COMPANY CONFIDENTIAL. DO NOT DISCLOSE.
How statistics changed the game at Moderna
In the face of a manufacturing challenge unprecedented in speed or scale, experts at Moderna saw that a data-driven approach to process characterization could accelerate the transfer of its ground-breaking mRNA vaccine against COVID-19. The rest is history.
Challenge: After securing emergency use authorization of the mRNA-1273 vaccine against COVID-19 from the FDA in December 2020, Moderna faced the singular challenge of rapidly transferring the company’s ground-breaking technology to manufacturing facilities at a global scale. “It was a very interesting convergence between trying to ramp up both the speed and the scale at the same time,” recalls Distinguished Fellow Julia O’Neill.
Solution: Moderna, O’Neill says, “didn’t have time not to take a quality by design approach” to process characterization and transfer. By standardizing around JMP® statistical discovery software – the leading out-of-the-box solution for design of experiments (DOE) – the company’s statisticians and scientists were able to generate deep process understanding quickly and provide optimal design specifications to manufacturing facilities around the world.
Results: Design of experiments and quality by design methods in JMP played a key role in accelerating technology transfer in the wake of the FDA authorization of mRNA-1273, ultimately enabling Moderna to make its COVID-19 vaccine publicly available several months earlier than originally predicted. Moreover, O’Neill says, the company’s use of a DOE-based approach to the process characterization of its original vaccine has created a framework that is now enabling scientists to quickly adapt the technology to defend against new variants of the virus.
_____________________________________________________________________________________
The COVID-19 pandemic posed an exceptional test not only of the ingenuity of the global scientific community but also of the extraordinary resourcefulness of those responsible for manufacturing vaccines at scale. By moving swiftly on both fronts, American biotechnology innovator Moderna in 2020 developed and manufactured a COVID-19 vaccine from sequence to authorization to distribution in 12 months.
With the announcement in December of that year that Moderna’s vaccine, mRNA-1273, was found to be highly protective against COVID-19 in Phase III clinical trials and subsequently issued Emergency Use Approval (EUA) by the US Food and Drug Administration (FDA), Moderna CEO Stéphane Bancel declared the company would continue to focus on developing manufacturing capability both in the US and around the world. “It has been a 10-year scientific, entrepreneurial and medical journey,” he announced in a press release. “But we remain focused on scaling up manufacturing to help us protect as many people as we can from this terrible disease.”
By the end of 2020, Moderna had committed to supply agreements totaling more than 600 million doses including 300 million to the US government, 160 million to the European Union, 50 million to Japan, 40 million each to Canada and Korea, and at least 25 million to other countries. Feats of technology transfer rarely make the front page in ways that clinical trial readouts might, however the less-told story of Moderna’s extraordinary efforts to rapidly transfer vaccine technology to production is no less compelling.
“It was a very interesting convergence between trying to ramp up both the speed and the scale at the same time,” reflects Julia O’Neill, Moderna Distinguished Fellow and CMC Modeling & Statistics Lead. “There were a lot of challenges as of course the COVID-19 vaccine is intended for essentially the entire population of the planet. And it had to be rolled out as fast as possible.”
Unprecedented challenges
Years before joining Moderna, O’Neill had been deeply involved with the first gene therapy to be approved in the US while at Merck. The company’s innovation targeted a rare disease with roughly 100 patients diagnosed nationally each year and had been approved for an accelerated production schedule based on its FDA breakthrough designation. Later, she worked on the rollout of humanized antibodies used in Merck’s cancer immunotherapy Keytruda. “I thought I had experienced acceleration when I worked on these breakthrough designation programs,” she says. “But in 2020, it was emergency use authorization, which is even faster.”
In the pharmaceutical industry, one of the most challenging aspects of scaling production lies in the establishment of the stability profile, a critical quality attribute of any vaccine or therapeutic product. Traditionally, scientists would determine a product’s shelf-life stability by monitoring product physicochemical degradation under tightly controlled storage conditions over time – for often as long as 12 months.
The extraordinary circumstances of 2020, however, meant that Moderna did not have 12 months to spare. They would need a more innovative solution to balance well-established stability analysis methodology and the quality assurance required by regulatory review boards. Statistical approaches provided a path forward to extract maximum insight from a combination of the team’s existing understanding of stability mechanisms and product degradation, pulling all the data they had available into a modeling framework.
No stranger to applied analytics, O’Neill had made a career out of advocating for comprehensive statistical programs in industry. Just prior to her coming on board at Moderna in 2020, the company’s forward-thinking technical leadership had standardized around quality by design (QbD) and design of experiments (DOE) workflows – and procured a tool their scientists would need to implement new best practices. That tool was JMP® statistical discovery software, a platform O’Neill says was an ideal choice for Moderna and central to the quick implementation of analytics workflows for DOE and stability analysis among other things.
“Stability analysis is the starting point to look at stability profiles for drug products or their components,” she explains. “With JMP, I can do a customized analysis using a fit model to help line up the modeling with what we know about the chemical kinetics of shelf-life and degradation. JMP lets me customize models for that and determine a product’s shelf-life.”
No time to skip Quality by Design
Despite the common misconception among scientists that statistical analyses add unnecessary complexity to traditional scientific workflows, O’Neill sees things very differently. Statistical methods in JMP, she says, augment – not replace – scientific expertise, reducing complexity, making processes more reproducible in the long term, and streamlining workflows by eliminating non-essential research steps.
