Natural Sciences
Life Sciences
Scientific Computing
Scientific Computing

Georg Zimmermann, Spinal Cord Injury and Tissue Regeneration Centre Salzburg, Paracelsus Medical University and Department of Mathematics, Paris Lodron University, Salzburg, Austria

ATV, Seminar Room, Im Neuenheimer Feld 242

Heidelberger Kolloquium Medizinische Biometrie, Informatik und Epidemiologie

Consider the situation of comparing two groups of patients with respect to a univariate outcome of interest, adjusting for one or several covariates. If the outcome variable is continuous, the adjusted group means are usually compared by using the analysis of covariance (ANCOVA) approach. The case of random covariates is of particular interest, because adjustments for baseline measurements are strongly recommended by regulatory agencies [1]. Moreover, controlling for additional variables that are supposed to be correlated with the outcome could reduce the bias of the effect estimators and increase the inferential power. However, methods for sample size calculation are only available for ANCOVA models with a single random covariate [2]. Therefore, we consider the case of a univariate ANCOVA model with multiple random covariates and possibly unequal group sizes. We derive an asymptotic sample size formula and propose some finite-sample adjustments. We evaluate the accuracy of our proposed sample size formulas in an extensive simulation study and discuss an internal pilot study design, which allows for blinded sample size recalculation and could thus be a potentially appealing alternative to the fixed sample size approach.

Event data:
Import event data into Outlook Calendar