Evaluating Future Biopharmaceutical Inspection Needs, Infrastructure Capability Gaps, and Technology Development Strategies by Marnix E. Hollander B.S. Mechanical Engineering, Columbia University, 2007 Submitted to the MIT Sloan School of Management and Mechanical Engineering Departments in Partial Fulfillment of the Requirements for the Degrees of Master of Business Administration and Master of Science in Mechanical Engineering in conjunction with the Leaders for Global Operations Program at the Massachusetts Institute of Technology ARCHIVES June 2012 C 2012 Mamix E. Hollander. All rights reserved. The author hereby grants to MIT permission to reproduce and to distribute publicly paper and electronic copies of this thesis document in whole or in part in any medium now known or hereafter created. Signature of Author -____ DepiUment of Mechanical Engineering MIT Sloan School of Management May 11, 2012 Certified by Roy Welsch Eastman Kodak Leaders for Glo ations Professor of Management, Engineering Systems Division Thesis Supervisor Certified by Douglas Hart Professor, Mechanical Engineering Thesis Supervisor Accepted by t'%%1a rdt, Chair, Department of Mechanical Engineering Professor, Aeronautics and Astrogautics and Engineering Systems Division Accepted by 'Maura flerson, Director of MBA Program MIT Sloan School of Management This page intentionally left blank. 2 Evaluating Future Biopharmaceutical Inspection Needs, Infrastructure Capability Gaps, and Technology Development Strategies by Marnix E. Hollander Submitted to the MIT Sloan School of Management and the Mechanical Engineering Department on May 11, 2012 in Partial Fulfillment of the Requirements for the Degrees of Master of Business Administration and Master of Science in Mechanical Engineering Abstract The biotechnology industry is undergoing a paradigm shift in the properties of the manufactured drug product. As therapeutic protein formulations change from agonist to antagonist methods of action, the concentration of proteins in each dose increased by orders of magnitude, and with it, the challenge of producing and inspecting the product. Current inspection technologies meet the requirements to properly inspect the existing drug product portfolio, but as new pipeline drugs enter commercial manufacturing, inspection will become a pressing issue from operational efficiency, compliance, and quality standpoints. It is known that the properties of some of these pipeline products render them "uninspectable" by currently installed Automatic Visual Inspection Machines (AVIMs) but the scale and scope of the challenge is not well defined currently. The process and approach outlined in this research focuses on distilling large datasets of future product forecasts together with product attributes and infrastructure capability to generate a quantitative understanding of the future challenge. Through this research, product attributes critical to inspection such as viscosity, presentation, and protein aggregation, are identified for each commercial and pipeline drug product. These attributes are paired with production forecasts to generate attribute focused inspection requirements through 2020, which are then mapped against current capabilities. Gaps identified between requirements and current infrastructure capabilities are determined and the scale quantified. These capability gaps are then segmented by potential solutions, complexity of solutions, and cost of inaction in order to give Amgen the best foresight into future decisions and investments. Based on the gaps identified and near term engineering challenges, several solutions are developed, proposed, and tested throughout the internship. These include the use of Surface Acoustic Waves (SAW) to agitate heavy particles into suspension through acoustic streaming, enhanced lighting and imaging techniques to better identify particles, and segmented machine vision algorithms. These approaches are part of a larger portfolio of technical solutions which must be developed to address future product attributes which render current inspection processes ineffective. Thesis Supervisor: Roy Welsch Title: Eastman Kodak Leaders for Global Operations Professor of Management, Engineering Systems Division Thesis Supervisor: Douglas Hart Title: Professor, Mechanical Engineering 3 This page intentionally left blank. 4 Acknowledgments Firstly, the author thanks internship supervisor Erwin Freund for his guidance at Amgen and patience in introducing the author to the entirely alien world of biotechnology. Many thanks are also warranted to Graham Milne, and Ryan Smith, for sharing their valuable office and lab space, as well as contributing to much of the research and insight this document is based on. The work in this thesis was further made possible by Leticia Saldain, Marc Doble, Oscar Gonzalez, Ron Forster, Deborah Shnek, Darcy Rosado, Miguel Carrion, the group formerly known at the RAT group, and many others at Amgen by their support. Thesis advisors Roy Welsch and Doug Hart were also invaluable as guides and compasses during the internship. Further thanks are warranted to his colleagues in the Leaders for Global Operations program, especially his fellow interns in Thousand Oaks, whose collaboration ensured the project was successful and the internship fun. The author cannot forget his significant other, Tracy, who endured many lost evenings and weekends to this work with patience and stoicism. Finally, the author would not even be able to read a thesis without the lifetime of unwavering support from his family, and to them he is always grateful. 5 This page intentionally left blank. 6 Table of Contents A b stract ......................................................................................................................................................... Acknow ledgm ents.........................................................................................................................................5 Table of Contents ........................................................................................................................................... List of Figures ............................................................................................................................................... Glossary ...................................................................................................................................................... These terms which may not be explicitly defined in the text but are useful in the context of the research presented. .................................................................................................................................................... 1 Thesis Introduction, Problem Statem ent and Internship M otivation ............................................... 1.1 Thesis Structure...........................................................................................................................13 1.2 Project Goals and General Approach..................................................................................... 1.2.1 Future Inspection N eeds (FIN ) Model ............................................................................ 1.2.2 Exploring technical solutions.......................................................................................... 1.2.3 Cost of inability to inspect .............................................................................................. 2 A m gen and Biotechnology Industry Background .......................................................................... 2.1 Biopharm aceutical Industry ................................................................................................... 2.2 3 7 9 10 10 11 13 14 14 15 16 16 A mgen Inc...................................................................................................................................17 2.3 Inspection at A mgen ................................................................................................................... 18 2.3.1 M anual inspection ............................................................................................................... 18 2.3.2 Com m ercial Autom ated V isual Inspection ................................................................... 20 2.4 Particulate m atter ........................................................................................................................ 22 2.4.1 Extrinsic particles................................................................................................................22 2.4.2 Intrinsic particles.................................................................................................................22 2.5 Drug Product Engineering Advanced Science and Engineering Group..................................24 2.6 Current capabilities across international facilities................................................................. 26 2.6.1 A lignm ent between clinical and various m anufacturing sites ........................................ 26 2.6.2 Im plications of redundancy strategy on inspection........................................................ 26 2.7 A m gen's inspection strategy as com petitive advantage........................................................ 27 2.7.1 Regulatory uncertainty ................................................................................................... 27 2.7.2 Inspection as com petitive advantage............................................................................... 27 2.7.3 Future state goal..................................................................................................................28 3 Future Inspection N eeds and Infrastructure Capabilities M odel.................................................... 29 3.1 Approach ..................................................................................................................................... 3.1.1 M otivation ........................................................................................................................... 3.1.2 M odel Param eters................................................................................................................30 3.1.3 Data gathering ..................................................................................................................... 3.2 M odel operation .......................................................................................................................... 3.3 29 29 33 34 M odel Outputs.............................................................................................................................35 3.3.1 3.3.2 Inspection needs..................................................................................................................35 Infrastructure capabilities............................................................................................... 37 3.3.3 Cost of inability to inspect .............................................................................................. 39 3.3.4 Gap m itigation analysis and recom m endations............................................................... 3.3.5 M odel expansion and scalability................................................................................... 4 Engineering solutions to critical challenges................................................................................... 4.1 Technical challenges addressed and solutions proposed......................................................... 4.1.1 Im aging high density particles........................................................................................ 4.2 A coustic stream ing......................................................................................................................47 4.2.1 Introduction.........................................................................................................................47 4.2.2 Experim ental setup..............................................................................................................48 7 40 43 44 45 45 4.2.3 4.2.4 Testing.................................................................................................................................49 Results.................................................................................................................................50 Discussion ........................................................................................................................... 4.2.5 Segm ented inspection regimes................................................................................................ 4.3 5 Recom m endations and Conclusion ................................................................................................ Operational Im provem ents..................................................................................................... 5.1 G lobal Infrastructure and Vendor Relationships............................................................ 5.1.1 Technology Investm ent ............................................................................................................... 5.2 5.2.1 H igh viscosity products................................................................................................... Presentation m ix..................................................................................................................58 5.2.2 5.2.3 Proteinaceous products................................................................................................... 5.3 Conclusion .................................................................................................................................. 6 References ........................................................................................................................................... 7 51 52 54 55 55 56 57 58 60 61 Appendix.............................................................................................................................................63 7.1 High Density Particle Detection M odule Perform ance.......................................................... 63 7.2 FIN M odel Data A ggregation and Distillation Code .................................................................. 64 7.2.1 Com mercial and pipeline drug product im port: .............................................................. 64 7.2.2 Breakdown by protein..................................................................................................... 64 7.2.3 A ggregation by presentation ............................................................................................ 64 7.2.4 A ggregation by viscosity................................................................................................. 