The traditional gold standard of clinical trials with a randomized, placebo-controlled design involving hundreds of patients has inherently been unfeasible for the rare disease community. So researchers and biopharma companies are getting creative to move drugs through FDA scrutiny and into the hands of patients.
It's not only the small patient pools that make placebo-controlled trials impractical for rare disease drugs, but there's also the ethical dilemma of not giving patients with life-threatening diseases an active treatment.
At a National Organization for Rare Disorders (NORD) scientific symposium this week, academics and industry leaders shared how they are breaking these constraints with creative trial methodologies, hoping to garner more interest and support for their work from both peers and regulators.
A common method used in rare disease drug development is external control. Under this approach—which is often built on data from natural history studies or registries—a sponsor may create a synthetic control group based on available external data for comparison.
However, a few recent FDA actions raised concerns of decreased receptiveness to this method. The agency in November declined to approve Biohaven’s troriluzole in spinocerebellar ataxia, taking issue with the company’s external control design—including methodological limitations and biases—especially after a randomized trial flop. Then, in a high-profile dispute, the FDA snubbed uniQure’s attempt to use an external control comparison to support its Huntington’s disease gene therapy AMT-130.
The FDA is finalizing a guidance on external controls that will hopefully “give clarity to sponsors about what sort of evidence we would be looking for and willing to accept for approval of rare disease drugs,” Tracy Beth Høeg, M.D., Ph.D., acting director of the FDA’s Center for Drug Evaluation and Research (CDER), said at the NORD event.
During a separate presentation, Wonyul Lee, Ph.D., senior statistical reviewer at CDER, noted that “it is critical to ensure that the treatment and control groups are as similar as possible in many aspects.”
In its complete response letter to Biohaven, the FDA flagged that the company’s propensity score matching—a statistical methodology that uses patients’ baseline characteristics to mimic a randomization—was flawed because it lacks “important factors that could affect disease progression, such as baseline supportive treatments, comorbidities and geographic factors.”
At the NORD symposium, uniQure’s VP of clinical development, David Margolin, M.D., Ph.D., spent much of his presentation describing the company’s propensity-matching process, which, according to him, led to a standardized mean difference of below 0.25 between clinical trial and external control patients across eight covariances. The outlier was geography. To handle that, uniQure included a subgroup analysis by region.
But in another hurdle, FDA’s Lee also stated that: “In general, externally controlled trials are more suitable for well-defined and objective endpoints.”
That’s one key argument that a senior FDA official recently invoked to justify the agency’s rejection of uniQure’s external control proposal. During a recent press conference, the official argued that the company’s trial endpoint, the composite Unified Huntington’s Disease Rating Scale (cUHDRS), is too subjective, and because “the will to believe is strong,” patients are susceptible to a placebo effect, which an external control wouldn’t be able to capture.
The cUHDRS is a hybrid of both objective and subjective measurements. In response to a Fierce question about whether the objective components can be teased out, Margolin said: “I know there’s active dialogue with FDA and the Huntington’s disease organizations regarding how to interpret and best utilize these clinical scored measures, and that’s an ongoing process.”
Digital twins
Like the broader biopharma industry, innovation in rare disease clinical development is also taking a digital turn through the creation of “digital twins.” As Sanofi's Susana Zaph, Ph.D., explained at the NORD meeting, computational models known as quantitative systems pharmacology (QSP) now allow researchers to integrate preclinical data with natural history registries to simulate patient responses, potentially supplementing the limited clinical trial data from rare diseases.
These models can give rise to synthetic patients who are digital twins of a real patient with specific biomarker and clinical features, Zaph, the U.S. head of QSP at Sanofi, explained.
In fact, digital twins have already helped Sanofi get an FDA approval.
In the development of the enzyme replacement therapy Xenpozyme (olipudase alfa) for the rare genetic disease acid sphingomyelinase deficiency (ASMD), Sanofi managed to generate substantial evidence of effectiveness from a controlled pivotal trial in 36 adults. But in children, a 20-patient open-label study wasn’t enough.
To complement the pediatric data, Sanofi created digital twins to help extrapolate efficacy from adults to children.
“We basically took a subset of parameters that describe the model and personalized them to be able to capture the very specific biomarker and clinical development profiles of the individual patients to create digital twins, taking into account their body weight and their age,” Zaph explained. “Using this approach, we were able to quantify mechanistically how similar the disease—the treatment response—was between these two cohorts.”
The FDA accepted Sanofi’s QSP simulation results as offering valuable insights into ASMD progression and response to Xenpozyme, contributing to the drug’s pediatric approval.
However, as shown in the Xenpozyme case, unlike external controls that also draw on patient data, digital twins can only play a complementary role in a drug application, Zaph pointed out to Fierce on the sidelines of the conference. Besides, as the models rely on existing knowledge, they generally work best for monogenic diseases, where the biology is clear and straightforward, rather than complex conditions with unclear mechanisms, she added.
Time to get snSMART
While digital twins have cleared the regulatory bar to reach the market, a new adaptive trial method presented by University of Michigan biostatistician Kelley Kidwell, Ph.D., remains in its infancy.
Called snSMART—or small sample, sequential, multiple assignment, randomized trial—this multi-stage trial design allows individuals to be re-randomized to other treatments based on response to their initial treatment.
This way, all patients get to receive active treatment or various doses of the same therapy, allowing researchers to obtain more information from a small number of participants.
The idea is being utilized in a phase 2 study called Aramis for isolated skin vasculitis. Patients are randomized to receive one of three medications: colchicine, dapsone or azathioprine. If a patient doesn’t respond to one drug in six months, they are re-randomized to another treatment.
But a rare disease may not have the luxury of multiple treatment candidates. Instead, such a trial can examine different doses of the same treatment, and patients who initially respond to a higher dose may be reassigned to a lower dose, Kidwell said.
Still, the snSMART model is not universally applicable, either.
“The design and the analysis is really appropriate for relatively stable diseases over the course of the trial, or predictably declining, where there are no carryover effects of the treatment, or there could be a short washout period included,” Kidwell explained.
Because investigators would measure the same outcome at the end of each stage, the primary endpoint outcome cannot happen just once. That means it’s not suitable for gene therapies with curative intent.
The snSMART model has multiple advantages, according to Kidwell. By putting all patients on active treatment, it can help with patient recruitment and retention.
By utilizing Bayesian statistics, snSMART shifts its emphasis away from traditional p-values, as it’s very difficult to get a p-value below 0.05 in rare disease settings anyway. It instead focuses on “intuitive probability statements,” namely, quantifying the probability of a hypothesis being true, and therefore allows for the incorporation of natural history or prior trial data into the analysis, Kidwell said.
With Bayesian methods and re-randomization, a snSMART trial can reduce the trial’s sample size by 15% to 60% compared to a traditional one-stage design, Kidwell said.
The team has been developing methods to cope with multiple patient outcomes at the end of each stage of a snSMART trial, as any rare disease can be heterogeneous. It’s also possible to incorporate an interim analysis to stop a treatment arm early for futility.
However, a key question is whether findings from such a trial are acceptable to the FDA for supporting a drug application, especially as the Bayesian approach de-emphasizes traditional p-values.
“It’s a relatively new trial design, so I think there’s a little bit of a chicken-and-the-egg problem,” Kidwell said.
“The FDA has supported our development of the methods … but because there’s not one that’s already out there, people are hesitant to put it forward,” she added. “We’d love to see it go forward for drug registration.”