[Editor’s Note: In recognition of Rare Disease Day being observed on February 28, ACRP is pleased to present this, the second of two blogs contributed by subject matter experts offering insights on how rare diseases are being focused on by the clinical research enterprise. The first blog highlights how Net Treatment Benefit assessments may be used to support rare disease research.]
Randomized controlled clinical trials are considered the gold standard for understanding treatment safety and efficacy. Rare disease trials struggle to recruit enough participants to demonstrate efficacy, creating a barrier to research and approval. However, Bayesian statistics utilize existing evidence and expert opinion to quantify uncertainty and inform trial design, reducing necessary sample size.
The Advantages of Bayesian Methods
By combining historic data and prior elicitation—a scientific method to develop unbiased judgements from experts—priors can be formed. Bayesian statistics then incorporate observed events to update these priors as data are collected, reducing required participant numbers.
The use of prior information can reduce required sample sizes drastically, with some trials seeing reductions of up to 2,400% lower sample sizes than corresponding frequentist frameworks,¹ reducing the barrier on rare disease trials.
These priors can be used to simulate trial outcomes repeatedly, allowing for more accurate assessments of necessary sample sizes under different circumstances. This produces more reliable sample sizes, meaning less chance of a wasted trials where outcomes cannot be met.
Bayesian methods also do not follow strict power calculations, allowing for a posterior distribution of results at different strengths and meaning there is no hard cut-off and information is learnt regardless.
The International Rare Disease Research Consortium,² European Medicines Agency (EMA),³ and Parmar et al.⁴ have introduced frameworks for designing small-population trials. These include considering other information sources through Bayesian methods which are also recommended by the U.S. Food and Drug Administration⁵ and EMA⁶ for rare diseases.
Case Study 1
Adenosine deaminase severe combined immunodeficiency is a hereditary metabolic disorder that causes immune system abnormalities, which are usually diagnosed within a year from birth. Historically, the best treatment option was hematopoietic stem cell transplant, but it is unavailable to most patients and often leads to complications. Gene therapy offers a single treatment option intended to cure the condition.
Taking inspiration from Bayesian methods, a study was performed using within-subject comparisons before and after treatment compared to a historical control, allowing for more optimization of information and all patients enrolled to treatment, thus stretching the available sample size further.
Case Study 2
Juvenile localized scleroderma has an incidence of about 3.4 per 10,000 children. The most common first-line treatment has very poor tolerability; an alternative treatment is thought to have better tolerability, but there is limited evidence around efficacy.
A frequentist, non-inferiority trial would take 15 years to recruit the required 320 patients. Extracting information from experts using prior elicitation allowed researchers to reduce the required sample size to 240.⁷
While this level of recruitment is still not possible, researchers concluded it is feasible to use a Bayesian method to inform rare disease trial design, for example running simulations to further reduce the needed sample size and taking advantage of Bayesian methods which are not restricted by power calculations. The use of simulation can also allow decisions to be made in advance regarding whether it would be meaningful to conduct a study and what kind of results would be sufficient to inform change to current practices.
Need for Greater Utilization
Despite the advantages, Bayesian statistics remains underutilized. A targeted review of design and analysis features in small population and rare disease trials found just four of 64 eligible trials used Bayesian methods. The rest used frequentist approaches, with most not achieving their target sample size and thus effect size.¹
There are between 263 million and 446 million people worldwide living with rare diseases at any one time.⁸ It is vital that we optimize rare disease clinical trials by enabling effective and efficient data analysis. Bayesian statistics can and should be more widely used to support research innovations, increase trial success rates, and help those living with rare diseases.
References
- https://pubmed.ncbi.nlm.nih.gov/34910979/
- https://ojrd.biomedcentral.com/articles/10.1186/s13023-018-0931-2
- https://www.ema.europa.eu/en/documents/scientific-guideline/guideline-clinical-trials-small-populations_en.pdf
- https://bmcmedicine.biomedcentral.com/articles/10.1186/s12916-016-0722-3
- https://www.fda.gov/regulatory-information/search-fda-guidance-documents/guidance-use-bayesian-statistics-medical-device-clinical-trials
- https://health.ec.europa.eu/system/files/2022-06/medicinal_qa_complex_clinical-trials_en.pdf
- https://pmc.ncbi.nlm.nih.gov/articles/PMC11064983/
- https://www.nature.com/articles/s41431-019-0508-0
Contributed by Giles Partington, a Principal Statistician at Phastar, where he consults with multiple clients on trial design, especially in early-phase studies. He also works as part of the Bayesian elicitation research team within the company. Previously, he completed a predoctoral fellowship with the Imperial Clinical Trials Unit at Imperial College London, focusing on the use of Bayesian statistics for rare disease/small population trials.