65 7.2.5 A ggregation by protein aggregates ................................................................................ 65 7.2.6 Inspectability ....................................................................................................................... 65 Visual Basic Macro for commercial product forecast import ........................................ 7.2.7 7.3 Net present cost of m anual inspection of various capability gaps .......................................... 8 65 69 List of Figures Figure 1 -SKU proliferation of a single protein product, with fill stage highlighted and unlabeled, filled v ia l sh own ................................................................................................................................................... 11 19 Figure 2 - Phoenix Imaging manual inspection booth ............................................................................ 19 Figure 3 - Manual inspection performance by particle size and study [8]............................................. 21 Figure 4 - Static Division (SD) sensor operation[10] ............................................................................ 21 Figure 5 - Spin-brake-inspect cycle[10]................................................................................................. Figure 6 - Basic states of proteins and advancement of protein monomer to aggregated protein particle.23 Figure 7 - Plot of viscosity vs. concentration for various drug products ............................................... 31 Figure 8 - AVIM particle detection rate vs. viscosity at various drug product mimic viscosities and AVIM settin gs ........................................................................................................................................................ 31 Figure 9 - Protein aggregates in drug product........................................................................................ 32 Figure 10 - Flow of FIN model data inputs............................................................................................ 33 Figure 11 - GUI for product forecast data loader................................................................................... 35 Figure 12 - Percent of filled units per year which are not "traditional" 3cc glass vials or Icc glass syringes ..................................................................................................................................................................... Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure Figure 13 14 15 16 17 18 19 21 20 22 23 24 36 - Portfolio viscosity ranges by year........................................................................................ 37 - Filled units uninspectable by currently validated processes............................................... 37 - Filled units with observed protein aggregation ................................................................... 39 - FTE's and associated cost required for manual inspection................................................. 40 - Potential solutions to implementing AVI for future capability gaps .................................... 41 - Bottom of glass vial showing curved base and potential for hidden particles...................... 45 - Particle obscuration by base curvature ................................................................................. 45 - High density particle detection module ............................................................................... 46 - Diagram showing angled camera and lighting setup................................................................46 - Experimental setup for acoustic streaming in glass vial...................................................... 48 - Diagram of fluid flow and sequence of toroidal streaming stills........................................ 50 - A 100um glass bead and an air bubble manipulated about a node by varying wave amplitude ..................................................................................................................................................................... 50 Figure 25 - Sequence of nodal, then vortex steaming stills.................................................................... Figure 26 - Segmented inspection zones................................................................................................. Figure 27 - Detection scores of high density particles ............................................................................ 51 52 64 9 Glossary These terms which may not be explicitly defined in the text but are useful in the context of the research presented. Acoustic Streaming - This is the phenomenon of driving fluid using traveling surface acoustic waves. As the acoustic waves dissipate into the fluid, their energy is transferred into the fluid generating flow. Automated Visual Inspection Machine (AVIM) - An inspection system for filled drug product which requires no input by the user to determine individual accept or reject classification for a given unit. This machine may look for cosmetic defects, color, turbidity, or particulate matter. Cosmetic Defect - This includes scratches, dents, tears, misprints, cracks, or any defect visible to the eye which may or may not affect the integrity of the product. Eject - A classification of a drug product container which has been deemed questionable by an AVIM due to failing any of the inspection criteria. Ejects are sent to a 2 nd tier manual inspection for final judgment. Parenteral - Introducing a medication into the body through an injection Particle - Particles as defined in this research are materials within a drug product container which are not in solution with the drug product fluid. They may be proteinaceous or consist of rubber, fiber, hair, glass or any other foreign material. Pipeline - The drug product pipeline consists of all drug products which have not yet passed phase three clinical trials and are in commercial scale manufacture. Pipeline drug products carry significant uncertainty as to whether they will see commercial manufacture. Presentation - A presentation as defined in this research is the specific container, fill volume, material. Protein Aggregate - Protein aggregates are particulated clusters of individual proteins within a drug product. They commonly form into strands, globules, or crystals and can range in size from dimers to several hundred microns wide. Reversible aggregates can go back into solution under certain circumstances, which irreversible aggregates are permanent within a drug product. These aggregates form based on environmental and handling stresses, drug product formulation, and over time. Reject - A reject is a drug product container which has failed visual inspection and must be disposed of. Turbidity - A measure of the clarity of a drug product. Products should not be cloudy or opaque. 10 1 Thesis Introduction, Problem Statement and Internship Motivation The biotechnology industry is expected by consumers and regulatory agencies to maintain flawless quality in their products. To achieve this, there are numerous inspections, tests, and analyses done throughout the manufacturing process to ensure that each delivered product is free of defects and inspected at a rigor not seen in many industries. Appearance testing is critical quality attribute and one of the most vital inspection stages is the 100% visual inspection after a container has been filled, capped, and sealed. This inspection stage is the focus of this research and the internship behind it. The order of the fill stage in the production process is conceptually shown in Figure 1 and may have many Stock Keeping Unit (SKU) variations of the same basic protein in different fill levels, presentations, or protein concentrations. Example SKU Development of single protein product MANUFACTURE FORMULATE LABEL PACK Figure 1 -SKU proliferation of a single protein product, with fill stage highlighted and unlabeled, filled vial shown Currently, this visual inspection is done manually, by a certified inspector, or automatically, by an Automated Visual Inspection Machine (AVIM), and looks for defects in color, turbidity, particulate matter, or any cosmetic flaws. For Amgen, the most interesting, and challenging, inspection criteria is particulate matter due to its variable appearance, source, and size. Particulate matter occurs either as foreign matter, or proteinaceous matter in the case of biotechnology products. While current practices 11 meet regulatory requirements and ensure patient safety, there are operational drivers to improve, as well as future changes in the drug product pipeline that will challenge the existing inspection infrastructure. In the near term, as Amgen expands its manufacturing capacity to Ireland and future worldwide sites, there are further operational drivers for aligning technologies across sites and improving the throughput and accuracy of the inspection process. The high cost of manual inspection, and relatively low accuracy, offers both significant financial and operational incentive for automating inspection. As Amgen's pipeline proteins go from agonistic to antagonistic, a paradigm shift requiring protein concentrations two to three orders of magnitude higher than legacy products, the shifting properties of the drug products in the pipeline will challenge the technologies and processes in place. Higher concentration drug product can lead to higher protein aggregation and higher product viscosities. Furthermore, the introduction of plastic presentations requires different lighting and imaging techniques. Finally, regulatory momentum is pushing for higher levels of protein aggregate quantification than the current United States Pharmacopeial compendia limits require and a greater demonstration of a process control strategy. The desire is to translate the appearance test from a qualitative assay to a quantitative test as the basis of setting specifications. Regulatory bodies want to see more understanding and control of particle content in individual batches and it will be increasingly important to automate the task of sizing and counting individual particles.[1] Addressing these challenges will require new equipment, processes, and technologies. Specifying, acquiring, installing, and validating inspection equipment and processes is a multiyear process with many hurdles and potential setbacks. This long timeline requires that potential inspection issues are identified far enough in the future and foresight developed such that the necessary equipment or technology is acquired or developed well in advance of its necessity. For example, a typical timeline for acquisition of a new AVIM would be roughly three years: one year for specification and bidding, one for vendor construction and customization, and one for installation and validation. The purpose of this research is to 12 develop an understanding of the future requirements that will be placed on the inspection infrastructure based on the attributes of pipeline products, understand gaps that will appear, and develop strategies to proactively address the future needs of Amgen's final product inspection so that these gaps do not become operational inefficiencies. 1.1 Thesis Structure The document is organized as described below: Chapter 1 describes the problem statement and motivations for the thesis as well as an overview of the contents. Chapter 2 outlines Amgen's background and strategic goals for inspection as well as relevant literature. Chapter 3 details the Future Inspection Needs (FIN) model developed, its approach, operation, applications, and ramifications for the future of Amgen's final product inspection infrastructure. Chapter 4 contains engineering solutions developed to counter some of the major capability gaps determined through use of the FIN model and company data. Chapter 5 develops recommendations for investment based on a net present cost analysis of inability to inspect for attributes determined by the FIN model. Chapter 6 lists references used in the thesis. Chapter 7 contains Appendices for data presented in the thesis. 1.2 Project Goals and General Approach The goal of this research is a method of developing a proactive infrastructure investment strategy through future state understanding, problem identification, problem understanding, solution identification, and finally, solution engineering. Through this research, a comprehensive model was developed to predict future capability gaps based on current data. Based on gaps identified, the project turned to technology 13 development and research aimed at mitigating these gaps and improving the overall final drug product inspection process at Amgen's manufacturing sites. Combining gaps identified, and potential solutions conceived, the research provides recommendations for economically rational future decision making. 1.2.1 Future Inspection Needs (FIN) Model In order to build an investment strategy for long term inspection, it is critical to identify future problems. Many of the problems identified in this research were anticipated by management, but on in qualitative terms; there was no quantification of the potential issues in the future. Interviews with management and line personnel generated a reasonably consistent list of concerns: viscosity, protein aggregation, drug product presentation, and cosmetic defects. The first goal of the research was to develop a quantitative model to aggregate disparate data and translate these data such that they are relevant to inspection. This model combines commercial and pipeline drug product production forecasts from Amgen's Operations Strategy group, drug product attributes from Drug Product Team Leaders, and inspection equipment capabilities from manufacturers, operators, and test documentation. By comparing forecasts, drug product attributes, and capabilities, the FIN model is able to forecast what capability gaps will appear, when, and how many units will be affected. This model is discussed in Chapter 3 and is the basis for the engineering research conducted and the cost model developed. 1.2.2 Exploring technical solutions The challenges identified and quantified in the FIN model, directed the research and engineering approaches to mitigate these future challenges. Prior to the internship, Amgen had formed an internal development group, the Drug Product Engineering Advanced Science and Engineering (DPE ASE) group, to develop internal technologies to enhance Amgen's ability to inspect because no commercial solutions were available. Leveraging the knowledge of this group, the research pursued enhancements to existing equipment, segmented algorithms for vial inspection, inspection for high density particles, and use of acoustic streaming to agitate highly viscous products. Several of these engineering approaches will be discussed thoroughly in Chapter 4. 