Emil W. Ciurczak, DoraMaxx Consulting11.07.17
In this article, I will review two topics that are integral in a modern pharmaceutical process: flexible manufacturing and the manner in which we store, protect, use, and retrieve all our data—and apply these data to controlling the process. The word flexible may sound exciting, new, and modern—and it is all those things. However, the flip side of flexible is complexity. Complexity of hardware and complexity of the data streaming off the equipment, and the complexity of the algorithms needed to control the process.
However, before we address these topics in detail, we need to understand how much hardware and software is needed to co-develop since the “good, old days.” Back in the early 1980s, I was thrilled to be able to use an early version of HPLC integrating software, which made my life easier. However, the version we had on our LIMS (lab management information system) was written by computer experts who had no idea of what chromatography was. In this version, the cursor needed to be manually placed at the beginning and end of each peak to be analyzed. One PIXEL off the baseline and the system might integrate all the peaks as one and you could easily get a 500% assay or higher. This “tiny” flaw allowed each person to choose the area of each peak as he or she saw it, which could lead to errors or, gasp, the ability to fake data. As chemists were included in the software writing team, the results became both more accurate and more reproducible.
Clearly, primitive computer programs in the 1980s could neither monitor nor control any normal, batch-wise process, much less a complex, flexible system, which could easily be reconfigured daily. Add to that the simple, non-memory-capable computers available at that time and we have a “Leonardo da Vinci” scenario. I refer to the fact that Maestro Leonardo designed an airplane, centuries before an internal combustion engine was built. He also designed a computer, centuries before integrated circuits were built, but I digress. Computers in the 1980s didn’t even have memories; the one I was working with—NIR program—was equipped with an external 9-megabyte hard drive yet, bigger than our lab’s LIBS computer storage at that time. For comparison, a typical smart-phone has many times the capacity and speed of 35-year old computers.
Back to the crux of this article. The first half of the computer/process equation is based on good engineering and knowledge of the product line. Here’s where PAT and QbD rear their dual heads. The second half is based on excellent algorithms, hardware, and interfaces with the hardware. In truth, the production hardware fails miserably without the software. With over 2.5 million terabytes of data being created by the Pharma industry every day, the task is obviously not trivial. But, ignoring the massive amounts of data and the ability to crunch and store it, all those “0s” and “1s” mean nothing if they are used on a process based on 50-year old practices and “common knowledge.”
Remember, in the dim past when the first food and drug acts were published, there were few large commercial Pharma plants out there. Where they existed, 100,000 tablets or capsules were considered huge lots. A majority of medicinal products were still being concocted in the back rooms of your local pharmacist. Formulations were pretty much like your mom’s recipe for apple pie: peculiar to each pharmacist and his assistants. Since a patient was unlikely to use more than his/her neighborhood pharmacy, differences didn’t matter.
When the FDA was created to assure safety—only a second passage of the law added “efficacy” of the product as a requirement—of drugs, the government, in its wisdom, appointed an MD to head the agency. In turn, the head of the FDA turned to the people he was familiar with and stated that commercial production should be overseen by pharmacists. Now, the fact that a pharmacy degree at the time was a two-year curriculum with little to no inclusion of commercial hardware apparently didn’t factor into the decision. A typical school of pharmacy now grants degrees from five-year programs and most schools have sophisticated production equipment upon which they practice.
While there was a fair amount of carry-over from the back rooms of a pharmacy to small commercial lots made at a single facility, the paradigm was not built to handle more than a million lots, manufactured at multiple locations, in multiple process lines. In fact, statistically taking 20-30 dosage forms out of dozens of drums for assay is difficult and may not be done. And, even assuming that the samples sent to the lab are representative, two other minor facts need to be addressed:
Before you entrust your livelihood to a new technology, walk before you run. My approach would be to become familiar with the terms and tools of PAT, first. Attend meetings, read the literature, chat with peers, and yes, even invest in a consultant or two. I would strongly suggest attempting some of the oldest and most successful PAT steps, as a first step. Set up a raw materials program, using NIRS or Raman. The information gleaned from this will help both formulators and production staff. Simultaneously, work with your IT department to store and retrieve the spectra and data generated from this program.
