Background: Over the last few years, many efforts have been made to leverage historical information in clinical trials. Incorporating historical data into current trials allows for a more efficient design, smaller studies, or shorter duration and may potentially increase the relative amount of information on efficacy and safety. Despite these advantages, it is crucial to select external data sources appropriately to avoid introducing potential bias into the new study. This is where borrowing methods become useful. We illustrate and compare the latest methods of borrowing historical data in a single-arm phase II clinical trial setting, examining their impact on statistical power and type I error. Methods: We implemented static and dynamic versions of the power prior method, incorporating overlapping coefficient and loss functions and meta-analytic predictive priors. These methods were compared with standard and pooling approaches, in which none or all historical data are used. Results: Dynamic borrowing methods achieve lower type I error inflation than pooling. The power prior approach, integrated with overlapping coefficient, allowed for measuring the similarity of the subjects considering their baseline characteristics, thus the likelihood of the data contains information about both confounders and outcome. Using a discounting function to estimate the power parameter guarantees the similarity of historical information and current trial data. Conclusion: We provided a comprehensive overview of borrowing methods, encompassing frequentist and Bayesian approaches as well as static and dynamic technique, to guide researchers in selecting the most appropriate strategy.
Comparison of Borrowing Methods for Incorporating Historical Data in Single-Arm Phase II Clinical Trials
Berchialla, Paola
2024-01-01
Abstract
Background: Over the last few years, many efforts have been made to leverage historical information in clinical trials. Incorporating historical data into current trials allows for a more efficient design, smaller studies, or shorter duration and may potentially increase the relative amount of information on efficacy and safety. Despite these advantages, it is crucial to select external data sources appropriately to avoid introducing potential bias into the new study. This is where borrowing methods become useful. We illustrate and compare the latest methods of borrowing historical data in a single-arm phase II clinical trial setting, examining their impact on statistical power and type I error. Methods: We implemented static and dynamic versions of the power prior method, incorporating overlapping coefficient and loss functions and meta-analytic predictive priors. These methods were compared with standard and pooling approaches, in which none or all historical data are used. Results: Dynamic borrowing methods achieve lower type I error inflation than pooling. The power prior approach, integrated with overlapping coefficient, allowed for measuring the similarity of the subjects considering their baseline characteristics, thus the likelihood of the data contains information about both confounders and outcome. Using a discounting function to estimate the power parameter guarantees the similarity of historical information and current trial data. Conclusion: We provided a comprehensive overview of borrowing methods, encompassing frequentist and Bayesian approaches as well as static and dynamic technique, to guide researchers in selecting the most appropriate strategy.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