14 1.2.3 Cost of inability to inspect In order for Amgen's decision makers to rationally invest in inspection technologies, there must be an economic case for investment. While selecting individual technologies to invest in and equipment to purchase is beyond the scope of this research, by using the FIN modeling tool, an understanding of engineering approaches, and inspection cost data, we generate a net present economic cost of the inability to inspect products in each capability gap. Armed with this cost data, decision makers at Amgen will be able to pursue those investments which generate the highest returns on investment. These costs, and recommendations associated with them, are discussed in Chapter 6. 15 2 Amgen and Biotechnology Industry Background This chapter addresses Amgen as a biotechnology company and focuses most specifically on inspection at Amgen. Background information is given regarding inspection procedures, infrastructure, and history. Additionally, general background is given on inspection as a science and the rationale behind current inspection limits. 2.1 Biopharmaceutical Industry The biopharmaceutical industry took off in 1982 when the first therapeutic product was approved for human use. This protein, biosynthetic insulin, was made using recombinant DNA technology,[2] and has spawned the entire biopharmaceutical industry which today generates $91 B in worldwide revenue. [3] The biopharmaceutical industry is a subset of both the pharmaceutical industry as well as the general biotechnology industry, which includes companies like Monsanto dedicated to agricultural applications of biotechnology. Since the 1980s, the industry evolved considerably from its startup period and over the last few years entered into the "mature" phase of industry development. While the industry saw a decline during the recession in 2007 and was forced to reduce staff by 13%[4], the biopharmaceutical industry has rebounded since then largely due to stable patient demand and pricing power. There are currently over 150 products on the market with 633 in development and sales are projected to grow to $150 by 2015 leading the overall pharmaceutical market. While product growth is expected to continue, there are further signals that the industry's major players are transitioning from growth to maturity as R&D spending is decreased as a proportion of revenue and acquisition activity increases.[3] Like the pharmaceutical industry, the biopharmaceutical industry has unique risks due to the nature of the product it produces. Product defects, recalls, and shortages can cost companies into the billions of dollars prompting an intense focus on quality and availability of product. Most biopharmaceuticals claim to 16 carry over one year of product inventory to mitigate the risk of a shortage, contamination, or supply disruption. Biopharmaceuticals are differentiated from conventional pharmaceuticals in that they are considered "large molecules," with molecular weights hundreds of times larger and significantly more complex interactions than the "small molecules" of the pharmaceutical world. The large majority of these "large molecule" products are derived from live cell cultures and grown, rather than obtained through direct chemical synthesis.[2] The production process for producing proteins from cell cultures, while robust, offers significant challenges in terms of repeatability, sterility, and scalability that are often not present in "small molecule" manufacturing. The process is highly reliant on exact conditions, geometries, and manufacturing control parameters to be repeatable and changes in these conditions can alter the final protein drug product. Due to the intertwined nature of product and process, it is very challenging to transfer technologies between sites or companies and exactly replicate the product. The cost to develop a new therapeutic biologic has been estimated to be approximately $1.3 billion, potentially rising to $2 billion in 2015 and $2.5 billion in 2020. This process takes eight years or more to bring a drug to market.[3] A trend threatening to disrupt the market is the introduction of biosimilars, the "large molecule" equivalent of a generic drug. These are deemed biosimilars because it is effectively impossible to perfectly replicate the original drug. While currently no pathway exists in the U.S. for the approval of a biosimilar drug, and the barriers to entry are high, these products will play a significant role in the future of biopharmaceuticals as governments and businesses work to lower their healthcare costs. 2.2 Amgen Inc. Amgen Inc. is a leading biopharmaceutical company with ten principal products which treat conditions such as rheumatoid arthritis, psoriasis, cancer, bone disease and others.[5] Amgen was incorporated in 17 1980 in Thousand Oaks, California where its current headquarters and primary research facilities are located. Since 1980 it has grown to 17,000 employees worldwide with $15.6 billion in revenue. Amgen's mission is to serve patients, and does this through pursuing treatments which will dramatically improve patient's lives. To this end, their R&D budget, $2.3 billion in 2011, is largely focused on protein therapeutics, such as monoclonal antibodies, and, to a lesser extent small molecule development. This R&D budget constitutes roughly 20% of revenue, slightly lower than most major players in the market. [3]. 2.3 Inspection at Amgen Amgen, like all biopharmaceutical companies, performs 100% visual inspection on its products. These products are predominantly liquid filled vials and syringes but also include lyophilized product in vials as well as tablets in the case of small molecules. These products are inspected prior to labeling to ensure a defect free product that has no contaminants, discolorations, and no potential for a breach of sterility. The U.S. Pharmacopeial Convention Chapter <1> declares that parenteral products should be inspected such that they are "essentially free from visible particulates."[6] Particle detection is the primary focus of this thesis research as the technology for cosmetic inspection is already well advanced and cosmetic issues are not expected to be a significant future challenge. Other inspection criteria such as color and turbidity are issues that will occur across an entire batch, not probabilistically in individual vials, as particle do, and are currently easily detected using sampling methods. In addition protein aggregate particles can change in size and number over time, in contrast to foreign matter particles which remain constant. Particle detection is part of the 100% inspection process and Amgen uses a mix of manual and automated inspection systems to identify and reject products with unacceptable particulate matter. 2.3.1 Manual inspection Manual inspection at Amgen has been in place since the first products were manufactured in the late 1980's. Manual inspectors sit at a calibrated inspection booth with either top or top and bottom 18 fluorescent lighting and manually inspect products for particulate matter, turbidity, color, and appearance in front of both black and white backgrounds. Inspectors gently swirl the contents to agitate any particles into suspension and scan the vial or syringe for signs of these particles. As with many visual inspection processes, it is certain that Figure 2 - Phoenix Imaging manual defects will be not be recognized with 100% probability and inspection booth there is a chance that even a defective product which is inspected will not be rejected. Manual inspection is a variable process and each inspection event is recognized as a probabilistic event with less that 100% certainty of detecting a particle in the container.[7] Despite its unrepeatable and probabilistic nature, manual inspection is currently the baseline inspection system at Amgen and nearly all biotechnology companies.[8] To ensure that the 100% inspection was effective, a number of the 100% inspected and passed containers are sampled and subjected to a repeat inspection by the Quality Control group with the expectation that the findings do not result in false negatives beyond preset criteria based on statistics also known the specified Acceptable Quality Limits (AQL). -10 40 0 so 0 e Meichorm 10e Pare ISO 200 288 Sim. (um) Figure 3 - Manual inspection performance by particle size and study [8] Figure 3 shows the performance of human inspectors across a range of particle sizes over several studies. As can be seen for particle sizes which have several data points, there is a large range of variability, even in scientifically controlled studies, with a single 100um particle having anywhere from 20% to 75% 19 likelihood of being detected. Currently Amgen uses 125um as the threshold between what constitutes a visible versus subvisible particle. The visible threshold is set at the point where there is statistically a -70% chance that the single particle will be seen by a certified human inspector. Manual inspectors must pass annual eye exams and are required to have 20/20 near focus visual acuity. Inspectors are trained against a large "defect set" which contains an assortment of defects including particles, fibers, glass, rubber, chipped containers, defective stoppers or crimps, as well as a majority of defect free vials. In clinical manufacturing, due to low volumes, inspectors are generally cross-trained from other positions and perform inspection when required by the manufacturing schedule. Commercial manufacturing uses full time inspectors. A typical full time inspector works an eight hour shift, taking several "eye rest" breaks throughout the day and will be actually be inspecting product for -six hours. Vials are processed one by one, whereas syringes are held in a "claw" and inspected several at a time. 2.3.2 Commercial Automated Visual Inspection Automated inspection has been replacing manual inspection since the 1990's and currently Amgen has AVIM installed in both clinical and commercial sites inspecting both vials and syringes. These machines are generally much higher throughput (300-600 units/minute) and have visible particle detection rates around 90%.[9] Currently installed AVIM's use a Static Division (SD) type sensor to detect particles within the drug product. As shown in Figure 4, a light source is projected with a collinear beam through the container under test into a lens which focuses the light onto a vertical, one bit wide, strip of light sensors. If the sensor strip detects a change in light transmission during the period of observation greater than a specified threshold, it reports this as a particle detected. 20 v(light) WNW+ aweSignaS A' change of light intensit OV (dark) A Foreign Particle re~cIon Particle signal ---- -Sensitivity level 1btof Sensor 0V "sensor Figure 4 - Static Division (SD) sensor operation[1O] This inspection happens at the end of what we will call the "spin-brake-inspect" cycle shown in Figure 5. The container is first spun between 1,500-2,000rpm in order to generate a vortex inside driving the fluid to the outside of the container, and then the rotation is abruptly braked. After the brake is applied the vortex in the center of the container collapses upwards generating upward flow pulling particles into the center of the still swirling fluid. The desired outcome is to have sufficient liquid flow to move high density particles in the field of view. The container is then moved in front of the SD sensor which detects for the presence of moving particles. If a particle is detected the container is ejected into a separate bin to go to a 2nd tier manual inspection which will provide the final verdict on the container. Liquid Level Particle. Spin Slop inspect -Inspecl Figure 5 - Spin-brake-inspect cycle[10] The time between brake application and inspection, the spin speed, and sensitivity settings, are three critical parameters in setting up an AVIM for inspection and must be set through empirical testing for each container, product, and fill level. A higher spin speed is required for more viscous products and will increase the size of the vortex in the center but may also generate bubbles which can cause false rejects. 21 Similarly, a short time before braking may not allow the meniscus to fully collapse away from the inspection window causing false rejection, while a long time before braking may allow particles to cease motion before inspection or fall into the base of the container, which is excluded from inspection. [11] A challenge for properly evaluating AVIM is that 69% [8] of companies use manual inspection as the baseline standard to evaluate the efficiency of a machine, which is expected to meet and exceed human capability without excessive false rejects. Thus, for a given challenge set, only vials which are rejected more than 70.7% of the time by manual inspection should be rejected by the AVIM. In testing though, an AVIM will reject particles with higher efficiency and repeatability than a human inspector,[7], [9] leading to a catch-22 for evaluating AVIMs. 2.4 Particulate matter The primary focus of the inspection challenges and technologies in this research are focused on particles in final drug product containers. This section provides a brief overview of the types of particles seen and issues associated. 2.4.1 Extrinsic particles Extrinsic particles are defined as particles which come from materials which do not make direct contact by design during the production process. These particles consist of hairs or skin flakes from human operators, paint chips, clothing fibers, cellulose fibers, insect parts, or any other matter that comes from outside the theoretically closed system of drug product manufacture. If discovered, these particles are especially worrisome because they indicate loss of control of the manufacturing process whereby the drug product is exposed to elements from which it should be isolated. Upon discovery, there is an intensive investigation process to identify the source of the particle followed by mitigation efforts. 2.4.2 Intrinsic particles Intrinsic particles come from the closed system of manufacturing the product, generally surfaces used to contain or convey the product during production. These include rubber from seals and stoppers, stainless 22 steel from piping, glass from the vial, or protein aggregates from the drug itself. These particles are more common than extrinsic particles, though still rare, and are investigated thoroughly if outside compendial limits. Protein aggregates in particular are more complex and warrant their own discussion. 2.4.2.1 Protein aggregates The United States Pharmacopeial <788> limits subvisible particulate matter in a small volume parenteral drug product (under 1OOmL) to 6000 particles between 1Oum and 148um with no more than 600 particles between 25um and 148um. Any particles above 148um are cause for rejection or at least an investigation for the entire lot.