Revisiting a Seamless 2/3 Trial: The Amazing Journey of a GLP-1 Agonist
The Advent of Innovative Trial Design
In pharmaceutical research, the conventional phase-by-phase approach is slow and methodical. This was notably challenged in the Eli Lilly AWARD-5 trial for their GLP-1 agonist, dulaglutide. Initially envisioned as a three-dose phase two trial, the objective was to optimize the likelihood of success of the study, select the best therapeutic dose, and utilize less time and patients. Berry Consultants collaborated on the creation of the seamless 2/3 trial. The trial began with seven doses of dulaglutide, compared to an active comparator, sitagliptin, and a placebo, forming a robust nine-arm trial structure.
At the heart of the trial was its unique Bayesian adaptive design: a phase 2 stage allowed bi-weekly adaptive analyses to optimize the randomization among the seven dulaglutide arms and selection of two arms for seamlessly moving into the phase 3 stage with a final analysis including both stages.
Seamless trials like this redefine efficiency in drug development, cutting down on the timelines and sample size with improved power. The success of Dulaglutide, with brand name Trulicity, illustrates the practicality of designing and executing complex adaptive trials driven by prospectively set Bayesian models, integrating bi-weekly adjustments, and allowing for immediate transitions through phases that can slow down traditional phase-by-phase strategies.
Navigating Barriers and Outcomes
The trial's success was not without hurdles. The design involved multidimensional modeling of the longitudinal trajectory and dose-response relationship for four endpoints. The concerns of both regulators and the trial sponsor were aptly addressed by modeling safety parameters like blood pressure and heart rate, alongside the primary efficacy endpoint HbA1c. Regulators recommended explicit modeling and incorporation to the randomization and dose-selection of cardiovascular safety, quantified by blood pressure and heart rate changes.
Implementing a Clinical Utility Index (CUI) became crucial when juggling the multiple endpoints – for efficacy, safety, and weight loss. The dilemma of potential high efficacy at unacceptable safety levels was elegantly managed by combining the endpoints into a single index. The CUI modeled HbA1c reductions, blood pressure and heart rate change thresholds, as well as the potential for patient weight loss. The FDA's input was essential—they set limits to ensure safety, which were integrated into the CUI and heavily informed the dose selection process.
The trial also marked a significant advance in the application of response adaptive randomization. Initially fixed for the first 50 patients, the trial later adopted a dynamic allocation approach by updating randomization decisions every two weeks. Such frequent updates were made feasible by real-time data updates and automated Bayesian algorithms. This not only minimized patient exposure to inferior treatments but also magnified the precision in dose selection – the implementation of response adaptive randomization allowed the exploration of 7 doses instead of the original 3 doses, with minimal or no additional sample size. The eventual marketed dose of Trulicity may not have been explored if not for the adaptive design.
Mastering Trial Execution through Simulation
A cornerstone of this trial’s success was its reliance on in-depth clinical trial simulations for building the complex design. The clinical trial simulation construction of the adaptive design involved simulating many versions of the design – well over 500 potential trial designs across a range hundreds of possible scenarios. The iteration and optimization of the design process involved sequential exploration and evolution of the design – continually evaluating its ability to pick the best therapeutic doses, result in statistical success, with the most efficient use of time and patients.
The iterative design process through simulation is analogous to the design process of an airplane through the many simulations of its performance across multiple measures for optimizing its performance. The details of the optimization are exemplified by Eli Lilly’s strategic restriction on patient recruitment speed, ensuring appropriate decision making while balancing the time of the trial. The resulting AWARD-5 trial was optimized through millions of simulated clinical trials – setting the framework for the one real trial.
The structured simulation approach encompassed over 300 null scenarios—crucial in achieving regulatory buy-in. In a landscape where type one error control is mandatory and highly scrutinized, Eli Lilly's trial became a benchmark, demonstrating through rigorous clinical trial simulation how a balance of innovation and compliance could be achieved.
Further pivotal was the collaboration with the DSMB, tasked with overseeing trial integrity and patient safety. While the trial implementation was automated the DSMB formed the human review of the design. The design team utilized the comprehensive simulations to create “design movies” for the DSMB to envision exactly what, when, and how, for the actual trial. The multiple movie presentations ensured complete understanding by the DSMB that would “fly the plane.”
Bayesian longitudinal models, designed to model and predict outcomes over time, were critical. The longitudinal models incorporated prior trial outcomes to form the prior expectation of the shape of each endpoint over time. While the primary outcome was the 52-week change in HbA1c, the design was able to select the two doses and graduate to the phase 3 stage with minimal 52-week data.
Long-Term Implications and Conclusion
The seamless 2/3 trial's legacy extends beyond dulaglutide's market success. It represents a transformative approach in clinical trials, setting a precedent for future pharmaceutical research. Eli Lilly’s trial demonstrated that complex adaptive designs could allow optimal balancing of safety and efficacy outcomes while significantly reducing drug development timelines.
This trial illustrates the potential of innovative trial designs. While complexities and customization will always be required, the path forged here offers valuable insights and reassurances that adaptive trials can extend beyond theoretical optimism into practical reality—shaping a new course for pharmaceutical development.
For full details on the trial design and implementation see:
Skrivanek Z, Berry SM, Berry DA, Chien, J, Geiger MJ, Anderson JH, Gaydos, B (2012) Application of Adaptive Design Methodology in Development of a Long-Acting Glucagon-Like Peptide-1 Analog (Dulaglutide): Statistical Design and Simulations, Journal of Diabetes Science and Technology, Vol 6, No. 6, 1305-1318.
Geiger MJ, Skrivanek Z, Gaydos B, Chien J, Berry SM, Berry DA, Anderson JH (2012) An Adaptive, Dose-Finding, Seamless Phase 2/3 Study of a Long-Acting Glucagon-Like Peptide-1 Analog (Dulaglutide): Trial Design and Baseline Characteristics, Journal of Diabetes Science and Technology, Vol 6, No. 6, 1319-1327.