Determination of a pilot sample size to determine the sample size of a substantive trial

Obodo, Scholastica C.

Abstract

This research intends to determine the appropriate sample sizes for two-arm pilot studies to help correctly determine the required sample size for the corresponding substantive or definitive trial. Previous research in the area has been reviewed. R programming is used to undertake simulation studies. The first study is an investigation of the procedure proposed by Browne (1995). Results from this study confirm the work of Browne and show that alpha (nominal Type I error rate), beta (nominal Type II error rate), and effectsize do not affect the error associated with estimation of sample size using the method. Desired coverage has a moderate effect, and the pilot sample size has a big effect on the error associated with the predicted sample size when using Browne’s formula. In general, the approach of Browne (1995) is valid but gives very large incorrect estimates. For this reason, the Goldilocks (“just about right”) approach was developed from quantifying the degree of underestimation and overestimation and this new approach may be used to guide researchers in controlling the excess margin. Study 2 compared the performance of four methods of sample size estimation when the Minimum Clinical Importance Difference (MCID) is unknown. The results showed that Browne’s method does not work effectively when MCID is unknown. In these situations, the use of Hedge’s correction was the best of four formulae but still led to overestimation. Hence the need for Study 3 where new methods are proposed using upper confidence limit of the coefficient of variation which was presented using examples and simulations. Using this new approach, the parameter had similar impact in error margin as in Browne’s method and performed well for large effectsize only. The multiplier critical table was developed in study 4 however it results in big sample sizes and therefore would not be used in practice.

• Thesis Type: Thesis
• Deposit Date: Sep 30, 2023
• Publicly Available Date: Jul 25, 2024
• Public URL: https://uwe-repository.worktribe.com/output/11145301
• Award Date: Jul 25, 2024