[12]. This test is destructive and applied to a limited number of containers. The principle of the test is light obscuration which will capture both foreign as well as protein aggregates. This thesis is focused on visible matter using non-destructive tests. Monomer Dimer Unfolded proteins Aggregated protein particle Figure 6 - Basic states of proteins and advancement of protein monomer to aggregated protein particle Protein aggregation is the result of self-association of the individual proteins (monomers) and is defined an 'inherent' property of the drug product. Protein aggregation occurs when monomers aggregate together into globules, crystals, strands, or any of various possible configurations which have a wide range of appearances and characteristics. Protein misfolding is the main path leading to aggregation and aggregation is more likely to occur in high concentration drug products as the frequency of interaction between individual protein monomers is much higher. Aggregation can happen during the manufacturing process due to any number of mechanical, chemical, thermal, or handling stresses but is also time dependent and may occur after days or weeks in storage. These aggregates can be both reversible and 23 irreversible, increasing the challenges of inspecting for them. A drug product may be protein aggregate free at time zero after manufacture but may develop particles after sitting in cold storage for several days only to have them disappear under the agitation of inspection. The hazards of protein aggregates are currently not well understood but are thought to be capillary occlusion, immunogenicity, or, at least in theory, reduced effectiveness due to fewer effective proteins. The most hazardous of these is immunogenicity. As proteins aggregate, they can form repeating arrays, which the immune system may detect and respond by creating neutralizing antibodies, a reaction called immunogenicity. These antibodies can, at a minimum neutralize the active injected drug, and in a worst case scenario, generate an autoimmune response attacking the body's own native proteins identical to the injected drug.[13], [14] Currently, protein aggregates are treated the same as other particles in the compendia limits but this is changing. These compendia limits were designed to control foreign matter many years before the birth of the biotech industry. The significant uncertainties regarding the hazards posed by these aggregates is driving regulatory pressure to better understand and control particles.[1] Current technologies fall short of being able to reliably detect and characterize subvisible protein aggregates, which may be the most likely to cause immunogenic responses.[15]. The amount of protein that can cause immunogenicity can be as low as a microgram or in some cases even less. A typical protein dose ranges from 1,000 to 100,000 micrograms, illustrating the need for a sensitive assay. This has prompted Amgen to begin proactively exploring options for obtaining better data on the particles in its products and develop technology to support this. 2.5 Drug Product Engineering Advanced Science and Engineering Group The author worked within the Advanced Science and Engineering (ASE) group located within the Drug Product Engineering (DPE) group within Amgen Operations, for the duration of his internship. The group was formed by management in 2006 to evaluate opportunities for innovation in drug product 24 manufacturing and to improve operational efficiency and compliance. An example is the initiative to improve machine vision reflecting the importance of parenteral product integrity by developing a competitive advantage, and responding to an evolving regulatory environment which is gradually becoming more demanding specifically around the need for quantitative methods. The DPE ASE group is working to address these challenges mainly through the application of machine vision and image processing. For the past few years, this team has been focusing on several inspection efforts: Development of laboratory scale particle inspection devices: The group developed two generations of semi-automated and automated laboratory scale inspection devices. Dubbed ParticleVision 1 and 2 (PV 1 and PV2), these stations consist of high resolution videography equipment, strategically arranged LED lights, and sophisticated image recognition algorithms. The stations inspects individual vials or syringes through a similar process to what is done on clinical and commercial scale inspection systems, a spinbrake-inspect sequence as described earlier (though as much lower spin speeds), but in contrast to the static division based commercial system, the PV systems records a video of the subject under test as the drug product liquid medium swirls inside. Image recognition algorithms then process the video stills identifying particles, tracking them, and through their trajectories and appearance, attempt to size and classify each detected particle. This system has been in use for several years, mainly for drug product development and formulation testing and serves as a test bed for many of the approaches detailed in this research. Building on the knowledge gained through the development of PV2, the group also seeks to identify other opportunities for machine vision and optics throughout the drug formulation, manufacturing, and inspection functions. Providing support for internal inspection or imaging related projects: Acting as an internal consultant and "skunkworks" style laboratory for internal inspection or machine vision related projects, the group provides advice, technical support, and small scale demonstrations on an ad-hoc basis. 25 Working with third party vendors to create long term inspection projects: On projects which are beyond the scope of the DPE ASE group's capability, they partner with third party vendors and integrators to develop longer term projects which are focused on high throughput inspection systems. The PV2 system is most relevant to the research performed in this internship as it provides an excellent test bed and comparison tool. While the unit currently is designed for low throughput inspection (~100 units/hour) the technology it contains is easily scaled and it has the potential to be embedded in a larger manufacturing inspection machine as a modular addition. 2.6 2.6.1 Current capabilities across international facilities Alignment between clinical and various manufacturing sites To properly project future needs for the final product inspection infrastructure, a comprehensive understanding of existing capabilities is necessary. Amgen currently operates two sites where final drug product is visually inspected, Amgen Thousand Oaks (ATO) and Amgen Manufacturing Limited (AML) in Puerto Rico. ATO conducts clinical drug product production while AML's focus is on commercial product manufacture. Amgen recently acquired a site outside of Dublin, Ireland (ADL) and is in the process of building it up as a second manufacturing and packaging facility. Currently, each site has different inspection equipment. For instance, ATO can only inspect 3cc vials and syringes are done manually, whereas 3cc vials and syringes are both processed automatically at the commercial site. This requires increased validation and qualification during the technology transfer from clinical manufacturing to commercial manufacturing. Each site uses various mixes of human and machine inspection stations based on their product mix and volume. 2.6.2 Implications of redundancy strategy on inspection The acquisition of facilities from third parties increases the complexity of the inspection challenge. For instance, the new facility at ADL is equipped with a different AVIM brand than the AML and ATO facilities and they use different operating principles of detection. While the machines are roughly equally 26 capable, there are still operational and efficiency differences which must be understood. In the biopharmaceutical world this requires extensive testing and qualification to validate a new piece of equipment. This qualification is still underway at ADL for new Amgen products and exact performance numbers for the AVIM there are not yet available. Part of this research will discuss reasons for equipment standardization across facilities. 2.7 2.7.1 Amgen's inspection strategy as competitive advantage Regulatory uncertainty As discussed above in section 2.4.2.1 the regulatory pressures are increasing with regard to understanding and quantifying particulates in parenteral drug product. The rules regarding foreign matter particulate suspended in these parenteral products are extremely stringent. Currently, regulatory thresholds are largely driven by the accepted capabilities of human inspectors instead of a scientific evaluation of the risks of particles.[6], [7] Pressure is growing from regulatory authorities for the industry to explore, develop and adopt novel automated solutions for quality inspection.[13], [14] While requirements regarding particles have been in place for many years, the regulatory bodies are beginning to look more closely at particles as an indication of manufacturing process control as well as their potentially adverse health effects.[1] A more scientific approach to particulate inspection is setting a threshold that is more coherent with what is safe than just what is detectable. This does not imply inherent danger in the current standards, but the rationale is not sufficiently scientific to base decisions on as the pipeline evolves. 2.7.2 Inspection as competitive advantage As regulatory bodies begin to put pressure on companies to demonstrate better process control through quantification of the particle loads, Amgen is striving to be on the leading edge of this transition. Amgen has a strong incentive to understand its future pipeline, as well as evaluating technologies, and through this understanding take the lead on pressing for technological advancement and standardization of inspection processes. By being the leader in this space, Amgen will be able to work together with 27 regulatory agencies to raise the bar and set the standards by which all biopharmaceutical companies will be measured. 2.7.3 Future state goal The purpose of this effort is for Amgen to be in charge of its own, as well as the industry's, destiny. Through developing a strong understanding of its own needs and developing technologies that will meet them, Amgen will take a strong position as the leader in non-destructive inspection technology. The ParticleVision line of AVIMs has the potential to address the concerns of regulators by quantifying and qualifying particles in inspected products. These AVIM concepts still have shortcomings though and must be engineered to meet the needs of the pipeline and be paired with higher throughput technologies in order to be viable for commercial scale inspection. An ideal future state for drug product inspection would be an inspection system which is able to identify each particle within a container, size and count it, and relying on historical particle data in clinical lots, make a patient safety based assessment as to whether it warrants rejection. This implies that improved specifications are needed that reflect quantitative data as opposed to the current qualitative expectations. To get to this point, we need to determine the future state of the pipeline, as well as technologies to get there. 28 3 Future Inspection Needs and Infrastructure Capabilities Model This chapter details the development of the Future Inspection Needs (FIN) model. This model is designed to be the foundation of any technology investment strategy and future decisions made regarding inspection systems. The FIN model aggregates data from disparate company sources and distills the data through attributefocusedforecasting into requirements which are relevant to investment infrastructure decision makers. This dynamically updated model which gives investment decision-makers a quantitative assessment of the next ten years allows Amgen to make investments based on rational business drivers. 3.1 3.1.1 Approach Motivation As described earlier in the document, the primary goal of this research is to develop a long term strategy for, and understanding of, the final product inspection infrastructure. The FIN model is designed to give decision makers a tool with which to better understand both current needs and future ones when determining critical company investments. Currently, this process is reactionary when, due to the long cycle times in any investment, decisions need to be made proactively with an eye to future needs. Many of the challenges identified in the model were not surprising to managers, but the quantity and timing were new pieces of information that were of interest. An additional function of the prediction model is to determine costs of future manual inspection, in the case that adequate automated inspection systems are not in place. Manual inspection is currently the de facto fallback for any product which is too challenging for automated inspection and the cost of this is not always accounted for. The model shows, in each year and by category, the cost of using manual inspection due to lack of validated automated inspection. While it faces volatility due to competition, clinical product attrition and regulation, Amgen still has significant visibility into its future products. With over eight years to bring a product a market, Amgen has the ability to predict very well the characteristics of its future commercially manufactured products. While there is still significant uncertainty whether these products will pass each trial phase, many 29 products in development have similar characteristics and as an overall portfolio, the challenges that will be faced can be predicted with some certainty. The model presented here takes a different approach to capability forecasting and eschews a standard capacity planning approach for a more relevant capabilityplanning focus. Current AVIM are not limited by their speed (with respect to Amgen's needs), but by their capability to inspect products with certain attributes. The model generates an attribute focused forecast which is able to translate production forecasts, which are effectively irrelevant to inspection, into capability requirements for future inspection infrastructure. This research focused on determining the attributes of current and future products, which will be outside of the capabilities of existing infrastructure, and then distilling production data down through these attributes. Once the capability gaps are known, and a cost-analysis applied, decision makers can make effective investment decisions which are economical in the short, as well as long term. 3.1.2 Model Parameters The parameters for the model were selected after reviewing existing hardware, interviewing inspection managers, supervisors, drug product team leads, inspectors, and other relevant stakeholders. Those drug product qualities which are expected to drive the main technical challenges for future drug product inspection are viscosity, container presentation, and particle load. Cost and productivity data was determined from inspection managers, while equipment capacities and capabilities come from manufacturer information, inspection managers, and engineers. Production forecasts were obtained from the Operations Strategic Planning (OSP) group at ATO and updated regularly throughout the internship. 