Experts Urge Firms To Resist Temptations Of Overactive Process Control

As pharmaceutical companies work to control their manufacturing processes in an increasingly complex, data-rich environment, they need to keep their focus on the big picture and avoid the temptation to overcorrect.

That was a theme that emerged from the recent International Forum and Exhibition on Process Analytical Technology (Process Analysis & Control), known as the IFPAC conference, in Arlington, Va.

Like novices behind the wheel who haven't learned to look down the road to avoid hitting the curb, organizations that have launched into the continuous process verification phase of lifecycle process validation are finding themselves swamped with meaningless corrective actions.

Control Strategy Renewal Challenges

Whether by planning ahead or just reacting to circumstances, pharmaceutical manufacturers are going to recapitalize their assets, and those assets include not just equipment and facilities but also manufacturing processes and their process control strategies, said FDA's Jeff Baker.

"The control strategy is a very, very valuable asset because we're in the business of reproducibly meeting outcomes and the control strategy is the mechanism for assuring that we reproducibly meet outcomes," explained the Lilly veteran, who is deputy director of the Office of Biotechnology Products in FDA's center for drugs.

However, companies have difficulty renewing their control strategies because of an unresolved philosophical conflict, he said.

The conflict is between people who believe that making a process highly capable renders it non-critical and others who believe capability and criticality aren't linked, that capability relates to the likelihood of a problem, while criticality relates only to the severity of a problem, if it happens.

"Instead of really sinking our teeth into that discussion, we keep developing lexicons of terms that we keep redefining and redefining and redefining, and that's a hurdle to renewing our control strategies because as the process matures, I get a lot more data … and I have a better understanding of my variability, but I haven't come to grips with that relationship," he said.

Baker noted other challenges to the proper renewal of control strategies – statistical thinking that obscures deeper understanding, regulatory strategies that involve rollout of manufacturing changes that put processes in constant flux, and global regulatory complexity.

He observed that "the burden of global diversity of practice and lack of alignment is really, really becoming heavy for an awful lot of companies."

The Doing/Checking Cycle

But perhaps the biggest challenge preventing companies from renewing their process control strategies is what Baker calls the doing/checking cycle.

It starts when companies encounter a new challenge. There's a new product in the portfolio, a new plant, and the company is going to have to step up its productivity.

This triggers a chain reaction: "I do a lot more stuff, which gives me a lot more to check. And do you know what happens when people check stuff? They find things. Now you can't just let them sit there. So there's CAPA [corrective and preventive action] or something which gives them more to do. So what happens then is that these old boys up here, they suck it up and go for it and they make tiger teams and stuff and special project units, which gives them more stuff to check. So they bring in people from Corporate. So Corporate comes in and checks more. So these guys come up and they hire consultants. Then these guys come down here, and they're Quintiles or Lachman or whoever. And what you do is you have a doing/checking cycle."

Knowing is what breaks the doing/checking cycle, but it's in scarce supply, he said. "Doing and checking are proportional to head count. So I can jack up the rate of doing and checking really fast, like turning up the burner with money and people.

"Knowing is not proportional to head count. Knowing is a capability that has to be nurtured, you have to have stewards. … It has to be matched to your capability, it can't be jacked up in a hurry, you can't make Einstein on an assembly line, and it's a muscle that needs to be exercised."

"One of the things that's a challenge to recapitalizing our control strategy, a challenge to lifecycle management, is that we don't have this 'knowing' capability to break the doing/checking cycle.

"These guys aren't doing anything wrong, they're doing their job. But it's going to spin out of hand and instead of making medicine for sick people, we're doing and checking each other when the whole point of the control strategy is to obviate some of that."

Which Shifts Are Critical?

Bert Frohlich of Shire HGT, who also spoke in the Jan. 25 session, said Baker's discussion about doing and checking struck a chord with him.

"I see that all the time. The quality folks … don't know the difference between something that's a truly critical event and something that's not so critical and it all gets the same treatment and the resources go out of whack. Everybody's chasing investigations. Half of them have really no impact on product or patient. It's a huge waste of time."

Frohlich went on to express hope that as companies implement continuous process verification, it will force them to see what their critical process parameters are and when they're shifting out of control. "If I can at least detect the shift, then I can take action."

