An interview with Chao Richard Li of Eli Lilly and Company

Chao Richard Li is Senior Statistical Advisor at Eli Lilly, where he provides statistical support and training for monoclonal antibody drug substance manufacturing. With nearly 20 years of experience in the pharmaceutical industry, including previous roles with Merck, Sandoz and Allergan, Richard is an expert in statistical process control, process performance qualification, continued process validation, quality by design, design of experiments, Lean and Six Sigma, product and process transfer, among other wide-ranging applications of industrial statistics.

With the outbreak of the COVID-19 pandemic in late 2019, Richard and his colleagues worked tirelessly on Eli Lilly’s groundbreaking monoclonal antibody therapy Bamlanivimab, the first of its class to be authorized for emergency use by the U.S. Food and Drug Administration (FDA). By the beginning of 2022, the therapy had been used to treat more than 700,000 COVID-19 patients, preventing an estimated 35,000 hospitalizations. “I know that every pharmaceutical company says they put the patient first, but to truly have that mindset, and prioritize that passion, is incredibly meaningful,” Richard says. “At Eli Lilly, I think the culture, the mindset, everything put together – I feel I am lucky.”

JMP User Reference Manager Meg Hermes spoke with Richard about his work on monoclonal antibody drug substance manufacturing, how learning about DOE changed the course of his career, and why statistics became a life-long passion.

Meg: I understand that your interest in statistics really began with DOE. Tell me about that.

Richard: I was in a PhD program at the University of Tennessee in chemical engineering and my research was focused on molecule simulation – programming to simulate molecular movement at the micro level and trying to predict the system at the macro level. Sometimes I needed to do experiments as well to verify whether a simulation made sense – whether it could really predict the system real time conditions. But, without statistical knowledge, I could only experiment with one factor at a time (OFAT) and generated a lot of data! That’s when my advisor suggested I take some statistical classes to learn how to use design of experiments (DOE) instead.

Once I started to learn statistics, I thought, “Wow.” It opened my mind! I saw that if I redesigned the experiments I had done before, it could save me a lot of time while also producing much more information. That's the DOE mindset. Also, I realized that with the same data, I could look at data from different angles using statistical tools, like JMP, explore different hypotheses and get much, much more information. I told my advisor I wanted to get another degree from statistics! Because statistics and chemical engineering are a power combination.

Meg: I can sense your passion for statistics and I’m sure that having domain expertise in your background is an asset not only in understanding the scientific challenges you want to solve but also in communicating stats with scientists.

Richard: In my current role at Eli Lilly, I support monoclonal drug substance manufacturing and have worked extensively on our COVID-19 antibody treatments. We worked really hard through the whole pandemic – we couldn’t stay home. We had to work smart and efficiently and you can see we are proud to have made an impact.

As an internal statistical consultant, when I talk to process engineers or laboratory scientists, I explain problems in the language of engineering or science, not in the language of math. I’ve been on that side, so I understand [the engineering/science] and always want to keep the engineering/scientific perspective in mind when I design experiments. I tell them, for example, that this is the conclusion based on these assumptions, here are the limitations, etcetera. Once you start building the connection, DOE really feels so powerful.

Meg: How do you balance the dual imperative to support domain experts while also empowering them to perform their own statistical analyses?

Richard: In most cases, it really depends on the design. I try not to just generate a DOE and analyze for them; I try to help them develop some background knowledge in statistics. All they need to understand is the basic design, maybe a little bit of theory, and how DOE works. Then they can implement it and push a little further. Once they feel comfortable adopting DOE, we have a deeper conversation [on the design and analysis], they do a presentation to explain it to their team, and they start seeing results. That’s powerful.

We've got a lot of very smart people over here at Eli Lilly, so I keep emphasizing: “You know your process, you know your data best. Once you know a little bit of statistics, and you know how to use JMP, you will feel that you are delivering even more value. It's not just good for the company, it’s also good for your career.”

Meg: In other words, it augments the science and what they are able to accomplish. You mentioned learning to use JMP in conjunction with learning statistics. Is that by design?

Richard: Let's put it this way: every time someone comes to me saying, "This is my data, this is the Excel file, and this is the problem," I just take it, plot the data in JMP and show them. They immediately see the patterns, the correlations, the outliers, and all they say is, "Wow."

Most of my clients will switch to JMP straightaway. That's why we now have a JMP community [within the global Eli Lilly organization]. And I have a JMP user group that meets locally. I also collaborate with JMP to give talks to the company, to educate our engineers and scientists, and [explore] how to change the culture!