3.1.2.1 Viscosity Viscosity is one of the prime drivers of future inspection challenges, especially because currently the solution to this is not entirely known and will likely require research, development, and engineering effort. The rise in viscosity is due to the significant increase in protein concentration in drug products. Amgen's historical bread and butter proteins such as Aranesp, Epogen, and Neupogen have, at their 30 highest, protein concentrations of 0.50 mg/mL, 0.34 mg/mL and 0.96 mg/mL, respectively. Several pipeline proteins have projected concentrations of 140-150mg/mL and beyond. 353025j' 20- 1510- 50I 0 50 100 150 Concentration (mgimL) Figure 7 - Plot of viscosity vs. concentration for various drug products Current equipment becomes unacceptably ineffective for particle detection above viscosities of 3-4cp as shown in Figure 8. Effect of viscosity on detection rate for Polysorbate formulation 120 100 0 ~20 0 0 -- 2 4 6 Viscosity (cp) 0.3 ml, Spin 1 -e-0.3 ml, Spin 2 - -- 8 10 ml, Spin 1-x-1 ml, Spin 2 Figure 8 - AVIM particle detection rate vs. viscosity at various drug product mimic viscosities and AVIM settings The current reasoning for this phenomenon hinges on the relative motion between particle and container. When an AVIM spins the container to suspend particles for inspection, as detailed in 2.3.2, it requires that the particle continue moving for some time after the container has stopped spinning. In high viscosity 31 solutions, the particles within the drug product are either never lifted from the base of the container into the inspection window or stop too quickly after the spin brake is applied. Data on product viscosity was gleaned from company documents and Drug Product Team Leaders. 3.1.2.2 Presentation Amgen's is considering plastic vials but these containers present several inspection challenges: * Thicker walls and lower light transmission: This requires higher light levels than current processes are developed for. * Variable thickness walls: Currently proposed 3cc CZ plastic vials are blow molded into forms. This leads to variable lensing effects and distorted projections of particles in the product. * New dimensions and handling requirements: The new vials are of different dimension than existing vials and will require specific change parts and handlers to accommodate. 3.1.2.3 Particle Load Though discussed more thoroughly in 2.4, certain particles represent a distinct challenge for existing AVIMs. Products which have a tendency to form protein aggregates are primarily inspected manually for economic reasons. Installed AVIMs are not designed to count and look at each particle's size and composition, before classifying it as an Figure 9 - Protein aggregates in drug product eject or accept. For most products which do not form protein aggregates this approach is acceptable because there is no expectation of particulate matter in the drug product. If a product has a tendency to form numerous small particles, which are not necessarily harmful and fall within the compendial guidelines for size and count, they will still have a high rejection rate when inspected by current AVIMs because of the inability to size and count particles. Running these products through manual inspection, which is more able to size and count particles, is costly, inefficient, and lacking robustness. 32 3.1.2.4 Inspector Productivity and Cost Inspector productivity and cost data were obtained from inspection managers in AML. A breakdown of a fully burdened inspector's cost and productivity is given in Table 1 - Inspector labor data Inspector burdened cost Inspector Break/Downtime Shift length (hours) Shifts per day Inspector target utilization Working Days Table 1 and Table 2. Table 2 - Inspection rates for various container types $ 100,000.00 25% 8 2 70%, 241.00 3cc Glass Vial 5cc Glass Vial 10cc Glass Vial 20cc Glass Vial 50cc Glass Vial 1cc Glass Syringe 1.5cc Glass Syringe 1cc CZ Plastic Syringe 4.1cc CZ Plastic Cartridge 6 1 1 1 1 11 11 11 6 364,392 60,732 60,732 60,732 60,732 668,052 668,052 668,052 364,392 Different presentations require varying lengths of time to inspect based on container volume and required inspection method. Inspectors' hours, breaks, and shifts are closely regulated by Amgen to ensure uniform performance. 3.1.3 Data gathering Automated Data Gathering Data Source R a pid Res po nse N in0ewae004nn4att Figure 10 - Flow of FIN model data inputs The core of the FIN model runs off of the forecast sales data provided by the Operations Strategic Planning (OSP) group at Amgen. This data forecasts over the next 10 years the projected sales of each commercial product, as well as expected pipeline products. The model built has data from all commercial 33 products and five pipeline products. 27 pipeline products are excluded due to time horizon outside of model scope, product type, or low probability of eventual manufacture. In order to ensure the model continues to be useful in the future, the commercial forecast data from OSP can be automatically generated and loaded into the model requiring very little effort. This data collection was automated because of the standard output format and frequency of updates. Pipeline product forecasts are generally done manually and updated only every quarter to six months. 3.2 Model operation In order to improve future accessibility the model was built in Microsoft Excel 2007 instead of a standard database program. While constructing the model in Excel reduces the flexibility and speed of the model, it will increase the likelihood that future owners will be able to operate, maintain, and expand it. The model revolves around a central worksheet which aggregates the product forecast data from OSP for each SKU produced as well as the relevant attributes for that SKU. Individual worksheets distill this database into figures relevant for inspection. By looking across all products at a SKU level, the model discards information that does not affect inspection and provides a summary view of each inspection challenge by year. The model generates worksheets by protein, presentation, viscosity, particles, "inspectability," and inspection costs. At the root of each mathematical formula in each cell is a summation across all products in each year where certain criteria are met. An example is given below for predicting the number of units 'U' which will not be inspectable with current processes in yearj due to high viscosity: U = Pij *yi V i where Pj,1 is the supply forecast for SKU i in yearj 1, r; > 4cp 1 < 4cp y-0, in single attributecases this works out to be Uj = P1 Ty Equation 1- Basic formulation of matrix multiplication for data distillation 34 Each model module is based in on this general equation though more segments are defined, multipliers added, or parameters Load New ComnmW Data Sdect oDateod~akj iftftloda used. A full list of formulas used is given in Appendix 7.2. In order to facilitate maintaining the model with the constantly updated commercial product forecast data, a Visual Basic script is built in. This script is run from the model control panel and parses a standard output from the OSP group's Enterprise Figure 11 - GUI for product forecast data loader Resource Management (ERP) system and imports the data into the model in only a few seconds. This commercial data is automatically generated each month and emailed to the FIN model owner ready for quick import. See Appendix 7.2 for visual basic code on loading data and generating outputs. 3.3 Model Outputs The FIN model generates several outputs valuable to decision makers and planners. A snapshot of the model as of December 2011 is shown below, detailing the future states of Amgen's drug product line. 3.3.1 Inspection needs Figure 12 shows the growing complexity of the presentation mix, mostly associated with the introduction of plastic cartridges and syringes, but also higher volume intravenously delivered oncological drugs. This figure shows the percentage of the product portfolio filled by the "alternative" presentations to the 3cc vials and Icc glass syringes which currently make up the vast majority of filled drug product at Amgen. Figure 12 represents not just a growing diversity and scale of new presentations, but also the challenge associated with maintaining manufacturing capabilities across a portfolio of increasing complexity. 35 Presentation breakdown by year (3cc vial and 1cc syringe removed) * 5cc Glass Vial m 10cc Glass Vial N 20cc Glass Vial * 50cc Glass Vial 0 1cc CZ Plastic Syringe N 4.1cc CZ Plastic Cartridge 12% 10% 8% 6% 4%2%- 0% - ----- 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 Figure 12 - Percent of filled units per year which are not "traditional" 3cc glass vials or 1cc glass syringes In 2011, over 99% of product containers filled were either 3cc glass vials or Icc glass syringes, whereas by 2020 this will be below 90%. As these new presentations move to manufacture, they will require new validated inspection techniques and processes. In 2016 the emergence of plastic presentations begins to dominate the "alternative" primary container scene. This is especially challenging because of Amgen's lack of history with plastic inspection. Associated issues and solutions will be discussed in section 3.3.4. Figure 13 details the shifting viscosities of products from low, easy to inspect, viscosity to higher viscosities above 4cp. In 2011, high viscosity products account for only 20% of production by units filled, whereas in 2020 they will account for more than half of all units filled. Many of these products have viscosities beyond the validated range of currently installed AVIM, and if introduced without adequate inspection capabilities, will incur significant manual inspection costs. 36 Viscosity range by year as a % of portfolio 0 Low (0 - 2cP) E Med (2 - 4cP) L High (4 - 10cP) 100% 90% 80% 70% - 60% 50% 40% 30% 20% 10% 0% 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 Figure 13 - Portfolio viscosity ranges by year 3.3.2 Infrastructure capabilities Beyond understanding the product portfolio attributes which relate to inspection, it is critical to understand which attributes pose a threat to the future operational efficiency of Amgen's manufacturing centers. The data presented in Figure 14 compares the attributes above and the installed capability. Main drivers of product uninspectable by AVIM E Proteinaceous 0 Presentation U Viscosity 20 18 16 I- 0 14 12 10 8 6 4 2 0 2011 2012 2013 2014 2015 2016 Figure 14 - Filled units uninspectable by currently validated processes 37 2017 2018 2019 2020 This figure is one of the most critical outputs of the FIN model because it cross references the attributes of future products with the capabilities of currently installed infrastructure and yields, by year and category, the number of units which do not have validated automated inspection processes. The three primary attributes, viscosity, presentation, and proteinaceous, are currently limited as follows: Viscosity: As mentioned earlier, in high viscosity products it is more difficult to agitate particles into suspension and relative motion with the container in order to be imaged. This problem is most acute in syringes due to their narrow diameter. The maximum inspectable threshold for syringes is set at 5cp, while no expected vial products are expected to exceed the viscosities currently inspected successfully through automatic inspection and are therefore not included in this analysis. Presentation: Current AVIM equipment is configured to handle only 3cc glass vials or Icc glass syringes. Other vial sizes include 5cc, 10cc, 20cc and 50cc and currently make up a very small portion of Amgen's product and are inspected by hand. However by 2020 this will be over two million units, and because they are larger, they are more time intensive to inspect manually and allowing the eyes to fully scan larger vials slows the rate to one vial per minute. In addition to the larger glass vial sizes, as of December 2011, there are no validated processes at Amgen for automatically inspecting any plastic presentation. There are guidelines for manual inspection but these have not yet been adapted, tested, and validated on AVIMs. Plastic presentations have been successfully inspected in manufacturing environments for years in Japan and the equipment exists, but at the current time, there is no validated capability to inspect them at Amgen. Proteinaceous: Some products with a propensity to form protein aggregate particles in the drug product form must be inspected manually due to high reject rates associated with AVIM. USP guidelines allow for a certain number of particles within certain size ranges. However, currently installed AVIM used for 100% are unable to properly quantify the size and quantity of particles and will reject many product containers that would be deemed acceptable by USP guidelines. These products are generally inspected 38 manually to minimize the false reject rate. By current projections these products do not account for a significant portion of uninspectable product in the future, but this is based on limited study data. Filled units with observed particulation M Particle Free a Study Ongoing M Particles Observed 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 Figure 15 - Filled units with observed protein aggregation As shown in Figure 15, the aggregation tendencies of several pipeline products are still under long term study and will not be known for several years. While Amgen has focused on minimizing aggregation through robust formulation, these products still present uncertainties which, beginning in 2016 could require significant levels of manual inspection without improved AVI capabilities. The chart shown in Figure 15 and the FIN model in general, does not currently include the potential of these products with ongoing studies to impact the inspection capability gap, but if particulation is observed, the gap will grow. 3.3.3 Cost of inability to inspect Figure 14 shows one of the most critical outputs of the model, the number of units each year which cannot be inspected by currently validated automated visual inspection machines. In 2011 and 2012 these numbers are low enough that manual inspection has been able to shoulder the burden without significant cost to Amgen. Beginning in 2015 and accelerating in 2016, the number of units which, under current capabilities, will need to be inspected manually grows between 50-100% each year. Based on the rates 39 assumed for manual inspection, the cost of inspecting these drug products with human inspectors is shown in Figure 16. Inspectors required for non-AVI validated products " 3cc Glass Vial N 5cc Glass Vial U " 20cc Glass Vial 0 50cc Glass Vial N 1cc Glass Syringe " 1cc CZ Plastic Syringe E 4.1cc CZ Plastic Cartridge 10cc Glass Vial 60 50 40 30 20 10 0 Year FTE Cost $ 2013 2011 2012 2011 2 246,565 $ 2012 4 416,953 $ 2014 2015 2016 2017 2018 2019 2020 2019 2020 2018 2017 2016 2015 2013 2014 50 40 47 17 25 11 12 9 895,958 $ 1,079,532 $ 1,186,471 $ 1,743,044 $ 2,530,986 $ 3,950,964 $ 4,686,049 $ 4,960,666 Figure 16 - FTE's and associated cost required for manual inspection By 2020 the number of inspectors balloons to 50, which would require more than doubling the current inspector base at AML in Puerto Rico. Even by adding a third shift and extra inspection stations, the physical space is not adequate at that facility to house these new inspectors. This adds extra gravity to the necessity of automating the inspection process for the new presentations and viscosities that will be reaching the manufacturing floor within three years. 