Article
Sample size recalculation using partial data – what is the nature of the benefit?
Search Medline for
Authors
-
Silke Jörgens
- Janssen-Cilag, Neuss, Germany
| Published: | February 26, 2021 |
|---|
Outline
Text
Background: Many adaptive clinical trials suffer from an imbalance between recruitment rate and duration of individual study participation: The faster the recruitment rate and the longer the individual study duration, the higher will be the percentage of incomplete subjects at the planned time point of an interim analysis. Many suggestions have been made on how to incorporate the data of these so-called pipeline subjects into interim decision making. Often this is done by imputing missing data points based on early readouts of the primary endpoint using parametric modelling or other approaches, like MMRM modelling. The imputed data can then be used to enhance and complement the information available from complete subjects, e.g. by contributing to sample size recalculations based on conditional power.
Methods: Methods used for the adaptive analysis of such a trial which are based on partitions of the data set (e.g., stagewise p-value combination tests) need to make allowance for including incomplete patients. These neither provide final endpoint data for the first stage nor can they be analyzed as second stage subjects if their partial data were used to modify the trial. Different approaches are conceivable.
Intuitively, such a procedure incorporating additional data could be expected to increase the power of a study. However, sample size recalculations based on interim effect estimates often overshoot because of their truncated distribution and therefore the actual power is frequently found to be above the targeted one. Therefore, a more reasonable expectation is an increased precision of the new sample size estimate.
Conclusion: Procedures are illustrated using a case study. The amount of precision gained by including partial data is investigated, and we also describe at which point the additional uncertainty introduced by the imputation outweighs that amount of precision.
The authors declare that they have no competing interests.
The authors declare that an ethics committee vote is not required.