Variably Variable But In Control

In a presentation the next day that in many ways paralleled Baker's, Tara Scherder singled out a statistics issue that often makes processes appear to be out of control when they're not.

It's possible that these issues relate to some of the check-and-do churn that Baker described.

The issue arises from textbook interpretations of common cause variability, said Scherder, who is managing director of Arlenda Inc., which provides statistics consulting, software and training.

Because people assume, like the textbook says, that common cause variability is random across time, they sound the alarm whenever there's non-random variation, assuming incorrectly that such variation is always a result of the special cause variability that they're trying to avoid.

Constant Flux is OK

Scherder called attention to small changes that put manufacturing processes in a state of flux and make it difficult to see whether they're in control.

The operator will make adjustments, and there will be changes in raw materials and equipment. The changes will be discrete and overlapping – one batch of raw materials for some finished product batches, a second for others; one tablet press or bioreactor for some batches; a second for others.

None of this is random or predictable, but neither is it out of control, she said. "One time you're changing the raw material for six batches, but the reagent and the lab is changing for four of those six, the next time that's not happening at all. That's not a process that's out of control, it's a process that we don't have random common cause noise, and so that means we have to keep this data structure in mind when we interpret our control charts in the context of CPV."

Statistics software might flag with red dots certain observations in control charts as "special cause" variation, based on, for example, Western Electric Rules or Shewhart rules, even for a process that's in control, she stressed.

"It's critical that we understand our operations within pharmaceutical processing" and how they might affect control charts, she explained. "Some special cause variation is expected. That is our state of control."

It's important that pharmaceutical workers – and health authorities – understand this, she added. Showing a picture of the 1893 Edvard Munch painting called "The Scream," she said, "people see red dots and they want to scream, because they think it's alerting them to a process that's out of control."

Not All Signals Are Equal

Scherder went on to caution the audience against manipulating charts to make the red dots disappear because "you may forfeit learning."

She reminded them that the point of using control charts is to learn about the sources of variation, and that not every variation needs to be investigated.

To underscore that point, she quoted Alex Viehmann of FDA's Office of Pharmaceutical Quality, who told the International Society for Pharmaceutical Engineering's Process Validation Statistician Forum last year that "not all signals are created equally. The magnitude of the reaction depends on the severity of the signal."

Normality Is Overrated

Scherder also cautioned against the common practice of normalizing process data.

It takes just one click with statistical software programs to test data for conformance to the standard bell curve, or normality; another to transform non-normal data.

But you don't need normality to run control charts, Scherder said. In fact, "it doesn't serve the patient; it doesn't serve us as engineers and scientists. It's irrational. And it's also mathematically inappropriate."

She said transforming non-normal data can actually inhibit continuous process verification, the goals of which, she reminded the audience, are "to understand your sources of variation, and have ongoing assurance that your process is in a state of control."

Your subject matter experts, who have a feel for the process data, might have difficulty understanding what they're seeing after you've transformed the data.

Another issue is that when you transform the data, you're fitting it to a model, and that model is built on an assumption that your data are homogenous. "If they are not homogeneous, all of your statistics, all of your models and all of your predictions are going to be wrong."

Typically, process data is not homogenous, she said. Instead it consists of little clusters of subpopulations that aren't very predictable. "Next year, they could be different. And so it's quite dangerous to say, 'I'll transfer my data,' because you're just over-fitting it."

Red Is The New Black

Though a statistician herself, Scherder emphasized that process experts, not statisticians, are the ones who need to interpret statistical signals from manufacturing processes.

"You might consider the magnitude of the excursion. It becomes difficult because we have to understand historical behavior. This is not done by a statistician. This is done by an SME (subject matter expert) of the process. And process and measurement knowledge is key here. So it becomes a little bit challenging, because instead of just saying, 'red dot, big problem,' it's 'red dot, what do we do with it?'"

To drive her point home, she compared two similar-looking clusters of red dots. By adding historical data, she showed that one of the clusters actually depicted a state of control, while the other indicated a process shift that would have to be investigated.

Scherder concluded: "if we remember what the goal of CPV is, to understand sources of variation, and when we react, to react in a way commensurate with the patient risk, the new paradigm could be 'red is the new black.'"