gms | German Medical Science

Introduction: Which variables should we include in the regression analysis as adjusting variables? Using which criteria should we define the binary endpoint? These and more are questions we as statisticians often have to discuss when consulting clinicians who want to plan or analyse clinical trials or observational studies. In clinical trials, those decisions have to be made during writing of the study protocol before collection of any data or at latest during formulating the statistical analysis plan before database lock. In observational studies those choices are often made after data collection. We aim to systematically investigate the implications of analysis choices on study results in the sense of a robustness analysis using one clinical trial and one cross-sectional study as examples.

Methods: In the cross-sectional study we examine the influence of covariate sets to include in the regression analyses for adjustment. In the clinical trial we explore the impact of derivation of binary endpoints from numerical values. We inspect the implications on study results by comparing estimators and confidence intervals for all possible combinations of potential analysis choices.

Results: Depending on the respective study and hypothesis, we aim to combine the different results to obtain one answer to the initial medical question or to show that such a combination is not sensible. In addition, we visualize all results in a joint manner using an interactive R Shiny app.

Discussion: The intuitive visualization enables clinicians and collaborators to understand the potential implications of analysis choices on study results and their influence on study interpretation. However, measures have to be taken to prevent malpractices like selective reporting or hypothesizing after results are known.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.

This contribution has already been published: DAGStat 2022