Simplification or simulation: Power calculation in clinical trials

Background and objectives: A justifiable sample size is essential at trial design stage. Generally this task is completed by forming the main research question into a statistical procedure and then implementing the published formulae or software packages. When these standard statistical formulae/software packages become unavailable for studies with complex statistical procedures, some statisticians choose to fill this gap by assuming an alternative simplified sample size calculation. Monte Carlo simulations can also be deployed, particularly for complex trials. However, it is still unclear on how to determine the appropriate approach under certain practical scenarios. Methods: We adopted real clinical trials as examples and investigated on simplification and simulation-based sample size calculation approaches. Results: Compared to simplified sample size calculation, the simulation approach can better address the non-ignorable impact of baseline/follow-up outcome correlation on study power. For studies with multiple endpoints and multiple co-primary endpoints, the sample sizes calculated by simplification approach should be scrutinized. Conclusions: Directly using the simplification approach for sample size calculation should be restricted. We recommend to utilize the simulation approach, particularly for complex trials, at least as a sensitivity checking and a useful triangulation to the simplification approach outlined.