Designing for Intent-to-Treat
- PDF / 566,342 Bytes
- 8 Pages / 504 x 720 pts Page_size
- 23 Downloads / 251 Views
Drug Informarion Joumnl, Vol. 35. pp. 1079-1086. 2001 F’rinted in the USA. All rights reserved.
DESIGNING FOR INTENT-TOWTREAT* PHILIPWILLIAMLAVORI, PHD Director, VA Cooperative Studies Program Coordinating Center, Palo Alto, California, and Professor of Biostatistics, Department of Health Research and Policy, Stanford University, Stanford, California
REE DAWSON, PHD Senior Statistician, Frontier Science Research Foundation, Chestnut Hill, Massachusetts
The principle of analysis by intent-to-treat ( I T T ) serves as the standard basis for design decisions as well as choice of analysis in clinical trials. I T T correctly contrasts the pragmatic consequences of the treatments offered in a study, as long as the study protocol accurately reflects the realities of clinical practice. We identify the study of ongoing treatment for chronic disease as the clinical context that most strains the ITTprinciple. In a placebo-controlled trial of a new drug in patients with a condition for which there are standard treatments, the ethical requirement to “rescue”patients who do poorly, and who might be taking placebo, causes “drop-in”from placebo to a standard treatment. We propose that this problem reflects a lack offit between the standardfixed design and clinical reality, rather than a weakness of ITT We propose that the adaptive nature of clinical decision making should be captured in the design of trials, and we show how the ITT principle can be used in such designs. Key Words: Intent-to-treat; Design; Adaptive treatments
INTRODUCTION THE PRINCIPLE OF ANALYSIS by I’IT requires that all protocol-specified measurements of outcomes experienced by subjects in a randomized clinical trial be analyzed according to the randomization, without regard to the subjects’ level of adherence to the experimental protocol. The principle has been described by Efron (1) as a “flagpole of certainty amidst the chaos,” and for many methodologists it serves as the unquestioned basis for design decisions as well as choice of analysis. It is grounded in randomization,
Resented at the DIA Workshop “Intent-to-Treat Analysis,” June 2000, San Diego, California. Reprint address: Philip W. Lavori, HRP Department, Stanford University School of Medicine, Stanford, CA 94305. E-mail: [email protected]. *Supported by a grant from the National Institute of Mental Health (ROI-MH5 1481) to Stanford University.
and as Tsiatis (2) points out, the I’IT test of the null has the right size, since adherence is irrelevant under the null. As a heuristic principle for statistical decision making, ITT has no serious competitor when it comes to comparing the pragmatic consequences of the treatments offered in a study, as long as the study protocol accurately reflects the realities that will be encountered in the world of clinical practice. Like all principles, however, ITT works better in some contexts than others. When the pattern of adherence to treatment in the trial is not what will be encountered in practice, the pragmatic argument weakens. Furthermore, in trials c
Data Loading...