3.3.4 Gap mitigation analysis and recommendations The primary focus of this research is to consolidate data and develop a forward looking model to identify capability gaps long before they become operational efficiencies, but it is important to also discuss potential solutions. Many of the solutions can be broken down into make-or-buy decisions. Applying this basic principle to the problems at hand, we can come up with four solution paths of varying 40 complexity. A chart showing the projected relative scale of each solution type between 2016 and 2020 is shown in Figure 17. Each category of solution will be addressed below but it is important to note the growth of the higher complexity challenges, developing technology and purchasing equipment, between 2016 and 2020. There is low hanging fruit currently but in the long term a proactive plan is needed. Potential solutions to AVIM uninspectable products 100% - 80% 60% 40% 20% 0% 2016 MPurchase change parts 2020 Validate process 0 Purchase equipment 0 Develop technology Figure 17 - Potential solutions to implementing AVI for future capability gaps Each solution category identified, in order of generally increasing complexity, are discussed below: 1. Purchase change parts: This "buy solution" applies to the 5cc, 10cc and 20cc vial sizes which will be sold by Amgen primarily for oncological drug products. Currently installed AVI infrastructure at the manufacturing sites have the technical capability to inspect these vial sizes as the products are not overly viscous or proteinaceous but they lack the vial handling hardware required. By purchasing the correct change parts to enable the AVIMs to handle larger vial sizes, nearly 1.5M units in 2016 can be inspected automatically. 2. Validate processes: This "make solution" is fairly straightforward in execution and applies entirely to the new 1cc CZ plastic syringes which will be introduced in 2016. Currently, there is not a sufficient level of data on plastic syringe inspection to determine the optimal AVIM settings 41 for these products. By running a detailed study to optimize these settings for plastic syringes such as that performed for vials, these products can be inspected on current machines with minimal investment. 3. Purchase equipment: Two presentation based challenges, 50cc vials and 4.1cc CZ plastic cartridges, require more aggressive purchasing approaches but the technology exists and is already in commercial use. It is important to begin working with vendors today though to ensure that a customized solution can be designed, installed, and validated for both the 50cc vial and the 4.1cc CZ plastic cartridge. In 2020 it will take six additional inspectors for the 50cc glass vials and 18 for the 4.1cc CZ plastic cartridges; at a cost of $100,000 each per year, there is a strong economic incentive for proactively investing in this equipment. 4. Develop technology: The most challenging aspect is to predict the path to acquiring new technology which represents the highest risk for Amgen in terms of inspection cost. A failure to develop or acquire adequate inspection for high viscosity will affect 8.4M units in 2020, entirely Icc CZ plastic or glass syringes. Fortunately, most of the high viscosity products in syringes won't reach commercial production until after 2016, giving Amgen some time to focus on acquiring the capability. Engineering solutions to this problem are posed in Chapter 4. A more detailed discussion of the timeline for solution investment and implementation is discussed later in this document but there are a few general points that are important to note now: Solution complexity is increasing. By 2016 there will be 3.5M units which will need to be inspected manually, but 3. 1M of these have off-the-shelf AVI solutions available or just need validated processes. While the purchase of any new piece of equipment or conducting a study is time consuming and costly, these are relatively low risk approaches and unlikely to fail to produce results. By 2020 though, the number of products which need entirely new technologies to inspection expands to over 8M units. This cost will only increase each year as more high viscosity syringe delivered products are developed. Research and development must be done proactively in order to avoid this cost. Some technical solutions 42 are explored and proposed as part of this research but all current paths contain high levels of risk and several should be undertaken concurrently. Whether Amgen chooses to pursue this in-house or in partnership with 3.3.5 3 rd party vendors is discussed later also. Model expansion and scalability The FIN model proposed and demonstrated in this section has been specifically built for inspection needs, but its basic operation can be expanded and tailored to fit many different prediction needs at Amgen. By leveraging the attribute focused method, the model is able to take readily available forecasts and combine them with any product attributes, materials, and limitations, to generate much more relevant data. It is irrelevant for inspection planning to say there will be a certain number units of a particular drug produced in a particular year, when what is critical is that there will be, for example, 8M high viscosity 1 mL syringes, a data point that spans multiple products and is relevant to the infrastructure capabilities. This method of data aggregation, translating from product forecasts to capability requirements and focusing on the really relevant data is valuable beyond inspection. Simple modifications to the model allow it to be used for a wide variety of applications where the true interest in a forecast is more focused than just raw SKU level data. Conversations with managers in primary containers, raw materials, demand and inventory planning generated strong interest and requests for variations on the model. The model can quickly be tailored to an area's interest by assigning the appropriate attributes to each SKU and then taking a summation of the product forecast over the proper SKU attributes. This attribute based data aggregation allows decision makers to see the data most relevant to their function, instead of that data which is relevant to a certain product line. The FIN model attribute based approach is applied to raw materials by attaching a bill of materials to each SKU and allows the purchaser to quickly look across five different pipeline products. In this case the model identifies raw materials common to each product and shows consolidated future purchases, giving the purchaser a more relevant dataset than just a production schedule. 43 4 Engineering solutions to critical challenges This research aims not only to discover the gaps in inspection capabilities that will arrive in the next decade but also to investigate engineering solutions to mitigate those gaps. Identified using the FIN model, the most significant capability gaps are high viscosity products, plastic presentations, and the type of protein products. In this chapter, the research explores technical solutions to both high viscosity and proteinaceous product inspection but because off the shelf technology exists for plastic container inspection, this was not pursued as a critical challenge to be solved through engineering. Amgen has worked to develop in house technologies for the inspection of drug product, focusing on particle detection, qualification, and quantification. The research undertaken while on internship at Amgen sought to open new avenues into reliably detecting high density particles, particle quantification, lighting techniques, segmentation algorithms, and acoustic agitation for high viscosity products. Table 3 - Proposed engineering approaches to current and future challenges N.-A~ High density particle module * 0 Segmented inspection regimes 0 Particle illumination enhancements Acoustic streaming ___ *_____ Simulated integrated inspection line The engineering work done during the internship was conducted together with the DPE ASE group in their lab as well as with assistance from the Robotics and Automation Team (RAT). Table 3 shows some of the engineering approaches undertaken and their application in improving inspection capabilities. 44 4.1 4.1.1 Technical challenges addressed and solutions proposed Imaging high density particles The shape of currently used glass vials can prove challenging for inspection systems which look directly orthogonally to the vertical axis. Testing performed on Figure 18 - Bottom of glass vial showing curved base ParticleVision 2 (PV2) showed that high and potential for hidden particles density particles have a tendency to sink to the bottom of vials and become obscured behind the curvature of the base. Image recognition algorithms run on the machine designed to mimic the operation of the static division detection technique failed frequently due to this obscuration effect. This problem occurred most frequently with 100um sized glass beads in higher viscosity solutions as the spin cycle of the vial would fail to suspend the particle long enough above the base curvature to image the particle. CCD Camera Telocentric Particleobscured lens behindbase Figure 19 - Particle obscuration by base curvature While on the commercial scale equipment it is difficult to judge how many of the false negatives are due to this obscuration, the PV2 system allows us to record video and manually examine failures to detect. Experimentation on PV2 showed that for dense particles such as glass beads, obscuration was the most common mode of detection failure. One solution to this, proposed here, is to use a downward angled camera to detect these particles at the base of a glass vial as shown in Figure 21. 45 F~1 Figure 21 - Diagram showing angled camera and lighting setup Figure 20 - High density particle detection module By angling the camera relative to the vial's central axis, we are able to see a full projection of the base of the vial including the previously obscured area. Using 3cc vials filled with 1.7ml Using the approach to detect 70, 100, and 400um glass beads in a mock solution, we are able to achieve 100% particle detection at the 100um and 400um level, as compared to only 90% in the existing horizontally configured SD inspection system.[9] This system uses a basic image processing algorithm named Maximum Image Projection written by the DPE ASE team. This algorithm effectively overlays a sequence of video stills on top of each other, removes static features and noise below a certain threshold, and outputs a score based on the number moving pixels above the threshold brightness. It does not track number of particles or size because in this base zone we are less interested in these properties because a particle in the base zone is more likely to be large, heavy, and extrinsic, warranting rejection. Due to the relatively low cost of each camera system, this high density particle detection module can be used to augment existing inspection systems. The camera system used here, and shown in Figure 20, can be installed above or next to orthogonal cameras already in place and the data can be used to more reliably detect high density particles. 46 4.2 4.2.1 Acoustic streaming Introduction As discussed in previous sections, the most significant challenge in inspecting high viscosity products is effectively agitating contained particles into suspension and keeping them in motion long enough to be detected. Both Static Division (SD) and camera based systems require particles to move relative to the container to be detected because they rely on background subtraction techniques. By removing static features from the signal or image, these inspection techniques ensure that surface flaws and external defects do not cause false positive rejects. Additionally, particles stuck to the inner wall during the spinbreak-inspect cycle will not be detected. With the existing spin-brake-inspect approach though, high viscosity solutions either do not generate enough relative motion between the fluid and the container, stop moving too quickly after the spin cycle has stopped, or do not lift negatively buoyant particles off of the bottom of the container. In SD type systems, the inspection area is only a thin vertical along the length of the container axis and if the particle does not stay suspended long enough, and spin at high enough speed through that inspection line, the SD system will not detect it. Similarly, even though CCD camera imaging is able to view the entire container, if the particle is not moving relative to the vial, is located at the bottom, or does not move enough to escape background subtraction, it is not detected. In order to mitigate the issue, it is ideal to develop a way to agitate particles and induce movement relative to their container without moving the container relative to the imaging system. Systems where the camera is mounted fixed relative to the container, and the entire assembly put through orbital mixing were considered, as well as several others which all presented significant challenges. The best solution considered was to induce agitation in the product without any detectable movement of the vial at all. After considering non-contact technologies for inducing mixing, it was decided to use Surface Acoustic Waves (SAWs) to generate acoustic streaming within the container. 