Comparison of Methods for Missing Data Imputations in the Analysis of Radiographic Data from a Large Phase III Trial

• Activity Number: 368
• Type: Topic Contributed
• Date/Time: Wednesday, August 11, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Guowen (Gordon) Sun*+ and Pilita Canete and James Whitmore and Gary Aras and Nick Fotheringham
• Companies: Amgen, Inc. and Amgen, Inc. and Amgen, Inc. and Amgen, Inc. and Amgen, Inc.
• Keywords: clinical trial ; missing data ; simulation ; estimation ; sensitivity analysis

Abstract

The treatments for rheumatoid arthritis have advanced dramatically in the last few years. These treatments are indicated not only for reducing signs and symptoms but also for inhibiting the progression of structural damage in patients with active rheumatoid arthritis. The progression of structural damage due to rheumatoid arthritis is generally slow and a validated index/composite score is commonly used to evaluate the progression based on the radiographs taken at different time-points. Since the index/composite score is based on the radiographs that include more than 80 individual joints in the hands, wrists, and feet, handling missing radiographic joint scores is a challenge. In addition, missing radiographic data would also occur due to missed visits, early withdrawal, or lost to follow-up. These types of missing data pose challenges in estimating the true treatment effect of new therapies. In a large phase III clinical trial, analyses of the radiographic data were performed using four missing data imputation methods (linear interpolation or extrapolation, worst-case imputation, multiple-imputation, and last-observation-carried-forward).