Sensitivity Analysis of Missing Data: Case Studies Using Model-Based Multiple Imputation
- PDF / 11,941,808 Bytes
- 10 Pages / 612 x 792 pts (letter) Page_size
- 16 Downloads / 179 Views
DATA
475
Sensitivity Analysis of Missing Data: Case Studies Using Model-Based Multiple Imputation
'"
Novartis Pharmaceutical Corporation
Key Words Missing data; Model-based multiple imputation; Sensitivity; Clinical trial Correspondence Address Jie Zhang, Novartis Pharmaceutical Corporation, One Health Plaza. 405-3083,East Hanover, NJ 07936- 1080 (email: [email protected]).
When undertaking confirmatory analyses of data ji-om clinical trials, statisticians fiequently are confionted with having to assess and address potential biases introduced by missing data. New developments for handling missing data have prdifcated in the litemfure. Sensitivityanalysis, which allows the assessment of the impact of a wide range of nonignorable missingness mechanisms on the robustness of the statistical results, provides a
INTRODUCTION Despite careful planning, missing data will be an issue in practically all clinical trials. It is critical to assess and address any potential bias introduced by missing data, especially in confirmatory trials such as pivotal phase 3 trials in drug development. Based on the missingness mechanism, missing data are often classified into three categories (1).Data are said to be missing completely at random (MCAR) when the missing data are independent of observed and unobserved data. Data are missing at random (MAR) when the missing data only depend on the observed data, and the missing data are missing not at random (MNAR) when they also depend on the unobserved data. If missing data are MAR and the parameters of the measurement process and the missing data process are distinct, the missing data mechanism is said to be ignorable in the framework of likelihood-basedinferences. In addition to the validity of likelihood-based analysis under the MAR assumption, it can be reasonable to assume MAR in some situations. Therefore, MAR may provide a sensible middle ground for analysis of missing data and can be a choice for main analysis in the study protocol after careful consideration. For example, for continuous endpoints in longitudinal clinical trials, the likelihood-based mixed-effects model repeated measures (MMRM) approach has received con-
reasonable alternative for analyzing trials with missing data. Two case studies utilizing sensitivity analyses in pharmaceutical industry clinical tials are presented. The first is based on an Alzheimer disease trial with a time-to-event endpoint, and the second is from an osteoporosis trial with a repeated binary outcome. The practical issues associated with the application of sensitivity analysis are discussed as well.
siderable attention in the literature and has started to be accepted in regulatory filing as the main analysis (2,3).The log-rank test or Cox proportional hazard model has been widely accepted in analyzing survival data to address censoring data under the similar assumption of MAR (4).Generalized liner mixed models and weighted generalized estimating equation under the assumption of MAR are proposed to analyze repeated categorical data (5
Data Loading...
