European Statistical Forum - Programma

Patients suffering from rare diseases deserve the same quality of treatment as other patients within the European Union (EU Regulation 141/2000 on Orphan Medicinal Products). The medical need in many rare diseases is high. Consequently, clinical development of treatments in rare diseases is important – but limited sample sizes, heterogeneous conditions and often scarce historical information to build upon can pose substantial challenges to the design of clinical trials.

These challenges are well reflected in regulatory guidance (e.g. CHMP/EWP/83561/2005), yet the specifics of what can be acceptable for a regulatory decision vary across different rare diseases and require in-depth discussion. The presentation will reflect on experiences from Scientific Advice and Marketing Authorization in Europe, and will address some of the recurrent challenges in balancing design uncertainties and the need for robust evidence.

The talk will touch upon design choices made more frequently in rare diseases, such as single arm studies, and their implications for regulatory decision making. It will reflect on the concept of conditional marketing authorization, and aim to illustrate with examples of developments of advanced therapies.

By definition when investigating rare diseases, one is confronted with limited patient availability. Accelerated evolution of in depth understanding of diseases and therapeutic opportunities further subdivides populations and demand an evolution in regulatory approaches.

Basis for the correct-robust application of statistics is the formulation of a proper “probabilistic experiment” with probabilities modelling uncertainty on different levels: the level of patient recruitment, treatment allocation, and measurement errors.

The presentation will focus on patient recruitment and its influence on trial validity, extrapolation, and robustness. Through illustrative examples specific mechanisms of bias originating from trial interdependencies and recruitment in the setting of rare diseases are considered. Strategies to mitigate will be explored and discussed from a regulatory viewpoint. Approaches to optimize patient recruitment will be presented.

Duchenne muscular dystrophy (DMD) is an X-linked neuromuscular disorder with progressive functional decline; long-term placebo-controlled assessment of milestone delay is not feasible. Treatment comparison with natural history data can provide useful information for long-term drug evaluation in rare diseases. This analysis examines givinostat in DMD using this approach.

Long-term efficacy of givinostat added to corticosteroids was assessed post hoc using external natural history (NH) comparators. Patients treated in EPIDYS and the OLE (database lock: 31 Dec 2021) were compared with NH patients on corticosteroids alone from ImagingDMD and CINRG, selected to meet key EPIDYS criteria and without other investigational therapies. Propensity score matching addressed confounding using baseline timed function tests (4stair climb (4SC), time to rise from floor, 10m walk/run) and corticosteroid regimen. One-to-one nearest neighbor matching on the logit propensity score was applied. Time-to-event analyses were conducted for major disease progression milestones: persistent loss of ability to rise from the floor, loss of ability to perform the 4SC test, and loss of ambulation (LoA). Kaplan–Meier estimates, Cox proportional hazards models, and nominal two-sided p values were used.

Propensity score–matched NH comparisons support a clinically meaningful delay in key DMD progression milestones with givinostat plus corticosteroids and provide an alternative option when long-term placebo-controlled studies are not feasible.

Federica Alessi(1), Giulia Zardi(2), Sara Cazzaniga(1), Paolo Bettica(1).
1.Italfarmaco S.p.A., Milan, Italy.
2.Alira Health, Italy.

Randomized controlled trials (RCT) play a key role in the evaluation of new therapies.

However, they are often resource intensive and take considerable time to conduct.

With fast evolving technologies and development cycles traditional evaluation approaches might simply be too slow.

Furthermore, small populations such as rare diseases and paediatrics challenge established approaches to the assessment of efficacy and safety given the limited numbers of patients.

In this presentation we provide an overview of techniques to make RCTs more efficient and illustrate the approaches through clinical trial examples in rare diseases. In the first part, we discuss how adaptive designs can be used to make RCTs more robust and yet more efficient. In the second part, we demonstrate how RCTs can benefit from the inclusion of real world data (RWD) or digital evidence using Bayesian dynamic borrowing techniques (shrinkage estimation).

Furthermore, we comment on the implementation of such designs including the role of data monitoring committees as well as Monte Carlo clinical trial simulations and their efficient implementation using artificial intelligence and machine learning techniques.

Finally, an outlook for future research methods in rare diseases is provided. Specifically, we will comment on the role in-silico clinical trials might play in small populations.

