Sample Size Determination

SAMPLE SIZE DETERMINATION

Originally, none of the general-purpose statistical packages contained software to help statisticians determine sample size requirements. As you have seen, sample size requirements are important for researchers and pharmaceutical and medical de-vice companies to assess the economic feasibility of a particular study, such as a phase III clinical trial for establishing the efficacy and safety of a drug.

To fill this void, Janet Elashoff of UCLA and the Cedars Sinai Medical Center wrote a small-business innovative research proposal to develop such software. The result was a statistical package called nQuery Advisor. This highly innovative prod-uct provided useful and correct results for a variety of important sample size esti-mation problems, a very user-friendly interface, and verbal interpretations for the resulting tables. The tool was so successful that a company, Statistical Solutions, decided to market the software. Now in version 4, the product has undergone sever-al improvements since its introduction.

nQuery Advisor now has several competitors. Some of the competitors that pro-vide sample size determination include StatXact, UnifyPow, Power and Precision, and PASS 2000. Version 4 of StatXact introduced sample size determination for ex-act binomial tests. Version 5, which is much more extensive, includes multinomial tests and a more user-friendly menu for the sample size options.

The SAS Institute is planning to produce a sample size estimation package and may buy the rights to UnifyPow. Chernick and Liu (2002) compare these various packages with respect to the way they determine the power function for the case of the single proportion test against a hypothesized value.

We list the user manuals for these products in the reference section of this chap-ter. Each product has its own web site. They are as follows: www.cytel.com for StatXact, www.ncss.com for PASS 2000, www.statsolusa.com for nQuery Advisor, www.PowerAnalysis.com for Power and Precision, and www.bio.ri.ccf.org/Unify-pow/ for UnifyPow.

Other sample size packages are PASS 2000 by NCSS, EaSt by Cytel, and S + SeqTrial by Insightful. These three packages are designed to handle what are called group sequential designs. Sequential methods are a special topic in statistics that is not within the scope of this text.

Group sequential designs allow the sample size to depend on the results of inter-mediate analyses. We mention them here because the fixed sample size problems that we have been discussing in the text are special cases of the more general problem of sample size estimation. The group sequential software can be made to solve fixed sample size problems by setting the number of interim stages in the design to 1.