Rare Disease Clinical Trials May Require Paradigm Shift Away From Traditional P-Values to More Adaptive Models, Experts Say
The future of clinical trials in rare diseases and precision medicine may require a paradigm shift away from traditional study methods. The shift trends toward the use of new and adaptive tools to squeeze every ounce of information from data gathered in shrinking patient populations, according to experts.
At a meeting on statistical methods and study designs in rare disease drug development, convened last week by Duke University under a cooperative agreement with the FDA, representatives from academia, the agency and industry sought to address the main challenges in the setting. Their feedback could form the basis of formal guidance documents, the university said.
One method to increase efficiency would be to borrow prior data from early phase trials and incorporate them into phase III designs, which traditionally begin de novo.
“Realistically, we’ve been using phase II borrowing for years,” to design subsequent trial dosing and regimens, for example, said Laura Lee Johnson, an acting division director in CDER’s Office of Biostatistics. “When you don’t, it goes haywire.”
However, this would go farther when formally incorporated into a Bayesian clinical trial approach, which aims to estimate the probability that a treatment is more effective, in a move away from the traditional reliance on acquiring two P-values of less than 0.05, described Karen Price, a research advisor at Eli Lilly & Co.
Evaluating, weighting and using data from a wide range of potential prior sources — including expert opinions; historical controls; natural history studies; summary-level data from randomized controlled trials or observational studies; pharmacokinetic and pharmacodynamic findings; and individual-level data from patient registries — can allow for greater flexibility in modeling and prediction in a completely transparent manner, Price said. Dynamic methods can allow sponsors to borrow more when current data are similar to historical data, and protect against over-borrowing.
Sponsors would need to assess the relevance and exchangeability of historical data compared to new data, Price said, including the similarity of indications, patient population and endpoints, as well as the time since collection. Data could be borrowed from either control or treatment arms, or both.
But borrowing historical data needs to be carefully considered. “The investigational sites may not be exactly the same, but are the regions the same?” added John Scott, acting director of CBER’s Division of Biostatistics. “If the regions aren’t the same, are there any substantial differences in the background care that patients are receiving? That can make huge differences in comparability.”
From the FDA’s perspective, the question of how much to borrow for phase III trials can become very complicated, said Scott, in addition to the practical trial constraints in the rare disease setting, with few patients available to participate in studies.
“I think that we have a moral obligation, especially in this setting, to use information as wisely as possible,” he said, as well as an obligation to not make erroneous conclusions of effectiveness.
Sponsors should plan their borrowing prospectively whenever possible, including when designing early-phase studies, and always ask if the external data were intended to be used in clinical research.
“If you have an overall development strategy that includes borrowing, it’s going to encourage collecting data in a way that fosters more homogeneity — and it can also help avoid some blind alleys where you might run a risk of fooling yourself,” Scott said.