Single-arm trials (SATs), while not preferred, remain in use, especially in the Rare disease space. They may be accepted by regulators in specific contexts e.g. when the potential effects of new treatments are very large and placebo treatment is unethical.

This presentation will review regulatory and HTA positions on SATs; provide case studies in rare diseases where SATs could and could not address research questions, illustrating challenges posed when using SATs; evolving statistical methods to provide context for SATs; and strategies to optimize study design to address evidence needs.

Randomized controlled studies are considered the gold standard for demonstrating clinical efficacy of investigational treatments. However, in rare diseases, conducting long-term randomized clinical trials, or exposing patients to placebo, may be ethically or practically infeasible. This presentation describes two case studies in which alternative statistical approaches supported regulatory approval.

In the first case study, a Bayesian model-informed analysis was applied in myasthenia gravis to demonstrate the maintenance of efficacy beyond 12 weeks of treatment (1). A placebo meta-regression was used as an informative prior, combining placebo data from randomized clinical trials alongside external data from a systematic literature review and a disease registry. By integrating real-world evidence with data from randomized studies, this novel method enabled estimation of long-term treatment effect while reducing patient exposure to placebo in the phase III study.

In the second case study, an external control arm was used to evaluate the survival benefit of an investigational treatment in an ultra-rare and life-threatening disease with a high unmet need, using data from a single-arm open-label study. To mitigate potential sources of bias, multiple matching methods, including percentile matching and sequential emulation, were applied with different survival models. These analyses were used to assess the robustness of the observed survival benefit while accounting for potential selection bias, immortal time bias, lead time bias, survival bias.

Together, these two case studies illustrate how statistical methodologies may support drug development and regulatory decision-making in rare diseases, where traditional clinical trial data may be limited.

Reference
1. G de la Borderie, D Chimits et al., Maintenance of zilucoplan efficacy in patients with generalised myasthenia gravis up to 24 weeks: a model-informed analysis, Ther Adv Neurol Disord, 2024, Vol. 17: 1–15

There has been a lot of interest in so called hybrid trials, where power is gained by augmenting the control arm with external data. This often leads to designs where the randomisation ratio is not 1:1. The most common and likely acceptable use case is when it is not possible to recruit the expected number of participants due to the rarity of the disease. It applies across all clinical scenarios where randomisation is possible and necessary.

The assumption underlying this is that regulatory authorities will accept such designs regardless of the results observed. This is a strong assumption that is unlikely to hold in practice. In this talk I will discuss the likely power gains when the assumption of exchangeability is not accepted and argue that although there may be such gains, the benefits may in reality may be more modest than hoped for.

Specifically, I will expand on a particular aspect of regulatory risk, where the power of the method takes into account that such borrowing may not be permitted, depending on the results observed. The chances of this happening are evaluated, and I will show that the issue is particularly acute when sample sizes are small and there is uncertainty around the true magnitude of effect – a scenario common in rare disease development. I will also discuss what alternatives might offer.

Background
Huntington’s disease (HD) is a rare, fatal neurodegenerative disorder characterized by progressive motor, cognitive, and behavioral decline, ultimately leading to total dependence and death. With no disease-modifying therapies available, treatment remains limited to symptomatic management, highlighting a critical unmet need for interventions that slow disease progression. Conducting trials for rare diseases like HD poses unique challenges, like limited patient populations and the chronic nature of the disease, which involves slow progression over years. These factors necessitate longer study durations to observe meaningful changes in the placebo arm, raising concerns about prolonged placebo exposure.

Many studies rely on high-quality, longitudinal, non-interventional data from the Enroll-HD and Track-HD natural history data using baseline-matched cohorts and observed longitudinal averages as contextual benchmarks. However, the HD field and regulators increasingly recognize the limitations of conventional external-control analyses, given the differences in the conduct of observational and interventional studies. Further, placebo and external-control trajectories usually diverge over time despite appropriate baseline matching. Moreover, external-control arms are frequently overweighted, with far higher patient numbers than in active-treatment arms. Using historical/external-control data requires transparent handling of these challenges and sources of bias.

Methods
PIVOT-HD was a Phase 2a randomized, placebo-controlled study evaluating the splicing modulator votoplam in HD-ISS Stage 2 and mild Stage 3 HD. After the primary 12-months, all LTE-eligible patients received votoplam, with no placebo arm. In the PIVOT-HD LTE 12-month interim analysis, representing 24 months of votoplam exposure for most participants, analytical approaches incorporating historical/external-control data were used to estimate treatment effect and contextualize outcomes.

