Basket Trials in Gynaecology: A Simulation Study Comparing Novel Designs with Current Practice

Background Basket trials provide a framework for evaluating an intervention across multiple subtrials within a single protocol. By enabling information sharing, they may improve efficiency and reduce reliance on small, underpowered studies. In gynaecological conditions, we often see patients with the same condition presenting with different symptoms. Although the symptoms are different, treatment is often the same, leading to small and often statistically underpowered trials. Master protocols such as basket trials have the potential to reduce waste in research by allowing the evaluation of a single intervention in multiple disease subtypes. In a hypothetical gynaecological basket trial, this would allow the inclusion of patients with the same underlying condition, but with differing most bothersome symptoms, to be included in the same randomised basket trial. In this article, we compare the use of basket trials in gynaecology to typical approaches to trial design and analysis. Methods Our simulation study evaluates the performance and utility of basket trial designs in hypothetical gynaecological settings. We compare information-sharing strategies treatment effect borrowing (TEB) and treatment response borrowing (TRB) to strategies closely emulating typical gynaecological trial designs: no borrowing (NB, analogous to separate randomised controlled trials) and complete pooling (CP, all patients analysed within a single trial). Performance was measured by key operating characteristics, including coverage of credible intervals, power, credible interval (CI) width, bias, mean squared error (MSE), and empirical standard error. We explore the impact of sample size, treatment effect heterogeneity and control group response, evaluating the balance between efficiency and the risk of misleading inference. Results Coverage was approximately 95% across all scenarios and methods. CP produced suboptimal coverage under treatment effect heterogeneity and yielded the narrowest CIs in homogeneous scenarios, while NB produced the widest. NB minimised bias under high heterogeneity but incurred the largest MSE, whereas CP exhibited the greatest bias when treatment effects differed across subtrials. TEB generally outperformed TRB in bias and MSE when control arm heterogeneity was present, whereas CP resulted in the highest bias. Conclusion Adaptive borrowing offers a more robust compromise compared to No Borrowing (independent trials) and Complete Pooling (a single large trial). We conclude cautiously that information borrowing offers efficiency gains, but basket trials require clear clinical and biological justification.