When you are feeling confident in your ability to generate and crunch data (Chemometrically), show that any or all information may be called up for an Agency inspector, and have expanded your storage capacity for the onslaught of terabytes of data to come. Then you are ready for steps two through “n.”
The second step in your process stream I would tackle is blending. All you have to do is try to mix cinnamon and sugar at home to show you may not have the knowledge to do it easily. In production, we still rely on the three demo batches as a guide, never making changes, based on materials information—gleaned from the first step above. By placing a NIR or LIF monitor on your blender, and controlling it with your brand spanking new computer system, you now will always achieve perfect blends. Expand this to granulating, tableting, coating, and even packaging and you will cut your OOS lots to near zero.
All of this is based on the belief that your generated data is both easily retrieved and secure. Why secure? Recent instances of cybercrime show that hackers can get into almost any system, any time they wish. In the case of a generic manufacturer, unless you are simply copying an existing, marketed product, your formulation information is your marketing edge and you do not want to share it. For a contract manufacturing organization (CMO) generating a product for an originator, you also do not want the formulation and process parameters to become general knowledge.
So, the bottom line is make your process integral with your data handling ability. See? IT is now part of Production. Expand and modernize, but work just as hard to secure the data as you did to generate and store it.
Emil W. Ciurczak
DoraMaxx Consulting
Emil W. Ciurczak 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. In 1983, he introduced NIR spectroscopy to pharmaceutical applications, and is generally credited as one of the first to use process analytical technologies (PAT) in drug manufacturing and development.
However, before we address these topics in detail, we need to understand how much hardware and software is needed to co-develop since the “good, old days.” Back in the early 1980s, I was thrilled to be able to use an early version of HPLC integrating software, which made my life easier. However, the version we had on our LIMS (lab management information system) was written by computer experts who had no idea of what chromatography was. In this version, the cursor needed to be manually placed at the beginning and end of each peak to be analyzed. One PIXEL off the baseline and the system might integrate all the peaks as one and you could easily get a 500% assay or higher. This “tiny” flaw allowed each person to choose the area of each peak as he or she saw it, which could lead to errors or, gasp, the ability to fake data. As chemists were included in the software writing team, the results became both more accurate and more reproducible.
Clearly, primitive computer programs in the 1980s could neither monitor nor control any normal, batch-wise process, much less a complex, flexible system, which could easily be reconfigured daily. Add to that the simple, non-memory-capable computers available at that time and we have a “Leonardo da Vinci” scenario. I refer to the fact that Maestro Leonardo designed an airplane, centuries before an internal combustion engine was built. He also designed a computer, centuries before integrated circuits were built, but I digress. Computers in the 1980s didn’t even have memories; the one I was working with—NIR program—was equipped with an external 9-megabyte hard drive yet, bigger than our lab’s LIBS computer storage at that time. For comparison, a typical smart-phone has many times the capacity and speed of 35-year old computers.
Back to the crux of this article. The first half of the computer/process equation is based on good engineering and knowledge of the product line. Here’s where PAT and QbD rear their dual heads. The second half is based on excellent algorithms, hardware, and interfaces with the hardware. In truth, the production hardware fails miserably without the software. With over 2.5 million terabytes of data being created by the Pharma industry every day, the task is obviously not trivial. But, ignoring the massive amounts of data and the ability to crunch and store it, all those “0s” and “1s” mean nothing if they are used on a process based on 50-year old practices and “common knowledge.”
Remember, in the dim past when the first food and drug acts were published, there were few large commercial Pharma plants out there. Where they existed, 100,000 tablets or capsules were considered huge lots. A majority of medicinal products were still being concocted in the back rooms of your local pharmacist. Formulations were pretty much like your mom’s recipe for apple pie: peculiar to each pharmacist and his assistants. Since a patient was unlikely to use more than his/her neighborhood pharmacy, differences didn’t matter.
When the FDA was created to assure safety—only a second passage of the law added “efficacy” of the product as a requirement—of drugs, the government, in its wisdom, appointed an MD to head the agency. In turn, the head of the FDA turned to the people he was familiar with and stated that commercial production should be overseen by pharmacists. Now, the fact that a pharmacy degree at the time was a two-year curriculum with little to no inclusion of commercial hardware apparently didn’t factor into the decision. A typical school of pharmacy now grants degrees from five-year programs and most schools have sophisticated production equipment upon which they practice.
