Flexible Sample Size Considerations Using Information-Based Interim Monitoring
- PDF / 1,168,530 Bytes
- 18 Pages / 504 x 720 pts Page_size
- 25 Downloads / 183 Views
Drug Informarion Journal, Vol. 35, pp. 1095-1112, 2001 Printed in the USA. All rights reserved.
FLEXIBLE SAMPLE SIZE CONSIDERATIONS USING INFORMATION-BASED INTERIM MONITORING CYRUSR. MEHTA Cytel Software Corporation, Cambridge, Massachusetts
ANASTASIOS A. TSIATIS North Carolina State University, Raleigh, North Carolina
At the design phase of a clinical trial the total number of participants needed to detect a clinically important treatment difference with suflcient precision frequently depends on nuisance parameters such as variance, baseline response rate, or regression coeflcients other than the main effect. In practical applications, nuisance parameter values are often unreliable guesses founded on little or no available past history. Sample size calculations based on these initial guesses may, therefore, lead to under- or over-powered studies. In this papel; we argue that the precision with which a treatment effect is estimated is directly related to the statistical information in the data. In general, statistical information is a complicated function of sample size and nuisance parameters. Howevel; the amount of information necessary to answer the scientific question concerning treatment difference is easily calculated a priori and applies to almost any statistical model for a large variety of endpoints. It is thus possible to be flexible on sample size but rather continue collecting data until we have achieved the desired information. Such a strategy is well suited to being adopted in conjunction with a group sequential clinical trial where the data are monitored routinely anyway. We present several scenarios and examples of how group sequential information-based design and monitoring can be carried out and demonstrate through simulations that this type of strategy will indeed give us the desired operating characteristics. Key Words: Sample size reestimation; Information-based design and monitoring; Adaptive design
INTRODUCTION AT THE DESIGN PHASE of a randomized clinical trial the total number of participants needed to achieve a certain level of significance and power frequently depends on nuisance parameters such as variance, baseline response rate, or regression coefficients other than the main effect. In practical applications, nuisance parameter values are often unreliable guesses founded on little or no available past history. As a result, if the initial guesses for Reprint address: Cyrus R. Mehta, Cytel Software Corporation, 675 Massachusetts Ave., Cambridge, MA 02139.
109.5
Downloaded from dij.sagepub.com at University of Texas Libraries on May 1, 2015
Cyrus R. Mehta and Anastasios A. Tsiatis
1096
the nuisance parameters are far from the truth, then the study may be under- or over-powered to detect the desired treatment difference. For example, in a clinical trial comparing dichotomous responses, suppose the investigators want to design the study to detect a 10% difference in the response rate between a new treatment and a control treatment with 90% power using a test at the 0.05 (two-sided)
Data Loading...
