Consideration of missing values in sample size calculation and re-calculation for clinical trials (COMIVA)

• Applicant: Dr. Marietta Kirchner
• Subject Area: Epidemiology and Medical Biometry/Statistics
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 544670723

Determination of the appropriate sample size to achieve high probability to detect a clinically relevant treatment effect is a crucial aspect when planning a clinical trial. Both, the choice of a too small or a too high sample size is to be criticized for ethical, scientific, and economic reasons. A problem that frequently occurs is that some data are missing so that outcome data important for the statistical analysis were not collected. Missing data may lead to a power decrease as the sample size of the dataset used for statistical analysis is smaller. Therefore, the possibility of missing data has to be considered already during the planning phase of the clinical trial. A common way to deal with the issue of missing data during sample size calculation is the inflation of the sample size to account for the expected drop-out rate. However, all included patients are to be considered in the analysis based on the intention-to-treat principle. Thus, imputation methods are typically used to replace missing outcome data so that patients with missing data are included in the analysis. As a consequence, treating patients with missing data in the sample size calculation as non-existent for the analysis does not reflect the actual situation. In the present project the following goals are pursued. First, the current practice for considering missing data in sample size calculation is systematically evaluated. Additionally, literature is systematically searched for available methodology for sample size calculation in clinical trials that use imputation methods in the statistical analysis. Second, the potential benefit of using imputation methods in terms of a gain in statistical power (or reduction in required sample size) is assessed with the goal to give recommendations beyond the commonly used inflation of the required sample size. This will be done first for clinical trials with a fixed design (no interim analysis planned) and second extended to trials with internal pilot study design which allows to adjust the initially defined sample size if one becomes aware that assumptions made during the planning stage are not valid. Here, the accumulated data is used to re-estimate the value of so-called nuisance parameters mid-course the trial and to update the initially selected sample size based on the information obtained so far. Finally, the developed methods will be implemented into a validated open-source software package in order to promote its use. To sum up, the present project contributes to the improvement of sample size determination for a clinical trial that uses imputation methods in the primary analysis to replace missing outcome data with the expectation that the required sample size can be reduced.

• DFG Programme: Research Grants