Comparative Study of Four Methods in Missing Value Imputations with Dropouts from Longitudinal Studies

• Activity Number: 538
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: IMS
• Author(s): Michikazu Nakai*+ and Din Chen and Kunihiro Nishimura and Yoshihiro Miyamoto
• Companies: National Cerebral and Cardiovascular Center and University of Rochester and National Cerebral and Cardiovascular Center and National Cerebral and Cardiovascular Center
• Keywords: imputation ; missing value ; dropout ; simulation ; longitudinal ; epidemiology

Abstract

In this talk, we present the investigation of the efficiency of four classic imputation methods (i.e. Complete Case method, Mean imputation method, Last observation carried forward method and Multiple imputation method) in longitudinal studies with dropouts. Simulation studies are conducted with 1000 simulations to produce longitudinal data with Auto-regressive (1) covariance structure and the percentage of missingness is simulated from a Bernoulli distribution from 5% to 30% with 5% increment. We conclude that Multiple imputation method performs better than other methods, whereas last observation carried forward method handles well for small time-interval. Also, Complete Case method includes bias from 15% missingness as well as Mean imputation method holds the accuracy up to 25% missingness. Furthermore, real data from our epidemiological database are analyzed based on these imputation methods to confirm the results obtained from the simulation studies.