“Whenever I speak to people outside of Moderna,” she says, “a lot of times they’ll say, ‘you probably didn’t have time to do quality by design [with the COVID-19 vaccine].’ But actually the opposite is true! We didn’t have time not to take a quality by design approach. In fact we did a much more thorough quality by design process definition [with mRNA-1273] than many other products I’ve seen.”
“It was challenging – no question about it – just to move so quickly. But we were as well prepared as we possibly could have been. And that plays a big part in why we were able to [get the vaccines into circulation so quickly] after the clinical trials,” she adds.
Statistical models help manufacturing sites around the world adjust process specifications according to available equipment
As Moderna worked to deploy vaccine production capabilities globally in anticipation of worldwide distribution, the company has transferred processes for each individual component of its mRNA-1273 product to geographically diverse manufacturing sites. Process understanding, generated through DOE, is vital to providing manufacturing partners with much-needed flexibility in accommodating variation in production conditions.
When DOE is used to characterize process controls and develop a control strategy, it outputs a design space within which set points can be adjusted without affecting the end product’s ability to meet specifications, O’Neill explains. “Instead of testing one or a select number of conditions, you’re looking at a factorial array of conditions and ranges that you might want to use full-scale to explore in combination with other factors. DOE gives you an operating region and proven acceptable ranges within it.” Moreover, she adds, highly predictable process validation results mean that fewer confirmatory validation runs are needed, thereby reducing transfer time and cost.
The ability to transfer well-characterized processes to diverse geographies is now serving the company well with its commitment to providing COVID-19 vaccine doses to low-income countries in partnership with COVAX. And with the late 2021 announcement that Moderna has committed to producing up to 500 million doses annually at a new manufacturing facility it will build in Africa, the statistical underpinnings of the company’s robust process controls are poised to have an even greater societal impact.
Process characterization generated by DOE lays the foundation for new iterations of the vaccine that will respond to emergent variants
Moderna’s data-driven approach to process development and characterization is also set to accelerate future iterations of the vaccine now in development, including boosters against the known SARS-CoV-2 variants. “Because we have a really strong foundation of well-characterized processes based on extensive design of experiments, we can now look at the control strategy to see which parameters are potentially dependent on the variant we’re trying to go after. It’s actually a pretty straightforward scientific discussion,” O’Neill explains. “Not only has this approach allowed us to move very quickly against the original strain of the virus, it is now allowing us to move quickly against the variants as well.”
In addition to reacting to emergent viral variants, all of the work done in 2020 will also feed directly into Moderna’s next line of products including a single shot vaccine for both flu and COVID-19 and a pre-clinical rare disease product. “With the knowledge we’ve gained with COVID, we know it should be very doable to dial in and figure out what a product for this rare disease would look like,” O’Neill says.
The statistical foundations of mRNA-1273’s stability profile have also been pivotal in enabling O’Neill to react quickly to unpredictable supply chain challenges and delivery incidents. When in early 2021, a severe winter storm resulted in widespread power outages in Texas, for example, she was called upon to assess how temperature fluctuations may have affected the shelf-life of doses in cold storage. “Early in the pandemic there was a huge focus on not throwing away doses unnecessarily – that is, doses that still had useful shelf-life left in them,” she adds.
By going back to the historical analysis to establish rate estimates, O’Neill was able to determine that the affected vaccine doses could still be used – and reliably predict new shelf-life expectations. “I set up a Monte Carlo simulator in JMP and used the profilers to find out how shelf-life would be impacted by a temperature excursion that was 24 hours but didn’t get any higher than room temperature,” she says.
An end-to-end workflow tool that evolves alongside the science
In addition to having standardized their statistical workflows around JMP, widespread support for applied statistics have catalyzed a democratization of both skills and tools to Moderna’s front-line scientists. As a result, any delays in waiting for expert support from statisticians has been virtually eliminated, O’Neill explains; “DOE has tremendous value and now that we’ve got lots of people using JMP, they can take it pretty far on their own.”
“If we really want to democratize DOE [and other statistical methods], we can’t expect everyone to become a Python or R expert,” she adds. “We do certainly have very skilled coders at Moderna, but you would have to hand off information or a problem to someone who can do the coding. And then they’d have to hand it back. That process would be so much less intuitive than JMP.”
As an analytics change agent and advocate for widespread analytics capability, O’Neill’s goal for Moderna is to move away from what she calls a transactional mode of statistics and toward a do-it-yourself model with opportunities for collaboration. “People should be able to set up models and analyze them, at least largely on their own,” she says, and JMP makes standard workflows like this far more accessible to domain experts, freeing them up to focus on the science – not the code.
“It’s all about piecing together different ways of using JMP from beginning to end, which is all very easy to do of course,” O’Neill adds. “Now I just can’t imagine doing my job without JMP, I really can’t.”
And as the science evolves, so too does JMP – what O’Neill says is one of the hidden value propositions of the tool. “We’ve all seen software packages that become stagnant and lose touch with the applications – or just don’t seem to work over the long term,” she says. But with new, feature-rich releases of the software every 18 months, she adds, “JMP has really been keeping pace with the evolution [of industrial statistics] – and helping to advance it.”
In a world where scientists are increasingly called upon to innovate against unprecedented calamity – be it climate change or the next pandemic – such advances couldn’t be more important. No rest for the weary, O’Neill says; “This past year, we really got it done [at Moderna]! But we could always standardize and optimize things further.”
