Statistical Method to Analyze Longitudinal Clinical Trial Incomplete Data

• Activity Number: 97
• Type: Contributed
• Date/Time: Monday, August 4, 2003 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Ohidul Siddiqui*+ and Hsien-Ming James Hung
• Companies: Food and Drug Administration and Food and Drug Administration
• Keywords: mixed model ; ignorable/nonignorable missing ; repeated measure ; endpoint analysis

Abstract

In randomized clinical trials, patients are often followed longitudinally. That is, each patient has multiple measurements on the scale of interest. Although each patient has multiple measurements on the same scale, the common practice is to evaluate treatment efficacy at the study endpoint measurement data only. One problem in the endpoint data analysis is that planned data are not available for some patients because of missing data due to dropout of the patients before completing the study period. In such situations, common practice is to impute the endpoint missing data using the last observation carried forward approach. An alternative approach is to do longitudinal modeling on the available data, and, based on the fitted models, compare and test the estimated treatment group differences. Longitudinal modeling can be carried out using likelihood-based repeated measure analyses, as well as using Generalized Estimating Equation analyses. Statistical inferences using each of these analysis approaches are valid under certain missing data mechanisms. A comparison among these analysis approaches is made in simulated data under different missing data mechanisms.