In response to concerns about data quality in real-world evidence (RWE) generation, including issues such as bias and small sample sizes, resulting in low precision estimates with questionable accuracy and thus interpretability challenges, regulatory submissions have increasingly incorporated advanced methodologies to enhance the robustness of RWE.

Among these methods, Bayesian borrowing stands out as an approach that can significantly increase the scientific potential of real-world data. By leveraging data from multiple sources that may all have different weaknesses, Bayesian borrowing can combine these and enhance the power of comparisons with trial data for comparisons beyond those from a randomized control trial. Bayesian borrowing can also be used to create hybrid control arms, enabling a smaller control cohort to address ethical concerns and patient availability issues.1

The Bayesian borrowing concept

Bayesian borrowing methods make use of external data, potentially from multiple sources, by using a prior distribution that adjusts for the possibility that this external data may come from a different population. While using external or historical data can enhance the precision and accuracy of parameter estimates in a study, directly simple pooling of this data could lead to bias if the external population differs from the current one.2,3,4 To address this, priors such as a power prior is used to adjust the influence of the external data, which is more diffuse than complete pooling of current study dataset and the external dataset, reducing the possible bias but also the eventual precision of the parameter estimate.

In drug development, Bayesian borrowing is primarily applied in situations involving rare diseases, pediatric trials, or when there are no existing approved treatments for the same conditions.5

Figure 1. Bayesian borrowing

Quantitative bias analysis (QBA) plays a crucial role in supporting studies that employ Bayesian borrowing by assessing the impact that the weaknesses in the data being integrated has on study results. When leveraging external or historical data through Bayesian methods, such as Bayesian borrowing, there is always a risk that the borrowed data may introduce bias due to elements that cannot be addressed directly in analysis specifications, such as missing or unmeasured data, or other quality issues. QBA helps to quantify the extent of these biases and provides a structured approach to adjust for them, thereby enhancing the interpretation possibilities of the results, ultimately supporting study validity and scientific integrity.

By applying QBA alongside Bayesian borrowing, researchers can transparently account for uncertainties in the borrowed data and ensure that the final estimates are more robust, credible, and defensible in both regulatory and clinical decision-making contexts.

Figure 2. Example of QBA for Bayesian borrowing

FDA and HTA submissions incorporated with Bayesian borrowing methods

In recent years, the acceptance of Bayesian borrowing approaches has been evolving from both regulatory and Health Technology Assessment (HTA) perspectives.

The FDA has highlighted this shift through initiatives like a podcast discussing the use of Bayesian statistics, including a case where Bayesian methods were used to borrow data from an adult trial to assess an asthma product’s treatment effects in pediatric patients.6 Additionally, the FDA recommended that GSK apply Bayesian dynamic borrowing to integrate adult trial data for a pediatric study for post-marketing activities, and these results were subsequently accepted.7

HTA bodies are also considering Bayesian methods; for example, NICE recommended using Bayesian hierarchical models, which are closely related to Bayesian borrowing, in the technical appraisal of larotrectinib for NTRK-fusion positive solid tumors in 2020.8

Furthermore, the FDA plans to release draft guidance on the use of Bayesian methods in clinical trials for drugs and biologics by the end of 2025.

The future of Bayesian borrowing

Although Bayesian methods have garnered increasing attention from regulatory and HTA bodies, their practical implementation has been somewhat limited. Challenges such as organizational resistance to novel approaches, resource constraints, and difficulties in applying these advanced methods effectively can hinder their adoption in regulatory and HTA submissions. However, as awareness grows and best practices are established, these barriers are likely to diminish, paving the way for more widespread use of Bayesian methods.

Notes

1 Dron, L., Golchi, S., Hsu, G., & Thorlund, K. (2019). Minimizing Control Group Allocation in Randomized Trials Using Dynamic Borrowing of External Control Data – An Application to Second Line Therapy for Non-Small Cell Lung Cancer. Contemporary Clinical Trials Communications, 16(1).

2 Viele, K., Berry, S., Neuenschwander, B., Amzal, B., Chen, F., Enas, N., Hobbs, B., Ibrahim, J. G., Kinnersley, N., Lindborg, S., Micallef, S., Roychoudhury, S., & Thompson, L. (2013). Use of Historical Control Data for Assessing Treatment Effects in Clinical Trials. Pharmaceutical Statistics, 13(1).

3 Struebing, A., McKibbon, C., Ruan, H., Mackay, E., Dennis, N., Velummailum, R., He, P., Tanaka, Y., Xiong, Y., Springford, A., & Rosenlund, M. (2024). Augmenting External Control Arms Using Bayesian Borrowing: A Case Study in First-Line Non-Small Cell Lung Cancer. Journal of Comparative Effectiveness Research, 13(5).

4 Mackay, E. K. & Springford, A. (2023). Evaluating Treatments in Rare Indications Warrants a Bayesian Approach. Frontiers in Pharmacology, 14(1).

5 Muehlemann, N., Zhou, T., Mukherjee, R., Hossain, M. I., Roychoudhury, S., & Russek‑Cohen, E. (2023). A Tutorial on Modern Bayesian Methods in Clinical Trials. Therapeutic Innovation & Regulatory Science, 57(1).

6 Clark, J. (2023). Using Bayesian Statistical Approaches to Advance our Ability to Evaluate Drug Products. CDER Small Business and Industry Assistance Chronicles, U.S. FDA.

7 Best, N., Price, R. G., Pouliquen, I. J., & Keene, O. N. (2021). Assessing Efficacy in Important Subgroups in Confirmatory Trials: An Example Using Bayesian Dynamic Borrowing. Pharmaceutical Statistics, 20(1).

8 NICE. (2020). Appraisal Consultation Document: Larotrectinib for Treating NTRK Fusion-Positive Solid Tumours.

Share Via