A Note on Stratification in Clinical Trials
- PDF / 57,488 Bytes
- 4 Pages / 612 x 792 pts (letter) Page_size
- 46 Downloads / 214 Views
Thomas Permutt, PhD Director, Division of Biometrics II, Center for Drug Evaluation and Research, Food and Drug Administration, Silver Spring, Maryland
Key Words Stratification; Interaction; Analysis of covariance; Imbalance; Adjustment Correspondence Address Thomas Permutt, Division of Biometrics II, Center for Drug Evaluation and Research, Food and Drug Administration, 10901 New Hampshire Ave., Bldg. 22, Room 3105, Silver Spring, MD 20993-0002 (e-mail: thomas.permutt@ fda.hhs.gov). The views expressed are those of the author and not necessarily of the Food and Drug Administration.
A Note on Stratification in Clinical Trials
INTRODUCTION “Baseline condition [or some other covariate] may have an important effect on outcome, so the trial should be stratified by baseline condition.” This statement is often heard at meetings to discuss the design of randomized clinical trials. It is usually met with general assent. The first part of the statement, however, may refer to one of three problems, and the second part to one of four solutions. There is rarely general agreement about what the problem is or which solution is being proposed. As a result, the agreed solution may often fail to solve the chief problem.
THE PROBLEMS AND THE SOLUTIONS The three problems to be discussed are (1) interaction, (2) interaction with a small stratum, and (3) stratum main effects. The four solutions are (a) stratified analysis, (b) stratified allocation, (c) stratified recruitment, and (d) adjusted analysis. INTERACTION The first problem to be discussed is stratum-bytreatment interaction. The treatment may have different effects on different groups of patients. Women may respond differently than men. Elderly patients may be more or less sensitive to
719
Stratification is sometimes proposed to deal with problems of influential covariates in clinical trials. The word stratification, however, may refer to any of four different methods of design and analysis. The methods are capable of addressing three different problems. Which problem and which method are being discussed is often misunderstood. Consequently, the method adopted may not solve the problem that provoked its consideration.
the drug. Pediatric patients may have a different form of the disease than adults. In extreme cases, we might consider evidence of efficacy in one group to have no relevance at all to another group. In other cases, we might find the concept of “overall” efficacy meaningful, but we might still have concerns about how the effects in different groups vary from the overall measure. To cope with interaction, stratified analysis is required. We must look at the effects by stratum and see whether they are the same or different. In the end, of course, we might decide that they are nearly enough the same and choose to represent them all by a pooled analysis, but the separate analyses are required first. Stratified allocation may also be used in this situation, but it cannot take the place of separate analysis, and it may be less important than is often believed. It
Data Loading...
