A Study In Optimizing Late-Stage Processes With QbD, DoE

A conversation with Lee Smith, GreyRigge Associates

As an innovative new drug product shuttles toward commercial launch, the very thought of reopening manufacturing process unit operations to find new efficiencies might make some developers queasy.

When Lee Smith, the founding consultant at GreyRigge Associates, saw an opportunity to apply quality by design (QbD) and design of experiments (DoE) approaches to improve operations from cell banking to cryopreservation with one of his clients, he leaped.

Smith recently talked about the project at the Cambridge Health Institute’s 2025 Bioprocessing Summit and offered to answer some questions about what it took to bring critical process parameters (CPPs) and critical quality attributes (CQAs) to heel. Here’s what he told us.

Can we start by setting the stage? What type of cell therapy were you taking to market? What disease were you targeting?

Smith: Without being too specific, the product we supported addressed immunomodulatory dysfunction across a diverse range of clinical indications using products based on stem cells.

Was QbD — and DoE in particular — built into the program from the start, or did its use intensify as you approached late-stage or ran into key challenges?

Smith: QbD, DoE, and statistical approaches for assays and processes weren’t built into the program from the start but were instead applied mid- to late-stage in the product’s development cycle. The QbD and DoE approaches were only implemented after we became involved. We were originally brought in to address a different technical challenge, and we subsequently suggested they’d benefit from QbD, even at this late stage.

QbD is a nice ideal, but it can be tough to implement in practice. How did you make it stick?

Smith: We identified a QbD champion within the organization in a senior position who, together with ourselves, began to discuss with, and influence, various heads of functions about the benefits of implementing QbD.

We initially began the work with the manufacturing team, who were very supportive. Success with this team drove implementation across assay development, process development, and into R&D.

Initially, we helped the manufacturing team to define their process risk assessments, process limit evaluations, statistical analysis, and design approaches to bring remedies and mitigation strategies from cell banking to product formulation through to cryopreservation.

The QbD approach required us to define the critical quality attribute of the product and the associated risk assessments. This was very useful for the parameter attribute matrix (PAM) but also consolidated the assay development for the assays that were ultimately validated.

The quality and manufacturing groups worked synergistically to devise very stringent controls to ensure safety and efficacy of the product, including tightening the limits for the product specifications based upon the data that was being generated. This in turn fed back to ensure greater control was achieved for all of the assays through the application of DoE.

We helped train staff on how to perform risk assessments and DoE. We also supported statistical analysis, as well as providing training on the statistical tools. This ensured that the client increased their independence and cemented a QbD mindset within the company, which helped improve their efficiency for QbD implementation. We also reinforce this approach with VCs, explaining why financial support to drive risk assessments, target product profiles, critical quality attributes, etc., pays dividends longer term as the company begins to integrate both quality and compliance with development, especially at the transition from R&D into product development. Therefore, if possible, the ability to speak to the senior teams within organizations, such as the CEO, COO, and CTO, on the benefits of DoE is important. If they buy in, as budget stakeholders they release the funds and resources because they understand there will be a payoff later.

We have had clients that found use of QbD combined with sound statistical analysis helped their interactions with CDMO management. This enabled the CDMOs to see that the client was data-driven, with sound data foundations behind any product development requests.

Overall, it takes good training and expert support to consistently deliver a safe, efficacious, and regulatory-compliant product. In this particular instance, all stakeholders were aware of the development across multiple portfolios of products from R&D through to commercial sites, including in different countries and regions. This efficient approach also helped reduce the overall development costs.

I would add, DoE does need to be used expertly. This involves knowing the best designs to apply while factoring in the practical constraints of the development environment by also considering project timelines and resource constraints.

How did you apply QbD principles to assay development? Were any assays ultimately elevated to release assays that weren’t originally developed with that intention?

Smith: Many of the assays used for release were already defined, but their variance and control space were unknown, with some assays having a particularly large standard deviation. Therefore, the assays required DoE to build robustness and variance component analysis to understand the noise in the assay and build a suitable replication scheme (format variability). This put the assays in good shape for the subsequent validation under ICH Q2(R2) guidance.

