 28% were missing ?1 HAI component at exam 5, 54% by exam 8. We faced a complex missing data problem with some data missing completely at random and other data missing due to non-attendance (sometimes due to comorbidity or death). For each HAI component, we applied multiple imputation using 2-fold fully conditional specification (2fFCS) which allowed us to handle longitudinal data and accommodate non-attendance that was death or morbidity related. Using 2fFCS increased sample size and events relative to complete-case analysis, and maintained stable hazard ratio estimates in analyses of aging studies where dropouts are common. We also compared imputing individual HAI components versus imputing the HAI itself: the latter resulted in faster convergence and similar HAI values.