Lou Bellafiore, Kevin Henretta, John Walker, Chris DeMello*
TechniKrom Inc., 1801 Maple Avenue, Evanston, IL 60201
* Centocor Inc. 200 Great Valley Parkway, Malvern, PA 19355

Abstract
The FDA's recent regulatory initiative, entitled OcGMPs for the 21st Century - A Risk-based Approach, has thrown the spotlight on the application of manufacturing science to pharmaceutical manufacturing processes. A more rigorous approach can be expected to unearth deficiencies in pharmaceutical and biopharmaceutical processes. This paper addresses one such area that has broad implications, namely, the concentration variations of bulk-mixed reagents such as buffers and solvents blends. We shall also discuss how the application of manufacturing science offers superior solutions to such problems, including better equipment designs and high product quality by design, while simultaneously improving ailing pharmaceutical manufacturing productivity dramatically.

An Industry in Major Transition
In 2002, the FDA announced a new initiative, called "Pharmaceutical cGMPs for the 21st Century - A Risk-Based Approach". This was the first major change in FDA drug regulation strategy in 25 years. Specifically, the change is aimed at:
a) Producing safer drug products
b) Creating a more efficient regulatory process.
c) Stimulating Continuous Process and Product Improvements

Once a pharmaceutical process reaches a certain stage of development, the gears of continuous improvement tend to seize. Understandably, part of this reaction is due to the perceived frigidity of the regulatory climate toward adopting new technology. Consequently, the pharmaceutical industry has been much slower than other hi-tech industries to approach manufacturing as a science, and no longer as an art. The FDA asserts that consistent and optimum product quality can be guaranteed only by first fully understanding and controlling the relevant processes.

Figure 1: The combination of modern quality management and risk assessment with
manufacturing science will make available more, better, safer and less expensive drugs.

Gaining such a comprehensive understanding of the process/product interrelationships will require drug manufactures to apply the tools of manufacturing science. To ensure the ongoing, consistent application of the knowledge so gained will require the application of modern Quality Management Systems (e.g. ISO 9001:2000) while the use of modern Risk Assessment tools will better focus the resources of both the drug maker and the FDA onto those areas that most affect patient risk. The benefits do not fall solely to the patient. Significantly less onerous regulatory oversight at all stages of drug manufacturing development awaits those drug manufacturers who can demonstrate in-depth understanding of how and why their processes produce good, consistent product. Figure 2 represents five levels of process understanding. The highest level, mastery of the process, is based on an understanding of the process 'from first principles'.

Figure 2: Current pharmaceutical manufacturing processes are
often based on a primitive understanding of the process.

The industry is operating at around Level 2 on this hierarchy, which probably accounts, in part, for why the industry is also running at an astoundingly low 20 - 30% manufacturing efficiency.

At lower levels of process understanding, unexpected outcomes in purity and yield, especially for multi-variant biological products such as vaccines or some therapeutic proteins, often arise. Consequently, extensive testing of the final product has been necessary to segregate out the bad material before delivery to the customer. This improves neither the product nor the understanding of the process. No other modern industry accepts the poor first pass yields that in our industry mandate extensive testing, pooling and re-work to achieve delivery quotas. This expensive, non-value added work is referred to as the "Hidden Factory", one of the major costs of poor quality (C.O.P.Q.) which can contribute as much as 50% of the manufacturing cost of goods sold (C.O.G.S.).

The industry's best remedy against rising COPQ is improved process know-how. As a deeper understanding of the relationship between the key process parameters and product quality is developed, a drug manufacturer ascends the process knowledge pyramid. An optimized process, controlled in real-time, pre-empts the need for the hidden factory and enables real-time product release while simultaneously reducing patient risk.

Process Analytical Technologies (P.A.T.) is a sub-initiative, promoted by the FDA, which fosters a toolset of measurement technologies that enables this real-time process control. Successfully implementing PAT requires the applications of three disciplines:
i) Manufacturing science to gain profound process understanding.
ii) Sensor/instrument technology to measure appropriate process variables.
iii) Control engineering to select and apply the best control strategy.

Six Sigma is a popular systematic approach to manufacturing sciences that is widely used in many industries. It is used to gain in-depth understanding and optimum control of processes. It is designed to systematically analyze and optimize the processes that drive product quality, by eliminating, or eliminating the opportunity for, variation and defects.

To do this the Six Sigma approach is designed:
i) To ensure that we measure all of the right parameters (known as key process output variables or KPOVs) in the right way (i.e. correctly use competent metrology), at the right places and right times (known as critical control points (CCPs).
ii) To then identify all the important, causative key process input variables (KPIVs) that govern each KPOV and to quantify the contribution of each KPIV to the final value of each KPOV over a broad range of operation. Once this is achieved we have all the knowledge required to achieve true process control.
iii) To reduce the variability of each KPIV in order to reduce the resultant variability in each KPOV, thereby tightening up the process control achieved.

