GET THE APP

Differences in Sample Size Requirements of Statistical Methods Involved in Clinical Trials with Baseline Imbalance Demonstrated and Quantified: A Simulation Study | Abstract
Journal of Clinical Trials

Journal of Clinical Trials
Open Access

ISSN: 2167-0870

+44 20 3868 9735

Abstract

Differences in Sample Size Requirements of Statistical Methods Involved in Clinical Trials with Baseline Imbalance Demonstrated and Quantified: A Simulation Study

Bolaji Emmanuel Egbewale

Background/aims: In trials with post treatment assessment of continuous outcome variable, sample size estimations usually do not make use of existing information on levels of baseline imbalance and correlation (r) between pre and post treatment scores of outcome variable. As a result, in trial scenarios where imbalance indeed exists, more or less sample units than the required may have been indicated, creating ethical issues and issues related to efficient trial design. This simulation study aimed at quantifying relative sample sizes required at differing levels of experimental conditions including baseline imbalance for statistical methods of analysis of variance ANOVA, change score analysis CSA and analysis of covariance ANCOVA.

Methods: Overall, 126 hypothetical trials were evaluated, each with data simulated by using several combinations of levels of treatment effect, correlation between pre and post treatment scores, direction and magnitude of baseline imbalance.

Results: Irrespective of both size and direction of baseline imbalance and level of effect to be determined, CSA when compared to ANOVA requires same sample size when r=0.5, requires less sample units when r>0.5 and more sample units when r<0.5. Irrespective of the level of baseline imbalance, depending on the level of correlation, reduction in the required sample size can reach more than 50% of the original for specifying ANCOVA.

Conclusions: Researchers should make use of a-priori specification of correlation in sample size estimations and endeavour to report information on the observed level of correlation in their trials. Such information is crucial to future design of efficient clinical trials.