When it comes to believing Certificates of Analyses (CoA), to quote the general from Star Wars, “It’s a trap!” This is especially true when attempting to integrate a new supplier’s goods into an existing product. Now, if you have been using vendor “A” for years and have been producing a quality product, the CoA is probably irrelevant. What you are assuming (hopefully correctly) is consistency over time, not the numbers on a page. Odds are a second lab, using the same analytical equipment may well generate different numbers, but the final product you generate would not change.

Most excipients are synthetically made, derived from natural products or synthetically modified natural products, and are available from multiple vendors. As a result, their physical/chemical properties can vary from lot to lot and vendor to vendor. Excipient variability has been studied and, in some cases, found to have a significant effect on finished product quality. The differences may not be apparent using “time-tested” (aka, old and non-specific compendial) methods, usually only applied when a CoA is not available. And then obtaining results based on a poorly designed sampling protocol. So, let’s look at the two areas where we need to be better.

1. Proper sampling

The broadly accepted method involves sampling some of the containers, usually sampling √N + 1 (e.g., for 200 bags of lactose, we only sample 16 bags) and then seldom done correctly. Books are written and courses given on proper sampling practices, but, in my more than 50 years in the pharma industry, I have seen very few instances of “proper” sampling techniques. For example, Ciba in Basle had a team, led by a PhD, performing sampling for both chemical and physical characteristics as well as for sterility. In defense of U.S. manufacturers, European companies obtain raw materials by rail, allowing for the aforementioned lactose in two or three containers, making sampling a wee bit simpler.

One potential problem with “statistical” sampling procedures is that you can miss a bad container, which could even be the wrong chemical. When I was developing the first NIR method for qualifying raw materials, I used individual grab samples from 220 bags of USP Lactose. The QC lab did the √N + 1 sampling, then analyzed three aliquots of the mixture. The mixed sample passed all the USP mandated tests, but the NIR qualitative method failed two of the samples when tested bag-by-bag. These were individually retested with USP tests and one failed moisture (strangely “anhydrous” means less than 4% moisture) and the other failed on the pH test. Were these added to all the other bags, a mixed sample would have masked the outliers.

All too often untrained operators or warehouse workers sample the requisite number of drums (only 11 per 100) for both excipients and finished products. Powders will segregate (both pure materials and mixtures/granulations) with finer particles moving to the bottom, while larger particles move to the surface. I have witnessed the operators merely using a scoop to remove surface materials—both excipients and finished products—and place them in a plastic bag.

Even when a sample thief is used, there is still the problem that 89% of the batch remains unsampled! And, when the thief is used for mixtures (blends, granulations), QC lab techs may merely use a spatula to take aliquots from the (by now) separated sample. Some ways to repair these problems:

1.  Sample EVERY container, not just some (this is where NIRS or Raman comes in handy).
2.  When using a thief, take samples from multiple locations, at multiple levels.
3.  Use the entire contents of each pocket within the thief (or mix it well), not merely take aliquots from the bag for mixes or granulations. Pure excipients can be mixed, since the average particle size is what is needed.

2. Performance properties

All too often, products are designed using only one or two suppliers of excipients and APIs. The need to get to market, because long development times, clinical trials, etc., all cut into the patent life of the drug. Considering the expense of R&D, it is understandable that we don’t spend a lot of time sourcing multiple alternate vendors. Since the current way that an alternate source can be approved involves making the dosage form with the new material and performing all sorts of stability and performance tests. This could keep a product off the shelves for weeks or months. For a single-source product, this could be a serious health issue for patients; for a generic product, a company chances losing a large part of its market share to competitors.

If, however, a company is using a QbD protocol for their production of a product, they would already need to have in place a number of meaningful parameters for an ingredient to pass. These are based on a design of experiment protocol and seldom include what are currently being done. Rather than just static tests (porosity, density, sieve analysis), PAT/QbD specs are based on performance criteria specific to each process step.

For example, mixing the components of a dosage form is the first step, but determines where the overall production will wind up. Imagine, if you turn South on the interstate when your destination is to the North, you might get there, but it will take far more effort and time. No matter which blender is used, the general idea is that several disparate powders are loaded into the device and some form of tumbling is used to affect a homogeneous mixture (of course, only a solution can be homogeneous, but tradition “is what it is”). Under GMP, you are required not to think, er, um, I mean required to strictly follow the written instructions (MMF). E.g., “mix for 20 minutes at 15 rpm, then shut power off” whether or not it is under-mixed, over-mixed, or perfectly mixed. Under PAT/QbD, you blend until the proper blend is shown to have been accomplished.

If you still insist on using GMP production methods (one step, transfer the materials, second step, transfer, and so on), consider using semi-PAT/QbD measures. It may convince you to go full tilt when you see how well it works. One example is an instrument designed at Rutgers by Dr. Muzzio’s group to check the “flowability” of pure powders or mixtures (see Figure 1 in the image slider at the top of this article). A large, transparent tube (1 meter long, .5 meter in diameter) is loaded to approximately half full, set on its side on rollers, and tumbled slowly. Lights are set at one end and an array of detectors at the other. As the tube tumbles, the powder moves up the side of the tube and, depending on the particle-to-particle friction, the height of the powder up the wall will differ.

Hence the “flowability” is determined and the mixing process adjusted appropriately.

The most common USP tests were designed for pharmacists to be run in their back rooms where they made meds for customers. Many of these “quick and dirty” tests were mainly to assure that the correct components were being used NOT for processability, because the powders were either wet, pilled, and dried or simply weighed into capsule shells by hand. Therefore, there was no need for the USP to generate any such tests.

Why are physical properties so crucial, you might ask? The API has a few critical parameters:

1.  The correct salt determines the pharmacological effect. That’s an easy one: NIR or Raman.
2.  The correct particle size affects both the homogeneity and speed of absorption into the blood stream. (Numerous ways, but I know, by experience NIR can be calibrated to measure average particle size.)
3.  The polymorphic form will affect drug solubility and uptake. (E.g., poorly soluble drugs need to be in the amorphous form.)

As far as the flowability of the blended materials, there are some major consequences for mixtures that flow poorly.

Cavitation:
a.  in the hopper will slow the process;
b.  in any conveyor stop the process, or, worse yet;
c.  in the tablet die will give seriously variable fills (and widely varying tablet weights).

Clearly, we need to move to meaningful tests for raw materials, based on performance, not outdated tests, designed just to prove the materials are what they claim to be. To do that, we also need to understand the basics of proper sampling theory. Many, many books have been written on sampling. Buy one and read it!

At some point, just about all products will be made by continuous manufacturing (CM) techniques. What is needed is the impetus; perhaps the price limitations that many countries are imposing will spur companies to move to QbD and CM protocols to maximize profits and quality. In advance of this endpoint, all personnel and hardware will need to in place. What is needed?

1.  Familiarity with Design of Experiments concepts and software.
2.  Acquiring and gaining experience with real-time processing monitoring equipment (i.e., NIR, Raman, etc.).
3.  Planning ahead: Who is needed? What space is needed? What hardware, software is needed? Which products are first?

In the words of the late Lee Iacocca, “Lead, follow, or get out of the way.” Oh, and by the way, almost none of the needed information can be found on the good, old CoA. Just saying. 

---

Emil W. Ciurczak, DoraMaxx Consulting, has worked in the pharmaceutical industry since 1970 for companies that include Ciba-Geigy, Sandoz, Berlex, Merck, and Purdue Pharma, where he specialized in performing method development on most types of analytical equipment. Email: [email protected]