When can global Pharmas return to normal?

The global Pharma industry will never return to the pre-COVID-19 world because it’s not there anymore. Neither does it make any sense to either wait for things to return to “normal” or for the dust to settle. Neither of those events is ever going to happen. THIS is our “normal”. THIS is what we have to work with! Let’s wake up and get to it!

Any direction is good if you don’t happen to know where you’re going

Since we cannot go backward, one good way to figure out the best direction forward for global Pharma right now is to look at nutty trends to see a direction that might be tempting but at the same time is also a waste of time and money. Luckily, global Pharma has examples of these in abundance. 

I confess to still finding one of these trends interesting and exciting myself. This particular Pharma trend is called “Continuous Processing” (CP). This Next Big Idea is that the industry should move beyond Old School batch-by-batch drug production and migrate to optimized continuous processing. Put simply, under CP, we should continuously manufacture our drugs like we manufacture our cars; if demand drops, we throttle back, but we don’t stop. FDA has made it crystal clear that they think that CP is a grand idea. 

To me, CP is missing just a single critical piece

It’s easy to demonstrate that CP will be a good idea in the future but one for which Pharma is currently very unprepared. My evidence is simply the terrible difficulties we frequently encounter when we try something that is technically much simpler than CP will be. Far too often, we fail miserably when we merely try to transfer a batch process from one Pharma site to a sister plant within the very same organization. We know that it fails, but we rarely have even a clue as to why. Once we can accomplish tech transfer between sites with assurance, then we will then be ready to identify, scientifically, the particular processes that will benefit most from CP, not before. The fact that FDA wants to throw CP into today’s global Pharma mix would just increase the present chaos. Now THAT is the nutty part!

But without nutty ideas, what are we left with?

Global Pharma is unprepared for CP so where can we go to find a strong and sensible direction for which we ARE prepared?  We will find some solid guidance if we return to one version of our Pharma First Principles: In July 1998, Robert Kieffer published a very short article entitled “Global Trends, Needs, Issues” [1]. In just two pages, he laid out a fundamental critique of 1990s Pharma practice and culture, both positive and negative. That critique still stands up very well today. Please take a minute, Google him, and read him for yourself. I’m betting that you will be impressed. Here is what he says in one particular sentence of his conclusion:

Robert Kieffer (way back in 1998!)

“We need to develop a holistic, integrated quality system; to use better quality tools, such as FMEA, process capability, process management; increase our understanding and improve control of critical process parameters so that we can move to parametric release.” 

For me, this sentence contained just one single unfamiliar concept. I might otherwise have easily dismissed Keiffer’s thoughts here as simply a series of marketing buzzwords except for the fact that I’d never heard the phrase “parametric release” before. When I asked around within my Pharma environment, everyone gave me a blank stare. If I hadn’t been quite lucky, I probably would never even have seen parametric release in person, even though it was happening only 30 miles away from my workplace at the time. 

By 1998, when he wrote this, Kieffer was winding up a 30 year career in Pharma. Yet he still managed to bequeath to us what I still believe is an ideal target to help us improve the global Pharma business model, right now. I hope to convince you that Keiffer’s idea is an achievable target that global Pharma can and should pursue, capture, and apply. Ironically, this is despite my belief that Parametric Release is going to be as technically difficult to achieve as Continuous Processing will be, when CP’s time finally rolls around. The difference between the two ideas is that Parametric Release will lead us directly to repairing and rationalizing our fundamental measurement practice, so we need to invest in it first. We already have all the tools necessary to succeed.

One concrete thing that convinced me that Kieffer’s idea about Parametric Release is still the way to go today occurred when I witnessed it in action in the Toyota Tacoma truck assembly plant in Fremont, California. 

What I saw in Fremont

Suppose that, upon entering the Fremont Plant, we stroll to a point in the assembly line that is about 75% along the path toward truck assembly completion. As we approach, we see a line of partially assembled trucks slowly passing by us with a small assembly team working on each truck as it passes their position. But you ask, where is the Parametric Release? It is right in front of us. If any member of the team doesn’t like what they see, if they cannot complete their team task, they “pull the noodle” and the entire line stops within about 60 seconds.  

This attribute gives each successfully assembled truck an identical membership in the sample space. This is because each has survived the same set of steps that lead to the end of the assembly line. The corollary to this fact is what leads us to the practice of Parametric Release. What is the “parameter” that Toyota tests at the end of the assembly line? It is simple: 1) if the assembled truck has survived examination by each team and 2) if the truck is present at the end of the line. If it is, then it’s ready and complete. Incidentally, this process also results in a truck that does not need to be inspected as if it had just been encountered by inspectors for the very first time. It has already been inspected at each assembly step. The truck’s presence here at the end of the assembly process means that each assembly team has approved its progress at each step. Otherwise they would have stopped the line.