Figure 3 shows two processes of differing capabilities. If the individual run-to-run distributions of a KPOV are aggregated the more-important long-term process distribution (shown in red) can be constructed. The diagonal, parallel lines represent upper and lower acceptable product specification limits for the variable. Distributions that consistently fall within the specification limits illustrate a capable process.

The application of the tools of Six Sigma, (such as Process Mapping, Pareto Analysis, Design of Experiments (D.O.E.), Taguchi Loss Function, Cause and Effect Diagrams, Quality Function Deployment (QFD), and Failure Mode and Effects Analysis (FMEA), transforms the unstable, variable process on the left to the same high capability as the one on the right.

Figure 3: Reducing or eliminating process variability will result in a stable, capable process.

This paper recognizes three levels of process control:
i) Manual (no automation)
ii) Fixed (dumb) automation, and
iii) Adaptive (smart) automation.

At the first level, manual processes are developed which depend upon the skill and vigilance of a human being. At the level of quality at which we aspire to work, simple manual operator control is insufficient. The next level of control implements fixed, or "dumb" automation of manual operations, in a simple mechanical way. It lacks the ability to accommodate real-time input variability. The ultimate or third level of process control is Adaptive Automation, which uses judiciously selected and placed real-time sensors to detect trends in real time towards out-of-specification (O.O.S) conditions. It further has the inbuilt intelligence to automatically, and in real-time, correct for them adaptively.

Shortcomings of Manual and Fixed Automation Blending of Reagents
Traditional large scale pharmaceutical manufacturing operations utilize manual or fixed automation to prepare large volumes of each pre-mixed reagent they are going to use, and store them in large tanks. Such production-scale mixing of process fluids is much more difficult than at the bench-scale where they are typically developed and optimized. Even the largest pharmaceutical manufacturers are burdened with a variability of the blends that is usually +/- 2 to 5% (a variability of up to 10%), and occasionally as bad as +/- 10%. Many individual mixing protocol steps can achieve no better than +/- 2% accuracy, e.g. the addition of large volumes of water. Another limitation is the measurement variability associated with large volume reagents that includes instrument resolution, sampling error and intrinsic temperature and concentration gradation within tanks, which together can add another 2-3% error.

Fig4: Variability in a typical process today starts with the buffer or solution
make-up, and the effects of that variability ripple through the entire process.

The effect of the feedstock variability is additive to the overall process variability. Process variability in turn is reflected in product variability in purity, potency, yield, recovery, etc. Large variability necessitates an extensive Hidden Factory and prompts closer regulatory investigation. Some manufacturers have used over-sized equipment to process an entire batch as one lot to minimize variability within a campaign. This significantly increases capital required (for both equipment and resin in LC), product risk (all of one's eggs…), and handling difficulties (due to equipment size and weight). It also eliminates the opportunity for gaining improved process knowledge from multiple runs.

Popular Fixed Automated Blending Approaches
Let's look at a couple of representative automated blending approaches in wide use today. At first glance they both appear seductively simple and self-explanatory. Both approaches indicate that their designers were unaware of, and/or unable to address, the +/- 2%-5% variability inherent in reagent feedstocks.

Fig 5: Approach 1 features two pumps, each controlled via its own mass flow meter.

In approach 1, it would be expected that these extremely accurate (+/- 0.1%) mass flow meters would permit equally accurate blending. Not so. The blending accuracy depends heavily upon where within the total flow range it operates and what % blend composition is being made. It has a limited "sweet spot" within which it adds a further +/- 3 to 5% of blend composition variation to the input feedstocks. Outside this sweet spot variation worsens.

Fig. 6: Approach 2 to blending uses a mass flow meter, a main pump and a digital proportioning valve.

Approach 2 shown in Fig. 6 adds up to 4%-5% of extra variation to feedstock variation. To improve this one manufacturer developed an upgrade. This involved the laborious measurement of the error profile to create a process- and equipment-specific look-up table of offsets by which the PLC controller was then instructed to modify the target process value. While this compensates for the systematic mechanical errors introduced by the blending configuration it is incapable of reacting to random equipment variations or input feedstock variations.

Among several reasons why these approaches give poor control are that feedstock compositions are not accurate, the mixing hardware is a source of error and the mixing physics across the compositional range of blends are not linear.

Applying Adaptive P.A.T.™ to Liquid Blending
For several years TechniKrom has used an approach to accurate and reproducible blending that fundamentally differs from the popular trend. An analysis of the overall process helped to clearly identify its intent; KPOVs are blend accuracy and reproducibility. This made it possible to then determine the best control strategy along with its critical control points.

In selecting the appropriate control strategy, we determined that we would guarantee failure if we assumed that bulk-mixed feedstock compositions (KPIVs) were accurate. We also recognized that it would be essentially impossible to identify and then individually measure and accurately control the myriad potential sources of feedstock variability.

Our solution was a PLC-controlled Adaptive P.A.T.™ system based upon the accurate, real-time feedback and adaptive control of the instantaneous composition of the blend as it is being created. We monitor an accurate, locally tuned, first-order measure of composition. Then, by preventing the release of blend to the process until its composition accurately matches the current set point, the blend accuracy delivered to the next stage of the process is guaranteed. Thus we have created point-of-use blending with an accuracy and reproducibility exceeding 0.1%: up to 100X more accurate than today's most commonly used approaches simply by using widely available, highly reliable, industrial grade automation technology.