Furthermore, understanding both the noise and drivers of variance for a particular assay used for formulation helped us improve the final product’s formulation and post-freeze alignment. Other in-process assays, such as visual inspection (VI), required a lot of work to understand the sources of variation and how to bring the assay under improved control and be performed to generate statistically significant results. VI is a common culprit for failure across cell therapy products, and we wanted to be sure we didn’t fail products unnecessarily.

How did QbD help when it came time to transfer to commercial manufacturing? Did your design space or control strategy hold up at scale, or did you have to revisit key assumptions during tech transfer?

Smith: We worked on a product that was rapidly moving toward commercial launch, with the successful outcomes of the QbD application being applicable across other commercial products with the portfolio using prior knowledge and prior data. The implementation of QbD within the R&D and process development groups working on the products allowed an easier scale-up as well as implementation of newer technologies into the commercial process without moving outside of the normal operating ranges (control spaces) such that regulatory applications for changes to process did not trigger a requirement for new clinical trials. That is, implementation of QbD greatly aided all aspects of the product portfolio and resulted in a reduced risk of failure at scale-up or tech transfer across sites as well as across different regions.

How do you see regulatory expectations evolving for cell-based products, especially when it comes to QbD documentation?

Smith: Due to the relatively new nature of the cell therapy modality, plus the inherent variability associated with these types of therapies, application of QbD, and particularly statistical approaches, such as DoE, can be difficult to implement in line with the regulatory guidelines. This also applies to stem cell derived products such as secretomes and extracellular vesicle products. This is due to the complex nature of the CQAs where proteins, RNAs, and cellular interactions all play a role in efficacy of the therapeutic.

The advent of a number of cell-based assay approaches with human readouts is helping drive safety, efficacy, and toxicology analysis of such products. The ability to potentially analyze interactions and how they affect aspects such as patient epigenetics or their potential for cancer induction over the longer term are already on the radar for a number of regulatory agencies.

For example, the PMDA in Japan has a long-term follow-up for subjects who have undergone CAR-T, CRISPR, or stem cell-based therapies, and they are always looking for ways to use more interactive analytical tools to assess long-term efficacy and late-onset adverse events.

In addition, how this drives clinical design and their readouts is also being thought about and addressed. The explosion in AI is allowing agencies to start assessing whether implementation of massively complex data sets across assays and the production processes, where safety and efficacy will be required as part of the approval process, could impact the pre-approval inspection and product assessment.

QbD can be used to generate data that feeds into machine learning to help anticipate critical process parameters as well as qualify and validate methods, which may ultimately percolate into the regulatory guidance. That is, using QbD effectively will increasingly become a standard regulatory requirement for cell therapy development and product life cycle management.

How did this work accelerate the path to commercialization for other drugs in the company’s pipeline?

Smith: As indicated earlier, by understanding the underlying noise, sources of variation, the process control and operating spaces, and the subsequent remedial and mitigation strategies used, this allowed the learnings to be transferred across all products in a portfolio. For R&D and process development, understanding the requirements based on the target product profile and CQA assessments is an important step in ensuring efficient use of resources and time.

While the work never ends, given that process platforms change, understanding how and when to use the tools around the QbD framework within an organization, from R&D through to clinical, can bring dramatic efficiency improvement to all aspects of a portfolio, including easier regulatory pathways and shortened development timelines that accelerate the BLA filings. There is also the prospect of AI performing real-time monitoring that may improve efficiencies across similar product portfolios. This has a profound impact upon an organization’s development spend. Shorter timelines to BLAs should result in an improved return on investment by reaching commercial launch sooner.

About The Expert:

Lee Smith founded GreyRigge Associates in 2010 after holding senior roles including CEO and VP level positions. Lee has worked both in small biotech startups and multinational corporations in the U.S., U.K., and Singapore, working for companies such as GSK, Emergent BioSolutions, and SingVax. Contact him at contactus@greyrigge.com. Learn more about Lee and GreyRigge at www.greyrigge.com.