What is a sample space and why do we need it?

A sample space is the virtual space in which a manufacturer places all their acceptable product results. Although each product located in this space is unique because of random variation, they are all identical to the customer in terms of their “fitness for use”. There is a dimension in this space for every critical customer requirement. If the customer must be satisfied with three different product attributes, then the sample space has three dimensions. In any case, no matter how many dimensions there are in this space, all the final products fit within its outer edges. Toyota can be said to have created a sample space to support its manufacturing process and global Pharma should imitate them as soon as it possibly can.

The concept of a “sample space” can help us to bridge from Toyota’s world to ANY Pharma process practice

To make Keiffer’s (and my) wish come true, we need to clearly see a sense in which the Parametric Release practiced in Fremont to produce trucks can be adopted directly and very advantageously to a global Pharma business model.

Every Tacoma truck arriving at the end of the line is unique. The plant couldn’t produce a perfect duplicate even if they ran the line non-stop for 100 years. But none of the tiny, unique, random variations in each Tacoma was ever big enough to attract the notice of one single assembly team. Toyota accepts any and all possible combinations of individual Tacoma outcomes into the sample space.  That’s because none of those differences ever rose to the level of rejection by a single team. Much more importantly, all the differences between any of the trucks within this sample space are completely invisible to Toyota customers.

In Fremont during the Toyota era, Tacoma trucks were assembled in series of what were ultimately “Yes/No” steps. Any truck can only arrive at the end of the assembly line if each and every assembly team made what is, in effect, a “Yes” decision. But millions of actual measurements are also baked into this process.Acceptance of some truck components requires measurement that resolves physical dimensions down into the range of a ten-thousandth of an inch (to put it in English not metric terms). 

What unifies Toyota and global Pharma measurement practice? 

All measurements that we make within a production system involve two tiers, 1) the first tier contains the continuous range of possibilities produced by the measurement process itself, and then 2) the second tier couples those measurement results to the binomial behaviors that ride above any measurement process when we must then decide “Yes” or “No” if these results fit within our needs. In addition, all measurement results have an uncertainty attached to them that affects both tiers. It is sometimes possible, in ideal measurement regions, to ignore the fact of that uncertainty with no consequences that are immediately apparent. However later on, that choice may prove to be extremely expensive.

So far, these facts apply equally to all Toyota Tacomas as well as all global Pharma drug products.

For example, somewhere in the world, there is a factory making pistons for the motors in Toyota trucks. When each of those pistons is made, the factory must not only measure them as carefully as needed, but also accept or reject each of them for placement in a Toyota engine. This is the coupling of continuous measurement data with the binomial behavior that helps us accept or reject the results.

We can find analogous situations in every Pharma, today. More particularly, this analogy applies not just to the measurement of global Pharma products and their components but to all Pharma measurements that get paired with accept/reject decisions. For example, consider a Pharma QA lab preparing to tell Manufacturing whether or not that last shipment of some raw material is good to go. The same analogy applies here, too. So far, global Pharma measurement practice still appears identical to any truck manufacturer. 

Where can we see differences? 

As far as Parametric Release and measurement practices are concerned where do we see differences between these two industries? One big difference is the way that measurement uncertainty gets handled. The second is what response a measurement consumer makes to the gradually increasing risk they encountered when their measurements approach any acceptance limit, no matter where they place that limit. To be frank, although I know some US auto-making history I don’t know the exact specifics of how Toyota does it. I know that it was first accomplished a century ago or else we wouldn’t have cars today. My understanding or lack of it pales in comparison to the opinion of millions of today’s truck buyers. Clearly the world truck market continues to think that Toyota has handled these two risks quite nicely indeed.

One thing that I can tell you about measurement uncertainty in the US auto industry is that by 1923 Henry Ford had already bought out C. E. Johansson and brought both him and all his “Jo Blocks” to the US from Sweden to fortify the Ford production process. (Jo Blocks were and still are very carefully machined dimension standards that come in a variety of block sizes. Their name honors Johansson). This fact implies that more that 100 years ago US auto makers knew very well that holding a critical dimensional measuring tool in your hand for too long as you measured auto parts would distort and bias your attempts to measure a critical dimension, like the width of a piston. 100 years ago, auto makers knew enough to do careful temperature control in areas where they used Jo Blocks for parts quality control. However the automakers managed to confront measurement uncertainty and risk, they accomplished that feat in an earlier time when competition was fierce, but practices were not formalized or regulated. Some of this institutional knowledge has been forgotten or was never recorded.