An Illustrative Application - Process-Scale Chromatography
The resolving capability of many chromatographic processes can be severely compromised through process variability. All the systematic causes of inefficient, inaccurate and non-reproducible separations can be traced to the increase in variability of some key process input variables (KPIVs).

Fig. The accuracy and precision of the mobile phase has a large impact on a
chromatographic separation procedure.

A small decrease in variability of the mobile phase can often leverage a large improvement in product resolution and reproducibility, which, in turn, dramatically affects product quality and profitability. Thus blend accuracy and reproducibility are both leveraged KPIVs and, hence, prime candidates for accurate process control.

By applying Adaptive P.A.T.™ - based blending to the creation of 0.1% accurate and reproducible mobile phase blends, it is possible to transform the process as illustrated below:

Fig. Adaptive PAT technology can be used to create very
stable and reproducible liquid chromatography processes.

Besides eliminating several hidden factory elements (tank farm, excessive QC testing and pooling/rework of product fractions) the increased resolution and process 'noise' elimination permits a much clearer view of the underpinning process that is so essential for even better process understanding and improvement.

By programming the LC system to not release the mobile phase to the column until it is in specification prevents failure from all the myriad potential causes of either isocratic or gradient inaccuracy. Adaptive control thus enables the most precise and cost effective resolution of product from impurities and also provides extremely valuable insurance against unintentional loss of the (often very expensive and/or difficult to replace) product. The potential business impacts associated with such losses usually dwarf the cost of the complete equipment associated with preparative scale-LC (e.g. each day a phase III trial is delayed can cost $1million U.S.).

This approach to blending is the only method that is truly portable from bench-scale through to full-scale manufacturing. Using an Adaptive P.A.T.™ -based chromatography system during the process development permits both meaningful parametric optimization, seamless technology transfer and guaranteed process scale-up.

The Bottom Line
The economic impact of an improved, reproducible process depends upon the value of the product being produced and the degree of improvement achieved.

A case study was performed to quantify the impact of using accurate and reproducible gradients using data from an actual high-value manufactured product. This study used a very conservative first-order estimate of the benefits of the improvements recognized by an FMEA (failure mode and effects analysis) study. As shown above, for the current popular approach, poor resolution has by far the most expensive consequences.

Using an isocratic or step gradient approach rather than an accurate linear gradient to separate closely related species (a typical attempt to create process robustness when existing equipment is incapable of generating controlled linear gradients) can easily reduce first pass recovery to below 70%. Pooling and re-working the fractions containing the remaining 30%, assuming similar recovery efficiency, would yield a total of 70 + 21= 91% recovered.

This pharmaceutical client manufactures product worth $200M USD per year. 9% of that would be lost and more crude starting material would be required to make up for this. 9% of $200M = $18M.

With the 0.1 % gradient accuracy capability the first-pass recovery is at 90% or higher. After reprocessing the side-cuts the recovery is 90 + 9 % = 99%, an improvement in recovery of 8%. This translates to an improvement of $16M per year, due directly to the improved gradient capability.

In this case study the owner also stated that assay samples taken throughout his total process chain represented 10% of the total drug produced, another hidden cost of the QC-based approach, ($20M/annum). In short Adaptive P.A.T.™ process controls makes very sound business sense.

Pharmaceutical Manufacturing Success in the 21st Century
The transition to a manufacturing science based approach requires new equipment with the accuracy and resolution we have discussed.

The Adaptive P.A.T™-based blending approach is widely applicable.
It can be:
i) Integrated into new LC equipment,
ii) Used as a freestanding blending system to blend or adjust large tank feeds.
iii) Retrofitted to existing equipment to provide significant economic and regulatory performance upgrades.
iv) Modified to control other aspects of process fluids such as temperature and pH

Thus this approach is expected to have a far-reaching impact on the quality of all pharmaceutical processes involving process liquids.

Conclusions
The FDA initiative "Pharmaceutical cGMPs for the 21st Century - A Risk-Based Approach" will have a far reaching impact on product quality, and manufacturing costs in the pharmaceutical and biotechnology industries. Its aims are fully congruent with those of Six Sigma.

Highly accurate (+/- 0.1%) point-of-use blending, including gradient capability, has been available for several years and can be used to upgrade a broad range of pharmaceutical processes through reducing, or in many cases completely eliminating, process variability and inaccuracies.

Adaptive P.A.T.™ controls prevent catastrophic product losses.

Variation in feed streams, i.e. the +/- 2-5% bulk blend variability, has not been fully acknowledged and so its impact is just starting to be calculated and assessed.

Manufacturing Science makes excellent business sense. It is a powerful tool for improving the industryıs productivity and reducing C.O.G.S.

By improving process controllability during blending operations, Adaptive PAT™ helps fulfill the FDA vision of reducing patient risk.