Submitted by A.S. Quenault on Tue, 02/12/2025 - 15:49
A new modelling technique has been developed by BSU researchers to improve the reliability of clinical trials.
Basket trials are a type of clinical study where the effect of a single treatment is tested across multiple groups of patients, referred to as “baskets”, with each group defined by a specific disease or condition. This trial design has been developed in response to precision medicine, which aims to target treatments to patients’ intrinsic factors as opposed to a disease type as a whole. As such, all patients in a basket trial require something in common (often a genetic mutation) that the treatment is targeting. By testing the treatment across multiple diseases, the usually lengthy drug development process is expedited, improving trial efficiency.
Basket trials also allow for testing treatments on small patient populations, such as those diagnosed with rare diseases. However, this poses substantial statistical challenges, as limited data can make it difficult to reliably estimate how well the treatment works. This can lead to misleading results such as claiming a treatment is effective when it actually is not, while also making it harder to detect real effects.
To address these challenges, Libby Daniells from the MRC Biostatistics Unit and colleagues, implemented a technique known as information borrowing, which allows data from one basket to help inform the results in another, but only when the data suggests it is appropriate to do so. This can improve the reliability of estimates, while still accounting for differences between the individual baskets.
Another way to improve reliability of results is to use historical data from previous studies. Often the treatment has been tested in at least one of the diseases under investigation in previous clinical trials. Therefore, data from previous studies can be incorporated into the analysis of the current study to strengthen the evidence of the treatment’s effect.
The paper, published in Biostatistics, explores methods that combine both approaches: borrowing information between baskets in the current trial while also incorporating relevant historical data. The proposed models adjust how much information is borrowed based on how similar the data sources are, so that only similar data contributes to the estimates of the treatment effect. The performance of the approaches is assessed through numerical studies, that simulate what would happen in a real-life clinical trial. The results of the studies demonstrate that the proposed model can more reliably identify promising treatments and improve precision of estimates when the historic data aligns with the ongoing trial. When the historic data is very different, the proposed model performs similarly to standard methods that ignore historical data.
First author and Research Associate at the BSU, Libby Daniells, said:
“Basket trials frequently involve very small patient groups, making every piece of information matter. Developing these models that amalgamate different forms of borrowing took time and patience, but seeing how they improved the reliability of trial decisions made it all worthwhile.”