How innovative trial approaches are advancing rare disease research

Rare disease is a major – and growing – area of clinical research and drug development. Of the FDA’s 46 novel drug approvals in 2025, more than half (26) were designated as rare disease (orphan) treatments.¹ This growth in rare disease research is being driven by several factors, including a commitment to improving the lives of patients, and regulatory policies, including financial incentives.²

However, traditional clinical trial designs are often ill-suited for rare disease research with common challenges including small patient populations, limited historical data, and high variability. In contrast, alternative parsimonious and innovative methods, such as external control arms (ECAs), Bayesian borrowing, and AI-enabled digital twins, use advanced modelling and simulation to strengthen limited data, shorten timelines, and reduce risk.

AI-powered digital twins are an emerging extension of model-based evidence generation, alongside Bayesian borrowing and other methods of data augmentation. In rare diseases, synthetic patients and digital twins can simulate disease progression and treatment response, helping explore trial scenarios and endpoints, supporting control arms, and informing dose selection and study design.

Regulators are increasingly open to such a model-informed drug development (MIDD) approach. Draft ICH M15 guidance recognises the ‘increasingly crucial role’ of MIDD in clinical research while the FDA has said the approach can improve efficiency, increase the probability of regulatory success, and optimise dosing and therapeutic individualisation.^3,4

The need for external control arms

ECAs provide a valuable alternative to standard randomised controlled trials in rare disease, offering comparator data without randomisation. They can accelerate development of new medicines, avoid ethical barriers and potentially improve generalisability to real-world populations. There are three primary methodologies for ECA development. Emulation sees historical trial data or RWD matched and re-weighted to emulate trial populations. Simulation uses predictive modelling approaches to generate trial outcomes based on population and intervention specifications. Synthesis uses generative AI and deep learning (DL) algorithms to produce realistic patient cohorts for trial stimulations.

For example, refractory precursor B-cell acute lymphoblastic leukaemia is a rare and aggressive cancer. A new treatment was granted accelerated approval based upon findings from a single-arm, open-label Phase II study supported by a historical control arm and ECAs using summary-level outcome estimates from previous trials.⁵ This innovative approach provided the robust evidence needed for regulatory approval and enabled faster patient access.

In another compelling example, low perinatal HIV transmission rates and evolving standards of care made randomised trial recruitment difficult. Real-world data (RWD) and historical data were combined to create an ECA within a Bayesian adaptive trial. This reduced the sample size fivefold and accelerated trial completion, influencing WHO guidelines and contributing to the eradication of perinatal HIV transmission in targeted regions of South-East Asia.⁶

When using ECAs, it is important to remember that methodological planning and critical reporting are essential. Guidance from the FDA and MHRA emphasises the importance of both planning and reporting to generate actionable and robust results.^7,8

Data augmentation and Bayesian borrowing

Data augmentation, particularly Bayesian borrowing, enables integration of historical trial data and RWD while maintaining statistical integrity. Bayesian trial designs can require 30-2,400% fewer participants than frequentist models, increasing efficiency, shortening timelines, and improving the likelihood of meaningful outcomes for paediatrics or rare disease patients.^9,10

An example of the value of data augmentation, comes from a Bayesian dynamic borrowing (BDB) study to support the approval of a drug in China leveraging trial data generated in western countries. Chinese patients had not been included in the initial global study to support drug approval. A BDB approach enabled bridging of the global study data to the new region and provided a safety valve to minimise the impact of inconsistent results between external and new study data. This rapidly generated robust evidence without the need for a fully powered standalone trial, accelerating medicine availability.¹¹

Synthetic patients and AI-enabled digital twins

Finally, AI-enabled digital twins (DTs), patients simulation, or synthesis can project treatment effects on individual patients or specific patient populations, but from short to longer term, taking into account varied characteristics and continuously updating with real-time data. This can reduce sample size requirements and trial duration while maintaining robust evidence and optimising medical practice.

For example, a study exploring the efficacy of a next-generation enzyme replacement therapy for the rare, progressive condition Pompe disease used DTs to compare the new treatment against the standard of care. By simulating patient response, researchers were able to gain new insights into the therapy’s effectiveness and demonstrate the value of applying DT analysis to support rare disease drug development.¹²

As with all AI-enabled models, the accuracy of DT twins relies on the quality and representativeness of the data used to generate them. If, for example, DTs are constructed using only clinical trial data, they may have the same limitations in generalisability as RCTs. However, these challenges can be mitigated using the data augmentation methods already discussed.

Innovative trial designs that harness data augmentation, RWD, and the power of AI are enabling a new era of rare disease research. However, seizing the opportunities on offer requires careful planning and execution. By working with expert statisticians and data scientists, pharma and biotech companies can de-risk development, improve evidence generation, and unlock a new era of research that accelerates rare disease drug development.

References

About the author

Billy Amzal is head of strategic consulting at Phastar. Over the past 25 years, he has developed statistical methodologies to inform and support strategic decision making in healthcare. Prior to joining Phastar, Amzal led the model-based drug development team at Novartis. Then, he developed and led implementation of statistical methodologies for high impact research sponsored by public and global Health Agencies (EFSA, Global Fund, US NIH). Amzal was then senior VP at Certara, in charge of data and decision analytics, and CEO/chief statistician at Quinten Health - an AI company pioneering digital twins and disease modelling. Amzal still acts as statistical expert for public health orgs (EMA, WHO, Gates Foundation).