A Simulation of Various Missing Data Imputation Methods in the Application of Composite Endpoint

• Activity Number: 77 - Complex Designs and Composite Endpoints of Medical Device Clinical Studies and Benefit-Risk Analysis of Diagnostic Tests
• Type: Contributed
• Date/Time: Sunday, July 28, 2019 : 4:00 PM to 5:50 PM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Ja-An Lin* and Rajesh Nair and Natasha Sahr
• Companies: FDA/CDRH and CDRH/FDA and St. Jude's Children's Hospital
• Keywords: Composite Endpoint; Medical Device; Missing Data; Imputation

Abstract

Composite endpoints often comes into play in the clinical study, when a single endpoint cannot comprehensively establish the effectiveness or safety profile of the proposed device well. A composite endpoint typically consists of several clinically relevant components. The components can be either from the effectiveness or safety perspective. Meanwhile, the types of component also can be categorical, continuous or survival. However, more data points need to be assessed, often means higher possibility for more missing data. To avoid the estimation bias due to missing data, the missing data imputation is usually suggested. To our understanding, it is not clear yet if the imputation directly to the composite endpoint or to the components instead lead to different conclusion. In this study, we conduct a simulation to investigate this issue with different commonly employed imputation methods. Hopefully the result provides insight to the researchers when adapting composite endpoints and selecting imputation method in the study.