A Note on Missing Data in Noninferiority Trials
- PDF / 6,994,179 Bytes
- 6 Pages / 612 x 792 pts (letter) Page_size
- 48 Downloads / 228 Views
DATA
469
A Note on Missing Data in Noninferiority Trials
P o d Gollo Novariis
Christy Chrong.Stein Pfizer
Key Words Intent to treat; Per-protocol population; Imputation: Assay sensitivity; Active control trial; Noninferiority margin Correspondence Address Paul Gallo. Novartis. One Health Plaza, East Hanover. N] 07936 (email: paul.gallo~novartis.com).
Two important topics associated with the design and analysis of clinical trials, missing data and noninferioriQ, are currently receiving much consideration in the literature, in scientificforums, and in health authority interactions. As methods and practices for these topics are evolving, understandably most of the discussion and guidance have focused on them individually.Howevm in considm’nghow best to address these two issues from a standpoint of design or analysis within a particular
INTRODUCTION Strategies for statistical analysis of clinical trials in the presence of missing data have received a great deal of interest and discussion in recent years. For a variety of reasons, clinical trials are prone to nonavailability of some subset of planned outcome measurements. Missing data tend to occur with higher frequency in studies of long duration, or in trials where the outcome assessments are complex or invasive. It has long been accepted that analyses that include only patients with complete data often lead to biased inferences. Such biases may arise from the unaccounted-for relationship of “missingness” to treatment or outcome. In addition, excluding patients with missing responses will generally violate randomization and intention-to-treat principles. Furthermore, completers may no longer constitute a representative sample of the subject population enrolled into a study. Relatively simplistic approaches, such as worst-case or best-case imputation methods, or which utilize last available endpoint readouts for each patient carried forward (eg, last observation carried forward, LOCF), have historically played a large role in the analyses of trials with missing data. Among the reasons for their use is likely the ease of implementing such approaches. While by no means are such analyses im-
trial in which both are relevant, it is helpfil to consider them jointly. For example, how one develops an analysis plan to accommodate missing data in a noninfen’ority study may, or may not, be identical to the approach for a superiority trial. Here we describe how these issues might interact in a given trial; that is, we raise some specific points relevant to both of these topics that may be particulariy important to consider when both issues are present in the same trial.
mune to bias, they have been accepted in cases where tendencies toward bias would be expected to operate in a conservative direction. Lately, “statistically principled” approaches (I), such as multiple imputation procedures, likelihoodbased methods, and mixed-model longitudinal approaches, have received increased attention. These approaches generally have been found to perform favorably in methodological
Data Loading...