While there was a fair amount of carry-over from the back rooms of a pharmacy to small commercial lots made at a single facility, the paradigm was not built to handle more than a million lots, manufactured at multiple locations, in multiple process lines. In fact, statistically taking 20-30 dosage forms out of dozens of drums for assay is difficult and may not be done. And, even assuming that the samples sent to the lab are representative, two other minor facts need to be addressed:
- The tests are decades out of date and may or may not show more than the ones being assayed contain the allowed levels of API. The other “tests,” such as dissolution, hardness, weight variation, etc., give us data to fill in the blanks on the Master Manufacturing Formula (MMF). However, remember, data is not always information.
- However good or bad these tests may be, they cannot actually affect the outcome of the lot being assayed. After-the-fact analyses can only tell the company to sell or destroy an already completed batch. It’s not too expensive with a 100,000-tablet lot, but 5,000,000 tablets? Now that’s another story. Additionally, the fact that a process was Out-of-Specification (OOS) is all that is determined, not where it went wrong of how and when it could have been corrected to avoid a failing batch.
- They didn’t know as much about the physics of their process as they thought. Yes, physics and physical chemistry. The “traditional” chemistry is all in the synthesis of the API, which today, is seldom done in-house, anymore.
- They needed to expand their personnel repertoire. Aside for the traditional formulators (pharmacists), they were beginning to recognize the need to ether hire or consult with material scientists, process engineers, statisticians, Chemometricians, and so forth.
- The analytical tests they had long depended upon to assure a quality product were either not giving the answers they thought they were, were not timely—not fit for controlling a process—or both.
- They needed new technologies, skills, and a way to gather, crunch, and utilize the mounds of data generated by these new technologies. For example, a spectrometer can generate a new spectrum every few milliseconds, with its throttle wide open.
- They also needed computers and skilled operators and programmers. Obviously, the computers are “store-bought.” The operators of the computers may be hired or the skills taught to existing professionals, but few companies start out with either the people or the experience to install the monitoring/controlling systems on day one. So, more people and/or outside experts are needed.
Before you entrust your livelihood to a new technology, walk before you run. My approach would be to become familiar with the terms and tools of PAT, first. Attend meetings, read the literature, chat with peers, and yes, even invest in a consultant or two. I would strongly suggest attempting some of the oldest and most successful PAT steps, as a first step. Set up a raw materials program, using NIRS or Raman. The information gleaned from this will help both formulators and production staff. Simultaneously, work with your IT department to store and retrieve the spectra and data generated from this program.
When you are feeling confident in your ability to generate and crunch data (Chemometrically), show that any or all information may be called up for an Agency inspector, and have expanded your storage capacity for the onslaught of terabytes of data to come. Then you are ready for steps two through “n.”
The second step in your process stream I would tackle is blending. All you have to do is try to mix cinnamon and sugar at home to show you may not have the knowledge to do it easily. In production, we still rely on the three demo batches as a guide, never making changes, based on materials information—gleaned from the first step above. By placing a NIR or LIF monitor on your blender, and controlling it with your brand spanking new computer system, you now will always achieve perfect blends. Expand this to granulating, tableting, coating, and even packaging and you will cut your OOS lots to near zero.
All of this is based on the belief that your generated data is both easily retrieved and secure. Why secure? Recent instances of cybercrime show that hackers can get into almost any system, any time they wish. In the case of a generic manufacturer, unless you are simply copying an existing, marketed product, your formulation information is your marketing edge and you do not want to share it. For a contract manufacturing organization (CMO) generating a product for an originator, you also do not want the formulation and process parameters to become general knowledge.
So, the bottom line is make your process integral with your data handling ability. See? IT is now part of Production. Expand and modernize, but work just as hard to secure the data as you did to generate and store it.
Emil W. Ciurczak
DoraMaxx Consulting
Emil W. Ciurczak 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. In 1983, he introduced NIR spectroscopy to pharmaceutical applications, and is generally credited as one of the first to use process analytical technologies (PAT) in drug manufacturing and development.