Why Simulation is Needed to Optimise Both Adaptive & Fixed Clinical Trial Designs
By Caroline O’Hare
Simulation is an invaluable tool for evaluating and mitigating the risks of clinical trials before conducting the studies in real time. Whilst helpful for any type of study design, this powerful in silico approach is currently mostly used to support adaptive trial design due to the relative complexity of these studies. This is because adaptive trials often involve the consideration of multiple, inter-related factors which is typically beyond the capabilities of simple study design tools. Indeed, perhaps it is because of this additional complexity, and the associated need for more specialist statistical resources, that researchers remain hesitant in choosing both adaptive trial designs and study simulation approaches. This is unfortunate as despite their added complexity adaptive trials can offer a more efficient, flexible and informative approach to gathering clinical information on a therapeutic intervention than more conventional commonplace fixed designs, and simulation offers unrivalled insight into the best designs to choose.
Advantages of Adaptive Trial Designs
Adaptive trials are useful as they can build in a series of pre-planned modifications into key components of their design. These can be opted into during the study without compromising its validity and integrity based on information from scheduled interim analyses of the emerging trial data. This flexibility contrasts with the more linear approach offered by fixed design trials where the study is usually conducted according to the initial design and the resulting data is reviewed on its completion. Hence, adaptive designs offer an important and informative learning feedback loop over the duration of a trial which allows valuable amendments to be introduced during the study to key parameters such as the allocation ratio, the total sample size, the target population, and the continued inclusion of treatment arms. Such modifications can prevent a study failing due to an inability to adequately quantify and manage an important uncertainty at the outset. In this way, adaptive trials can offer a much more responsive and efficient way of generating clinical evidence, driving down both clinical resource wastage and costs. Their use can also reduce patient burden by enabling the use of fewer participants, limiting the allocation of patients to ineffective treatments or shortening trial duration.
Barriers to Adaptive Trials and Simulation
Despite the many benefits offered by adaptive designs and simulation, these approaches are currently not widely embraced by research communities. Industry inertia may play a part here, especially where there are a lack of statistical skills or awareness. Traditional fixed designs seem to be an intuitively easier approach and so more accessible. Less specialist statistical input is required for these at the planning stage due to the apparent simplicity of their designs and so there is a common perception that you can hit the ground running faster. However, like Aesop’s fable of the hare and the tortoise, study attrition rates seem to indicate that for many of these studies, a fast start doesn’t necessarily equate to a successful finish. Indeed, the currently accepted failure rate for clinical studies of around 90% suggests that clinical studies of all types may benefit from closer consideration of their designs before initiation.
There are also undoubtedly operational barriers to adaptive designs. Ultimately, studies involving any adaptation need to stop at an interim point or points and conduct an analysis before proceeding. This may sound simple, but it can be time consuming and resource intensive in practise as all the data validation procedures need to be completed for the interim analysis. In some cases, this may result in study delays, additional cost and extended duration. Any consideration of an adaptive design needs to weigh up the benefits and the risks to ensure it is the right approach for a given situation. In this respect, simulation is a powerful tool as it enables close prospective evaluation of the clinical, statistical and operational risks.
Are Current Fixed Designs Failing?
Although adaptive designs are gaining traction, most planned trials still use conventional fixed designs. Whilst at first glance a fixed design seems to offer a much simpler strategy to gleaning important clinical information, it is heavily reliant on having a good knowledge at the start of the assumptions and estimates on which to base the study design as little can be changed after this point. However, in many cases these assumptions are far too simplistic, often assuming only one factor influences variability in treatment response. This does not reflect the reality of actual studies where there are usually multiple sources of uncertainty and complex inter-relationships between influential factors. Such failure right at the start to fully capture the real uncertainty of multiple study factors means that many of these apparently simple studies are less likely to succeed than those running them believe. This comes at great cost to the sector, both financially and more importantly to the individuals involved in them. With the pharma industry now striving to deliver a more patient-centric approach to clinical development to overcome some of the barriers involved, surely it can no longer afford to accept this.
Simulation Can Improve All Trial Design
It seems abundantly clear that it is now time to shake up how clinical studies are designed. Quite simply, simulation is no longer a specialist luxury but a real need in the design of all types of clinical study, not just adaptive trials. Fixed designs are often based on fixed assumptions and frequently do not consider or evaluate the impact of these assumptions being incorrect. Here, simulation is a valuable tool that allows researchers to consider high numbers of what-if scenarios before settling on a final optimised study design. This process can often throw up completely unexpected associations that would have never been considered, significantly reducing the risk of failure. Moreover, the complexity and multitude of data types now being collected even for conventional fixed design trials means that it is becoming essential to use simulation to assess the ability of a study to generate a broader evidence package rather than focussing on a single primary endpoint. Importantly, study simulation also allows researchers to compare the benefits of fixed and adaptive designs to identify which approach would be the most suitable for their study. There is often just as a much confusion as to what adaptive trials can’t deliver as to what deeper insights they might provide despite guidance from regulators such as the FDA.
From Trial Tortoises to Simulation Swans
But does this extra exploratory work upfront risk researchers becoming trial tortoises, slow to initiate a study and reach their research goals? Next-generation simulation is incredibly fast now and costs have dropped dramatically. This means it is possible to assess the relative benefits of thousands of designs in silico in just a few hours. The outcome of such a small investment upfront is to comprehensively de-risk and optimise a clinical study, ensuring that the insights it delivers will be statistically robust. Given the enormous costs of clinical research and running trials, this is surely time and money well spent. Indeed, given the large numbers of patients involved and the challenges of clinical trial recruitment it seems positively reckless not to use simulation in the design of most studies. However, this requires a fundamental change of mindset.
The increasing complexity of all types of clinical studies means that it is time for statisticians to be routinely brought into the design process at the very start and for simulation to be integrated into this process. The evolution of advanced cloud-based simulation packages now supports this by negating the need for time-consuming bespoke statistical programming and validation, skills which not all statisticians have first-hand expertise in. These simulation packages provide a useful sandbox environment for statisticians to pin down design options through quantitation and share visually with all stakeholders, including clinical project teams, funding bodies, ethics boards, regulators and independent data monitoring committees, the key criteria for study success. Their use not only demonstrates a robust statistical approach to development to reduce timelines and costs but more importantly that all options have been considered to ensure the wellbeing of the patients these clinical studies involve. Consequently, cloud-based simulation offers a new gold standard for designing all types of clinical trials whether fixed or adaptive. Indeed, Aesop aside, this ugly duckling data-driven approach to decision-making now has the power to bring the swan out of every clinical development project.