Sample Size Estimation when Comparing More than Two Treatment Groups
- PDF / 426,309 Bytes
- 7 Pages / 504 x 719.759 pts Page_size
- 61 Downloads / 279 Views
0092-861 5/98 Copyright 0 1998 Drug Information Association Inc.
Printed in the USA. All rights reserved.
SAMPLE SIZE ESTIMATION WHEN COMPARING MORE THAN TWO TREATMENT GROUPS ALANPHILLIPS, BSc, PHD Wyeth Research (UK), Berkshire, United Kingdom
A large amount of literature has been published on methods for determining sample size and powerfor clinical trials. Althoughformulae exist to determine sample size for different designs, in practice the methodr are rarely used. Instead, sample size estimates are usually based on simple formulae derived for the comparison of two means or binomial proportions. This paper describes some useful theory for sample size estimation when comparing more than two treatment groups. Where the theory is not applicable, it is demonstrated that simulation can be used as an alternative approach.
Key Words: Sample size; Analysis of variance; Dose-response; Simulation
INTRODUCTION THE BASIS OF THE sample size calculation is usually needed in protocols for clinical trials and when publishing results in respected journals. Although a large amount of literature has been published on the problem of determining sample size and power for different trial designs (1-5), in practice the methods are rarely used. Sample size estimates are usually based on simple formulae derived for comparing two binomial proportions or two means. This paper describes some previously published theory that has practical relevance for sample size estimation when comparing more than two treatment groups. Where the published method is not applicable, it will be shown how simulation can be used as an alternative approach to estimating the sample
Reprint address: Alan Phillips, BSc, PhD, Wyeth Research (UK), Huntercombe Lane South, Taplow, Maidenhead, Berkshire, SL6 OPH, United Kingdom. E-mail : [email protected]
size. In particular, dose-response studies where patients are allocated to either placebo or one of several fixed dose groups will be considered. Using simulation this paper will show that for such studies the sample size calculation needs to take into account the method of analysis and the expected response of each dose being studied. Although throughout this paper only outcome variables which are assumed to be continuous and normally distributed are considered, the issues discussed apply equally to binary and ordered categorical data.
SAMPLE SIZE ESTIMATION FOR TWO GROUPS Sample size estimates for a clinical trial depend on several factors, including but not exclusively the method of analysis, the clinically meaningful difference to be detected (6), the type I error (a)or probability of detecting a significant difference when the treatments are really equally effective, and the type I1 error (p) or the probability of not
193
Downloaded from dij.sagepub.com at NANYANG TECH UNIV LIBRARY on May 21, 2015
Alan Phillips
194
detecting a significant difference when there really is a difference of magnitude (6) and level of statistical significance. For instance, consider a six-week randomized, doubleblind
Data Loading...
