Many industries have long since adopted the practice of modeling and simulating experimental scenarios. And despite initial hesitancy in the clinical trials space, simulation-guided design has revolutionized, and will continue to transform, the landscape of clinical trial development. We are already seeing incredible gains in productivity, efficiency, and in our ability to ask and answer increasingly more complex and novel questions.
For this edition of the Industry Voices series, Cytel’s Chief Scientific Officer Yannis Jemiai discusses the current landscape of simulation-guided design, the enhanced role of biostatisticians in drug development, applications of SGD, and a vision of the future in which SGD enables new statistical methodologies.
How would you say simulation-guided design is changing the landscape of clinical trial strategy?
What we are seeing in the landscape of clinical trial development is a paradigm shift akin to that experienced in other sectors of the industry (energy, aeronautics, etc.). In those disciplines, it has long been a standard practice to model and simulate a plethora of experimental scenarios with the goal of managing, if not mitigating, risk and avoiding costly errors. However, when designing a clinical trial, one of the most complex experiments we run in society, teams have been averse to doing the necessary homework that would maximize the probability of success while balancing business considerations like time and money. Perhaps this is somewhat unfair since the technology has not always been available to enable this exercise in trial design. But this is changing, and trial design is becoming more intentional, considered, and collaborative as people realize the value of doing it right. The needs of clinicians, of clinical operations, and of investors can now be accounted for during the design stage, creating efficiencies throughout the development process. The ramifications are many as newfound confidence in the robustness of a design affects patient participation and informs the risks investors are willing to take. Wouldn’t you be willing to invest more in new medicines if you had a clearer sense of how the trial would play out? At the industry level, I believe this will speed both the discovery and testing of new medicines, which in turn will affect how quickly (and cost-effectively) new therapies get to patients.
Simulation-guided design is not entirely new. Could you tell us a bit about its history and what is different this time around?
The key difference is the compute power available to us today, and what this power enables us to accomplish. Compute power is a tool and an enabler, but a very transformative one. It is what drives the ability to simulate virtual trials, hundreds of thousands in minutes, many millions in a quarter of an hour. This means not only more designs can be evaluated but more insights into their behavior across assumptions can be made by the team and inherent uncertainty in the experiment being carried out. By extension, we are now able to tailor designs to more specific questions. We can suggest solutions for delayed treatment effects, or view trade-offs between greater power and timelines. If there is a competitor on the horizon, maybe a design can be found that will move up decision points without compromising on statistical power or type-1 error (a point of particular interest not only to the trial sponsor, but to regulators and patients).
How can sponsors work with biostatisticians to best make use of simulation-guided design principles?
In this new paradigm shift, biostatisticians become strategic partners, rising from technical experts to drug development professionals. Statisticians play a critical role as a member of the clinical development team by understanding the business needs of the organization and figuring out how to balance that with statistical rigor in the development plan. Simulation models can help assess uncertainty in the plan and explain or demonstrate the risks more plainly to non-statistical team members. The resulting designs outlined by biostatisticians and aligned on by the team then reflect not only rigorous statistical operating characteristics like sample size and power, but also the holistic needs of sponsors, especially as they pertain to timelines and resourcing.
What opportunities open up when simulation-guided design is utilized appropriately?
Opportunities of multiple kinds arise from the appropriate use of simulation-guided designs. Firstly, we see gains in productivity as technology accelerates the design process, giving statisticians the ability to explore and select designs faster, as well as the time to focus on more complex challenges. Secondly, the sponsor organization’s governance process improves off of this informed decision-making process, becomes more consistent and of higher quality. Finally, we expect drastic R&D savings over time and across the portfolio when this process is consistently used. We routinely observe simulation-guided design resulting in months shaved off of clinical trials and millions of dollars saved. Teams feel greater confidence in the chosen design. This is different from de-risking a clinical trial. We see that, too, but mostly when a powerful algorithm chooses your design from literally 100 million others, there is a sense of safety in what you are about to embark on.
Will simulation-guided design enable better uses of real-world and historical data?
It certainly has the potential. Figuring out what designs make the best use of external data and where the pitfalls lie when leveraging them is a great application of simulation-guided design. In many cases, protecting type-1 error (false positive rate) is challenging when using external data, whether historical or real-world, but simulating the vast possible outcomes of the trial and how the external and internal data come together offers the team insight in what may go wrong, whether consistency exists and whether any kind of inference will be valid.
What would you tell sponsors who are anxious and leaning toward more traditional methods?
Traditional methods serve their purpose, no doubt. Simulation-guided design is not meant to be alternative per se in the way that adaptive designs were. Simulation-guided design will simply tell you whether your traditional method is the one that should give you the most confidence. There is a quantitative element that tells you the strength of your design by pressure-testing it under various assumptions and inherent risks.
Will simulation-guided design help generate new statistical methodologies?
In the long run, of course it will. We are not there yet, but I imagine in a few years the adoption of this process will lead to ever more complex questions being asked, novel designs being suggested, and methodology needing to be developed to address the challenges. We already see some of these problems in the form of complex innovative designs, master protocols (basket, umbrella, and platform trials), and Bayesian methods, especially to leverage external controls.