The primary digital-twin/placebo-projection approach uses predictive modeling, based on 12-month placebo trial data and historical data to project expected placebo-arm trajectory beyond Month 12. A secondary analysis using only external-control data applies propensity score methodology to address baseline comparability, and drift-bias correction to handle divergent placebo and external-control trajectories over time. These methods address residual heterogeneity, placebo-effects, assessment-frequency differences, cohort-size imbalance, and weighting strategies.

Results and conclusions
This talk’s aim is to describe the advanced analytical approaches for contextualizing PIVOT-HD LTE outcomes and estimating treatment effects while addressing challenges of using external-control data due to the lack of internal concurrent placebo arm for the full study duration. The critical discussion of the limitations and properties of the different alternative approaches may help inform future HD research using external data to adequately quantify treatment effects.
The digital twin and the external control arm approaches effectively address feasibility and methodological challenges in rare disease trials. By integrating high-quality historical data with advanced statistical methods, the proposed methods might establish a robust framework for generating evidence to support the efficacy assessment of investigational drugs investigated in non- or partially controlled trials and offer a scalable model for other trials in HD and similarly rare, progressive conditions.

Authors
Angelika Caputo(1), Yihan Sui(2), Rutvick Parlikar(3), Robin Dunn(2), Beth Borowsky(2)

Author affiliations
1. Novartis Pharma AG, Basel, Switzerland; 2. Novartis Pharmaceuticals Corporation, East Hanover, NJ, United States; 3. Novartis Healthcare Pvt. Ltd., Hyderabad, India

An engaging, live quiz designed to test and reinforce key concepts in the design and analysis of rare disease and gene therapy trials, covering innovative methodologies, regulatory challenges, and statistical approaches to evidence generation.

Join us to interact, learn, and compare your answers with peers in real time.

One of the key challenges in rare disease clinical trials is to capture patient-relevant outcomes within small, heterogeneous populations. This often leads to trial failures due to insufficient accrual or underpowered designs. Traditional trial frameworks frequently rely on a single primary endpoint, which may fail to reflect the multidimensional nature of treatment effect and patient priorities. Prioritized multi-component endpoints offer a more comprehensive and patient-centered approach to improve treatment assessment.

The Net Treatment Benefit (NTB), estimated using Generalized Pairwise Comparisons (GPC), provides an innovative approach for assessing the overall treatment effect by integrating multiple outcomes into a single metric.

This approach aligns clinical relevance with patient preferences to simultaneously evaluate the benefits and risks of a treatment. Embedding the patient’s voice into the statistical design ensures that trial success is defined by what truly matters to the community.

Furthermore, this approach facilitates transparent discussions across different stakeholders, including regulators, payers, and clinicians, by capturing the full medical value of a new treatment.

Reference
Buyse M, Verbeeck J, De Backer M, Deltuvaite-Thomas V, Saad ED, Molenberghs G. (eds.) Handbook of Generalized Pairwise Comparisons. Methods for Patient-Centric Analysis. Chapman and Hall/CRC Press, New York, 2024.

Oncology trials especially in rarer cancers can be difficult to run, both in producing results that can be meaningful but also in discussions with regulators. With the FDA’s move in recent years towards oncology trials with meaningful efficacy data even from early phases, it has become more important to fully utilize the information available within the trial.

The BOIN12 method – Bayesian Optimal Interval Phase I/II (BOIN12) design – is an early phase Bayesian method taking cohorts of patients through a range of dosages to find the optimal dosage whilst considering both efficacy and toxicity. Phase 1 of the trial looks at the utility scores of cohorts of patients at different dose levels to determine the optimal dose to recruit to next, whilst phase 2 looks to expand upon the two best performing doses to increase understanding of efficacy and tolerability ahead of taking a dose through to phase 3.

One recent trial in a rare cancer adapted this method, including an Accelerated Dose Titration phase to rapidly reach a therapeutic dose to use as a starting point for the BOIN12 design. This method and its adaptation allowed for a larger range of doses to be tested without overly increasing patient numbers, a key requirement when limited number of patients would be able to be recruited.

Whilst the BOIN12 design is a relatively novel method, agencies have been receptive to its inclusion in trials and have shown a good level of understanding of Bayesian concepts required to ensure sensible regulatory discussions during trial design phases.