47 Acoustic streaming is effectively the opposite of sound generated by a fluid flow and occurs when fluids absorb the energy of acoustic waves traveling through them. This acoustic momentum within the fluid generates Reynolds stresses. As acoustic waves travel and dissipate into the fluid, their momentum decreases, creating a gradient in the Reynolds stress and thus driving the fluid in the direction of wave propagation.[ 16] SAWs have been shown to disrupt laminar flow in microfluidics and lead to the formation of mixing vortices.[17-20] and are used in many microfluidic applications to induce mixing and acoustic streaming effects. This approach has also been applied to large scale and highly viscous liquids, such as corn starch at 2,100cp.[21] The ability to apply this acoustic streaming phenomenon to drug products within vials or syringes would allow for fluid and particle agitation concurrent with inspection imaging. Manipulating the contents of a container without moving it on any scale sensitive to an imaging system allows for particles to be agitated into suspension while they are imaged and the static background of the container itself is subtracted. While there are still concerns, warranting further investigation, about the effect of the acoustic energy on the proteins in the drug product, acoustic technology has been applied to cell flow cytometry for the purpose of aligning the cells prior to imaging [18] as well as in protein detection [22] implying that the effect is not harmful to the proteins. 4.2.2 Experimental setup - Rigid frame holding vial and piezoelectric crystalunder como~ressive load Vial undertest piercelectcrstl Figure 22 - Experimental setup for acoustic streaming in glass vial 48 As shown in Figure 22 an experimental setup was built to test the effectiveness of acoustic streaming on glass vials filled with particulates. Testing was primarily conducted on a 5cc vial filled with a mimic drug product solution. The vial was stoppered but not crimped with the standard aluminum seal. The vial was spiked with 100um polystyrene beads which are negatively buoyant in the solution but do not sink as rapidly as glass beads. A frame was constructed out of MicroRax@ extruded aluminum framing with an adjustable upper bracket which allowed for a range of vial heights as well as varying compressive stresses to be applied to the vial. The piezoelectric crystal and the vial are inserted inside the frame and the upper bracket is pressed down with 5-1Olbs of force and fixed to the frame. This force ensures good contact and acoustic transmission between the vial and piezoelectric element, while the rubber stopper at the top allows for a degree of compressibility such that the vial is able to vibrate with the crystal. The piezoelectric crystal is silver soldered to two 18 gauge copper leads on each side which are connected to a ±200 volt amplifier with a 20x gain. The amplifier is in turn connected to a function generator generating square waves. During testing the amplitude and frequency of the output waves are varied. 4.2.3 Testing Based on research done in flow cytometry, the goal was to achieve acoustic streaming at the resonant frequency of the vial. The propagation of the acoustic waves generated within the vial are difficult to predict as the curvature of the base and the air gap at the top are non-uniform, and while equations exist to determine these frequencies, none were found which would have adequately approximated the situation. It was decided that it would be more effective to empirically determine the most effective frequencies for acoustic streaming by slowly sweeping the frequency through the range of the signal generator. The driving voltage was swept from 0.1Hz to 10MHz and frequencies were noted when agitation and streaming occurred in the vial. No significant effect was seen below 20kHz and above 150kHz during primary sweeps and the region of interest was narrowed to this spectrum. Within this region, significant streaming effects were seen at 31kHz, 56kHz, and 86kHz with the strongest occurring around 56kHz. 49 4.2.4 Results Figure 23 - Diagram of fluid flow and sequence of toroidal streaming stills By sweeping driving waves through various frequencies, a sequence of modes would develop as the pressure waves within the vial developed. Several modes of activity were seen during testing: toroidal streaming, vortex swirling, node aggregation, and localized mixing. All of these modes lifted the polystyrene beads from the base of the vial for easy imaging, and all except the node aggregation kept them in constant motion relative to the container. During node aggregation, a standing wave is created within the vial and particles are driven towards low points of pressure and energy as shown in Figure 24. Typically, the first mode seen, approximately 1kHz before reaching the target frequency is nodal aggregation where low pressure nodes appear in the sample and particles seek these points out, clustering into bunches. As the target frequency was reached these nodes would dissipated and toroidal acoustic streaming would begin as shown in Figure 23. In Figure 25, vortex streaming was achieved at approximately 86kHz whereby the particles lifted into a horizontal nodal plane at approximately 85kHz and then begin swirling vigorously at 86kHz. e-ar Figure 24 - A 100um glass bead and an air bubble manipulated about a ode by varying wave amplitude 50 I Figure 25 - Sequence of stills of nodal aggregation followed by vortex streaming A final test was done with a single 100um glass bead in a 5cc vial filled with the same mock solution. In this test, we were able to achieve altitude control over the particle and an air bubble. By increasing the driving voltage amplitude of the applied acoustic signal could drive the particle and air bubble towards a node located in the center of the vial and by releasing the bead and the bubble would sink and float. This result suggests potential for particle manipulation and clustering through acoustic forcing. 4.2.5 Discussion The application of acoustic streaming to automatic visual inspection could be a very valuable tool, especially when dealing with high viscosity fluids and heavy particles. As shown in these tests, surface acoustic waves applied to a vial are capable of lifting negatively buoyant particles from the base of a vial and set them in motion in the region of interest. By generating this motion, imaging algorithms can effectively background subtract while still retaining the particles of interest. Generally, agitation occurred within a few seconds of sweeping into the target frequency and sustained the particles in motion as long as the surface acoustic waves were applied. While the forced nature of the acoustic streaming would render most particle tracking algorithms useless in differentiating particles during acoustic forcing, they could be effective as the forcing effects dissipate quickly when the acoustic waves are removed, and the particles settle back to the bottom. While this approach is more time intensive than the standard spinbrake-inspect approach and is likely not appropriate for low viscosity products, which can already be effectively agitated, it demonstrates potential value for the future high viscosity products. 51 In order to fully validate this as a non-destructive means of evaluation, studies are needed to examine whether the applied acoustic waves have any detrimental effects on the proteins contained within a product, but as noted [22] these technologies are already used on protein pharmaceuticals and whole cells, implying the risk is likely low. Since the conclusion of this testing was largely positive and the internship was projected to end within the near future, a third-party has been identified to continue investigation into this technology. The company has significant experience with acoustic systems and Amgen has formed a partnership to continue driving the technology forward as applied to inspection. 4.3 Segmented inspection regimes As seen with the high density particle detection system, different areas of a container display different behaviors under inspection. This is dependent on the type of particle under inspection as well as the geometry and material of the container itself. Current inspection systems treat the internal fluid as a single body, attempting to apply the same inspection procedure across the entire volume. Research conducted during this internship suggests there are at least three discrete regimes within a container under inspection. We define these as the Meniscus Zone, Body Zone, and Base Zone as shown in Figure 26. Meniscus Zone Body Zone Base Zone Figure 26 - Segmented inspection zones Meniscus Zone: This zone consists of the uppermost section of the drug product in the container and the meniscus formed by adhesion of the product liquid to the glass containers. This is currently the most challenging region for effective inspection, both manual and automatic. The meniscus reflects and distorts the light coming into it making it difficult to see particles that are in this region. Similarly to the 52 base zone, floating particles, such as rubber from the stopper, will be obscured behind the meniscus. As seen in Figure 26, the meniscus can become "washed out" by the inspection lighting making it difficult to detect any particles trapped in this zone. Additionally, bubbles and foam in the meniscus on certain products can obscure particles, or be detected as particles themselves. In the spin-brake-inspect cycle, one critical factor is waiting for the meniscus to recede so that the maximum amount of product can be imaged in a static inspection window. During the recession of the meniscus though, heavier particles can drop back in the base zone again and be lost. A dynamic meniscus region as depicted in Figure 5 would allow for maximum inspection time and assuming the behavior of every container of the same product and fill level is repeatable could be hardcoded into an inspection algorithm. As far as detecting particles trapped in the meniscus, this may require angled camera's, enhanced lighting techniques, as well as new methods to remove bubbles and foam which cause false rejects. Body Zone: This zone is currently the best understood and most straightforward to inspect. It is characterized by a large volume which is easy to light. Particles found here are generally neutrally buoyant, such as cellulose, protein aggregates, hair, and glass lamellae. Both existing automated inspection systems as well as those developed at Amgen do well in detecting particles in this zone. Amgen's approach to particle tracking and quantification may prove more fruitful in the long term here, but existing SD systems are effective at detecting most particles in this zone as long as the particles pass through the line SD sensor. As this region is relatively well advanced, this research will limit its focus here. The SD technology tends to be less sensitive to semi-transparent objects such as protein aggregates. Base Zone: As discussed in 4.1.1, this zone has a tendency to collect heavier, denser particles with low drag coefficients which are not drawn into the vortex swirling during the spin-brake-inspect cycle. Curvature in the base of a vial, or the "funnel" shape of the base of syringe can easily hide particles from inspection. Particles in this base region are especially susceptible to lensing effects, either making them appear larger or smaller or, especially challenging for tracking systems, particles change size and shape as they move in this region. It is proposed in this research that the standard tracking or imaging algorithms 53 are discarded in this region and a binary approach is taken, attempting only to determine whether a particle exists but not its size or shape. If discovered in the base zone after spinning, it is likely that the particle is heavy, large, and extrinsic and the vial should be ejected for further inspection. Approaching each of these areas distinctly will provide better overall inspection efficiencies because this segmented approach takes into account the nuances of each region. Using an approach like that outlined in 4.1.1 for the base zone can increase the overall accurate reject rate for heavy, dense particles from approximately 90%, to close to 100%. 5 Recommendations and Conclusion This research explores the future inspection capabilities required by Amgen's evolving drug product pipeline and identified the magnitude of the challenges revolving around the products' increasing viscosity, changing presentation material, and quantification of particulate matter. By compiling raw forecast data with information gleaned from interviews, research, and experimentation, a forward looking, attribute based forecast is created. This forecast method is more robust for infrastructure planning purposes because it looks past the typical metrics of quantity and location, focusing on what attributes of future product must be taken into account. In the case of drug product inspection at Amgen, capacity planning alone would not reveal the future capability gaps, leaving the company vulnerable to the high cost of defaulting to manual inspection on AVIM uninspectable products. This manual inspection cost not only arises from the direct labor costs calculated in 3.3.3, but also from the significantly increased footprint of inspection space, reduced throughput, less accurate detection, and lowered repeatability or robustness. The biopharmaceutical industry is moving towards more automated inspection in order to combat these costs and inefficiencies but, as seen in this research, several technological hurdles must be cleared before AVIM are capable of inspecting these types of therapeutic protein products. 54 Amgen has been proactive in developing the technology to inspect for particulate matter currently beyond the capabilities of existing non-destructive visual inspection equipment. The research performed during the internship, some of which is presented in this document, worked to develop engineering solutions to improve upon that capability through inspection zone segmentation, improved lighting, new camera and lighting angles and algorithms, and new methods of forcing particle motion. These approaches help to move Amgen's technological capability closer to what will be required but there are still several paths that should be followed in order to ensure that when these pipeline products land on the manufacturing floor, it will be possible to efficiently automatically inspect them. Recommendations and pathways for operational improvements and technology investments are outlined in the following sections. 5.1 Operational Improvements This section addresses shorter term operational improvements that could be undertaken to improve inspection efficiency at Amgen. 