Looking at a visualization is also where engineering expertise comes in. We build JMP dashboards because the engineers always want to see certain things – this unit operation, upstream, downstream. They have a design in their mind and want to quickly see key process information. I can help them to design a JMP dashboard with all real time process information they want to see. That way, every day, they can have a process meeting and look to see [how things are going].

Previously, they had to use Excel and it just wasn’t efficient. Now with just a single dashboard, they can see everything in one place like a control board. If they want to see commercial scale reactors, they just one click and boom, they instantly get all the results they need. Or say you’re interested in one signal related to a process step, and you want to know how the signal is associated with the cycle. It’s so easy to diagnose on the same process. That's huge!

We have four API manufacturing sites at Eli Lilly, and we share the best statistical practice across all sites. I do a fair amount of statistical practice benchmarking so that we can standardize as much as possible to get good practice.

Meg: In other words, you’re saying it saves time but it's also about reproducibility and standardization.

Richard: Exactly. It's easy to transfer [best practices]. Sometimes there will be a global meeting across the API sites, and one group will say, "Wow. You guys developed that script – we want to use it too." We've got a lot of data similarity in monoclonal substance manufacturing, so it’s much more efficient to use a script developed by [colleagues at a different site] than to develop our own from scratch. All you need is very minimal modification to apply it to your process.

All of Eli Lilly’s four API manufacturing sites use JMP. That's why we have so many discussions through our network to share benchmarking best practice, training – everything promoting statistics and use cases for JMP. Compared with other software, JMP has that powerful combination of, on the one hand, being so easy to use, but on the other, could go deeper if needed.

We also get excellent support from JMP to see how we can apply newly released features, among other things. For example, at a recent JMP Day, @jiancao showed us the Functional Data Explorer and @ChrisKirchberg showed us the Model Driven Multivariate Control Chart. They just opened my mind! I didn't realize a model-driven solution was already available in the latest JMP Pro release. Because JMP has evolved so much and developed new capabilities, it has become pretty much the dominant software we're using now.

Meg: One of the hidden value-adds about JMP is how much JMP Development innovates the software between new releases. They listen to users and say, “What do you need? How can we develop this?”

Richard: JMP really tries to hear and understand the voice of the customer. That's the way I appreciate it. Another thing I really appreciate about JMP is that you don’t have to be good at everything so long as you know the resource exists.

I use JMP Pro a lot, and really enjoy using modeling and simulations to help build our process understanding. I will also say that I really like how JMP is very intuitive with its graphing. When I started my career at Merck, my mentor used to say to me, "Once you've got the data, what are you going to do first? A model or a simulation? No. You have to start by plotting the data.”

When you get the data, the first three things you need to do: plot the data, plot the data and plot the data! Looking at the data from different angles and trying to find the patterns. Use your process knowledge to get some initial insight. That's 70 or 80% of the information right there!

Meg: Speaking of learning new things, what advice would you give to students or early career scientists?

Richard: It doesn't matter whether your background is in chemical engineering, biology, or arts – whatever. If you have the chance, please, please take some data analysis courses. It doesn’t have to be a statistics class – just one where you will learn data analysis that can be applied to any science or engineering discipline. It's not just about mathematics... It's about systematically logic thinking.

Once students have an analytical mindset and start using data, it will profoundly impact their professional life, maybe even their life in general. They’ll be able to use that critical thinking, that logic, to make better decisions.

Meg: Do statistical thinking skills factor into your hiring decisions? When you’re considering early-career candidates, do you want them to, for example, come in with DOE knowledge?

Richard: Yes, that's a huge plus! At Eli Lilly, we have a manufacturing stats group and whenever we are considering adding a new member, we always look at resumes. It's pretty much required that candidates have both a statistics and engineering/science background.

Meg: Are you seeing a trend towards more data literacy among recent graduates coming into the business?

Richard: It's not a trend. I would say it's booming! Analytics is everywhere. Data is everywhere. In a way, everyone should be, to some extent, a data scientist. As long as you have the mindset – and your own expertise – you will go far. Knowing programming, on the other hand, doesn’t necessarily make you a better engineer! Expertise is essential!

So, while I would say that for the next generation, programming will be really important, I would also emphasize the data analytical mindset over programming. You don't need to be able to program to do analytics. Let JMP do that.

*Disclaimer: The views, thoughts, and opinions expressed in the text belong solely to the author, and not necessarily to the author's employer, organization, committee, or other group or individual.