Comparison of Missing Imputation Methods for Combining Multiregional Cohort Studies

• Activity Number: 270
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 2:45 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ayano Takeuchi*
• Companies: Keio University
• Keywords: missing ; imputation ; multi-regional cohort

Abstract

In recent decades, there are many cohort studies with similar purpose have been planned and conducted in many regions or countries. Sometimes, researcher try to harmonize and make consortium. When the collaboration has worked well, each researchers bring their data and integrate it together. In these case, many items and evaluation procedures does not match because their study protocol are planned independently. If item can be integrated directly or through the conversion process, percentage of item or bio specimen missing also vary across cohorts. We compare missing imputation methods for many situation at these kind of cohort consortium. Situation settings are about regional differences and missing percentage. We compare single imputation and multiple imputation by each cohort or by whole combined cohorts with and without considering regional differences using simulation data. When there are strongly difference between regional differences, multiple imputation by each cohort works best in bias point of view. When regional differences are not so large, multiple imputation by whole cohort with consideration of region have similar performance as imputation by each cohort.