There is just the tiniest little roadblock to trying to contrast current Pharma behavior with auto and truck manufacturing about this point. That is because the entire Pharma sector simply pretends that measurement uncertainty does not even exist. No hint ever appears in any SOPs, docs, practices or policies of any kind. Nor does the FDA ever bring it up.  Nor do Pharmas exhibit any incremental response to measurements that reflect a process that is heading for its limit. Pharma drug products are either “in” or “out” of specification. Period. QA Lab results are in perfect agreement with this “policy”, they’re either “in” or “out” of specification. Period. Above, I wrote about what I believe are the possible costs of this choice. Up until now, Pharmas have managed to avoid adding these “risk costs” to their internal balance sheet. They accomplish this by passing them right through to their patients whenever they occur. That way, Pharma can continue pretending that neither the costs nor the risks even exist, while pulling down annual margins of about 20% annually.

If, as I say there is a big problem, where is it easiest to see? It becomes easier to see as we progress outward toward the edge of any global Pharma process limit. Remember, we will always see it no matter where we have set that limit.

Meanwhile out at the acceptance limit…

Suppose that Joes Pharma has decided to never, ever, ship a vial of their blockbuster product if it weighs more than 28 grams or less than 20 grams. That policy could be based on science, or legend. Doesn’t matter, that’s Joes policy. Inevitably, Joe’s Pharma will produce a vial that appears to weigh exactly 28 grams. Right out of the gate, we know something specific about this particular measurement result and this vial that has landed exactly on our acceptance limit (no matter where we have placed that limit, and no matter which process we used to place it). It doesn’t matter whether we decide to accept this 28 gram result as “good” or reject it as “bad”, because to a first approximation, both decisions have the same 50% chance of being wrong. This is an inescapable fact that attaches to all measurements made at any product limit because of measurement uncertainty. Joe can ship 24 gram vials all day long without spending any time considering measurement uncertainty (and they do every day). But when at last Joe inevitably produces a 28 gram vial, we have a 50% chance of deciding wrong, no matter which choice we make; ship or reject. As we move from our best 24 gram vial to one that appears to weigh 28 grams while simultaneously maintaining that there is simply no such thing as Measurement Uncertainty, we are headed for trouble.  The existence of this increasing probability of being wrong about anything in a Pharma environment is not a fact that we would enjoy sharing with our doctors, patients or our regulators. At the moment, nothing compels global Pharmas to do so.

What’s my solution?

What do I think that global Pharmas should logically do next to move forward? 

·      They must adopt a sample space framework and start practicing Parametric Release. 

·      They should support this with method validation, SOPs, and tons of training. 

·      They should write an SOP that describes their fundamental measurement practice, one that fully describes how they produce and use measurement results including assessing estimates of Measurement Uncertainty (MU). This document should also include the criteria they use to decide when MU can safely be ignored, and the systems in place that control when it may not. This is going to be expensive. 

The first global Pharma to do this will jet light years ahead of the field. (Please see my LinkedIn profile for more on this strategic topic)

Where can we save on expenses for this improvement?

Most of the work of drafting this document has already been performed within a process that created an independent ISO guidance on Estimating Measurement Uncertainty (“The GUM”) in 1993. It has been out there for 27 years, frequently being updated and improved by a group of metrology people as qualified as any in the world.

Including or adopting an ISO document will also have the advantage of letting regulators like the FDA off an embarrassing hook because in the past they have actively supported this laughable process practice across the entire industry. FDA can choose to either rubberstamp this transition, as they should, or feel free to raise any technical problems that they detect with this ISO document (FDA doesn’t have these kinds of technical resources, or if they have, they should keep that fact very quiet!). I am pretty confident that they will raise no objections. What Pharma industry lobbyists will do in their turn will prove an interesting but separate question.

Humble bragging and blatant name-dropping

I owe a debt of gratitude to some LI colleagues. Over at Google, Cassie Kozyrkov’s wonderful Data Science posts have made me think long and hard about a lot of things including the current state of Decision Science and Data Engineering. Hadley Rees is currently “speaking truth to power” and has sharpened my ideas about how drug patents are screwing up the entire Pharma industry. Zoe Brooks is trying to develop an Analytical Lab Management support product and invited me to join her Science Advisory board. She then asked me a simple question that forced me to advance my education about the Binomial Distributions from where I left it parked upon completion of my ASQ CQE (Quality Engineer)  and CQM (Quality Manager) certifications way back in 2001. I blame her for making me learn the statistical programming language, “r”. Lastly, I always enjoy keeping an eye on Dr. Ajaz Hussain, former Deputy Director of Pharmaceutical Science at the FDA. I think of him as one of the principle architects of PAT(Process Analysis Technology), my most favorite FDA Guidance. When he left FDA in 2005, I thought that I would never hear his name again!

[1] Kieffer, Robert; 1998, “Global Trends, Needs, Issues”, PDA Journal of Pharmaceutical Science and Technology, issue # 52, pages 151-153.