5.1.1 Global Infrastructure and Vendor Relationships As noted earlier, Amgen has recently acquired a new facility on the outskirts of Dublin, Ireland. This new facility was acquired from Pfizer and, while an excellent facility, brings a new set of employees, processes, and equipment into Amgen. With this acquisition, Amgen will be using visual inspection equipment from a variety of vendors across different sites. Some machines are modern, and some are over fifteen years old, and each has different technologies, processes, and overall performance. This nonstandardized equipment means that each time a technology transfer is performed between sites, the AVI process and settings need to be retuned, tested, and validated. As inspection is a relatively repeatable process compared to drug substance and drug product manufacture, if machines are identical between sites, the complexity of a technology transfer could be simplified through the use of standardized AVIM across clinical and commercial sites. 55 Furthermore, as the drug product pipeline characteristics change, as detailed in this research, it will be important to develop long term relationships with vendors. Third party vendors are the most likely and qualified to be able to develop commercial scale AVIM with the requisite capabilities and it is in Amgen's best interest to promote an integrative relationship with these vendors. Recent AVIM purchases have departed from the installed AVIM base at Amgen and to some degree weakened the relationship with currently installed vendor. The decision was made correctly for technical reasons, but may have ignored the long term interests of the supplier partnership. It is the opinion of this author that with recent decisions as given, Amgen needs to fully embrace the new vendor as a partner. Amgen's mix of inspection needs is unique among drug companies and it is unlikely that an off-the-shelf AVIM will be developed that will meet its future needs. In this kind of environment, Amgen needs to develop a partnership with an inspection equipment manufacturer. They can leverage the vendor's expertise to develop the right technology in advance of the critical pipeline products. An integrated partnership model will allow Amgen to develop trust with the vendor, sharing technology and the FIN model. Amgen's DPE ASE group has developed some very advanced in house particle inspection technologies which will be very valuable in the future, but the ASE group lacks the experience and resources to integrate this technology into a commercial scale AVIM, though these are strengths of the equipment vendors. By moving to a more standardized platform across global manufacturing sites, and developing an integrated relationship with a single vendor, Amgen will both be able to reduce the complexity of the technology transfer process and begin a partnership to address its future inspection challenges. 5.2 Technology Investment Table 4 - Truncated net present cost of inability to automatically inspect various attributes (for full table see 7.3) Discount (WACC Post 2020 Ca Growth Est. Particulation 5cc 10cc 20cc 50cc Plastics High Viscosity -10% 5% 10% 20% 30% 30% 35% Net Present cost 0 +Risk 14% 14% 19% 19% 14% 14% 24% $ $ $ $ $ $ $ 1,805,400 3,417,959 39,966 1,737,528 1,940,572 6,566,540 1,641,037 2011 $ $ $ $ $ $ $ 02012 36,225 37,871 55,325 20,747 132,622 56 $ $ $ $ 932013 208,477 329,316 87,637 $ - $ $ - $ $ $ $ $ $ $ 02014 379,663 652,117 22,870$ 150,827 20,747 49,397 - $ $ $ $ $ $ 02015 495,702 735,764 11,269$ 287,657 20,747 24,094 $ $ $ $ $ $ a 499,737 804,327 46,901 247,481 5,434 82,329 - The FIN model developed in this research gives Amgen long-term visibility into its future inspection infrastructure needs but this model is only valuable if it is regularly updated and used to make proactive decisions. Currently the model focused on three main drivers of uninspectable product, viscosity, presentation, and protein aggregation and particulation. We will briefly outline general technology investment strategies for these areas and basic timelines for investment. The inspection costs of each capability gap have been broken down in this section to generate an economic incentive for investment in each of these gaps. The cost of manual inspection is projected through 2025, an approximate life for these pipeline drug products, based on the quantities and inspection throughput rates. For years where there is no data (beyond 2020) conservative growth estimates are made based on understanding of the product portfolio. Discount rates are based off Amgen's overall cost of capital derived from the equity market combined with an additional discount based on risk of that specific product or presentation actually passing all trials and going to manufacture. For instance, the high viscosity products' forecasts have a lower likelihood of materializing and therefore higher discount rate than some of the proteinaceous products which are already in production. 5.2.1 High viscosity products High viscosity products represent the clearest and most significant inspection challenge over the next ten years. This challenge has been recognized by Amgen but there has been little internal research focused on it, or evidence of communication with vendors. The research performed during internship on agitating particles in high viscosity fluids and imaging particles in the base of the container are first steps in developing a robust portfolio of possible solutions to the challenge. This issue is not expected to arise until 2016, costing only $0.1 to $0.2M in that first year, but is expected to grow beyond $1.2M in inspection costs each year by 2020. A first-order estimate of the 2012 net present cost, through 2025, is $1.6M. This is a conservative estimate using an elevated 24% discount rate due to the inherent uncertainty of these products passing clinical trials and going to commercial manufacture but if they are launched, the technology is not in place to automatically inspect them effectively and efficiently, resulting 57 in significant manual inspection costs. This cost implies that Amgen should be investing this order of magnitude into acquiring technologies for effective automated inspection of high viscosity products. This investment should be spread over a portfolio of potential solutions to protect in the case of the failure of any to mature in time. Currently, an investment has been approved to pursue further the acoustic streaming method discussed in this research, but it is critical that more options are considered to increase likelihood of a viable solution. This will be done most easily through coordination with suppliers and discussion of the potential risks and rewards of various approaches to high viscosity product inspection. 5.2.2 Presentation mix The increased diversity of presentations discussed in 3.1.2.2 means that the standard equipment for 3cc vials and Icc syringes will no longer cover the vast majority of products over the next ten years. In 2012, these two presentations will account for 99.6% of products sold, in 2016 it will be down to 95.8% and by 2020, only 67.2%. The difference is made up mostly in plastic cartridges and syringes combined with a few larger volume glass vial types as well. For larger glass vials, the solutions are fairly straightforward and require the purchase of change parts for existing, installed AVIMs. The economic rationale for doing this varies by vial size though. 5cc, 20cc and 50cc vials will be used in larger quantities in the future and the cost of manually inspecting these is high given the long time it takes to manually inspect such a large container. The net present cost of inspecting these presentations is $3.4M, $1.7M, and $1.9M. 10cc vial use is lower and, depending on the cost of the change parts, carries the least compelling economic incentive to upgrade with an NPV, through 2020, of only $0.4M. Most of these solutions are more straightforward than high viscosity products or proteinaceous particles and it is clear there is a lot of value to be captured through the pro-active procurement of change parts, inspection equipment, and processes for these alternative presentations. 5.2.3 Proteinaceous products Products which form protein aggregates have been under scrutiny for several years by Amgen and this is one area where there has been a proactive effort to mitigate the challenge of inspecting products which by 58 their nature have particles in them, even if the particles are not large enough to compromise safety or efficacy and warrant rejection. Currently, many of these products must be inspected manually because installed AVIMs are unable to non-destructively evaluate 100% of containers and effectively discern size and quantity of particles. This leads to unnecessarily high rejection rates of proteinaceous particles and high cost of lost product. Humans are currently better at determining whether the particles within a container fall within prescribed limits but are limited in their ability to count and size particles unaided. The ParticleVision systems developed in the DPE ASE group are focused on determining particle type, size, and quantity, giving readouts of the contents of a vial but are not yet ready for commercial scale inspection. Currently this system is best suited for small scale long term stability studies but has significant potential if paired with a high throughput system as well. As noted in the FIN model discussion, Amgen is actively working to eliminate protein aggregation through product formulation. The projected number of proteinaceous products is not projected to grow significantly in the future, though the potential exists if several products which are still on stability studies show signs of particulation. The net present cost of inspecting these products manually through 2025 is estimated to be $1.8M through 2020, the bulk of which occurs between 2014 and 2018 on products that are highly likely to be in commercial manufacturing during this period. This high probability and high cost in the near-term warrants significant investment in the infrastructure to develop these solutions. 59 5.3 Conclusion Much has been discussed about the changing pipeline of biotechnology products and the challenges they pose to currently installed Automatic Visual Inspection Machines, and the core recommendation is that targeted, pro-active investment can cut millions of dollars in operating costs each year and improve overall product quality, operational efficiency, and compliance. The capability gaps discussed in this research are understood and are expected by many biotechnology managers but the true scale was poorly defined. In order to define this scale better and develop an economic rationale for investment, large data sets were aggregated and translated into attributes relevant to inspection infrastructure. Generic product forecasts are unable to properly define the requirements of an inspection system if pure throughput capacity is not the limiting factor. This research takes these frequently updated forecasts and marries them with product attribute data to generate an attribute focused requirements forecast which gives a clearer picture of the challenges faced by inspection infrastructure. By looking at this data and the costs associated with each, we are able to generate a quantitative economic rationale for investing in solutions for each challenge. It is beyond the scope of this research to select specific technologies and vendors, but the FIN model, and associated engineering research, will allow current and future decision makers at Amgen to quantitatively understand the costs associated with lack of technology investment and create the most value through economically driven decision making. 60 6 References [1] R. Cordoba-Rodriguez, "Aggregates in MAbs and Recombinant Therapeutic Proteins: A Regulatory Perspective," BioPharm International,2008. [2] B. S. Sekhon, "Biopharmaceuticals: an overview Abstract :," Cytokines, vol. 34, pp. 1-19, 2010. [3] S. Silver and B. Analyst, "Biotechnology," Biotechnology, no. August, 2012. [4] DataMonitor, "Amgen, Inc.," 2011. [5] K. W. Sharer, R. Hoffmann, A. Hooper, and D. Mckenzie, "Advancing Science through Strategic Partnerships," Exchange OrganizationalBehavior TeachingJournal,pp. 1-2, 2012. [6] "General Chapter <1> Injections," in United States PharmacopeialConvention, 2011. [7] J. Z. Knapp and P. D. A. J. P. Sci, "The Scientific Basis for Visible Particle Inspection The Scientific Basis for Visible Particle Inspection," Development, 1999. [8] J. a Melchore, "Sound practices for consistent human visual inspection.," AAPS PharmSciTech, vol. 12, no. 1, pp. 215-21, Mar. 2011. [9] N. Rathore, 0. Gonzalez, P. Kolhe, and C. Meshki, "New Vial SKU evaluation for the EISAI Automatic Inspection Machine (EN 28157)," 2007. [10] "Particulates in Liquid," EisaiMachinery US.A. Inc., 2012. [Online]. Available: http://www.eisaiusa.com/technologies/particulates-liquid.asp. [11] J. A. Melchore, "Prerequisites for Optimized Performance of the Eisai 1088W Automated Inspection System," PDA JournalOf PharmaceuticalScience And Technology, 2010. [12] "General Chapter <788>," in UnitedStates PharmacopeialConvention, 2011. [13] B. P. D. Cherney, "Current Regulatory Considerations for the Assessment of Sub-Visible Particles," in WCBP CMC Strategy Forum, 2011. [14] J. F. Carpenter et al., "Overlooking subvisible particles in therapeutic protein products: Gaps that may compromise product quality," Journalofpharmaceuticalsciences, vol. 98, no. 4, pp. 120 1- 1205, 2009. [15] B. Demeule, S. Messick, S. J. Shire, and J. Liu, "Characterization of particles in protein solutions: reaching the limits of current technologies.," The AAPSjournal, vol. 12, no. 4, pp. 708-15, Dec. 2010. [16] J. Lighthill, "Acoustic Streaming," Journalof Sound and Vibration, vol. 61, no. 3, pp. 391-418, 1978. 61 [17] X. Zhu and E. Sok, "Microfluidic Motion Generation With Loosely Focused Acoustic Waves," Sensors And Actuators, pp. 4-5, 1997. [18] Attune, "Tuning acoustic forces for flow cytometry," vol. 1, no. November, pp. 16-18, 2010. [19] A. Wixforth, "Acoustically Driven Programmable Microfluidics for Biological and Chemical Applications," Journal of the Associationfor LaboratoryAutomation, vol. 11, no. 6, pp. 399-405, Dec. 2006. [20] K. Sritharan, C. J. Strobl, M. F. Schneider, a. Wixforth, and Z. Guttenberg, "Acoustic mixing at low Reynold's numbers," Applied Physics Letters, vol. 88, no. 5, p. 054102, 2006. [21] Resodyn, "Polymer-Liquid Blending," 2011. [Online]. Available: http://www.resodynmixers.com/technologies/technical-library-resources/mixing-liquid-polymerwith-liquid-colorant-data-sheet/. [Accessed: 09-Jan-20 11]. [22] Applied Biosystems, "Detecting Red Fluorescent Protein using the Attune @ Acoustic Focusing Cytometer," 2011. [23] Microsoft, "How to use macro examples to delete duplicate items in a list in Excel," Knowledge Base, 2011. [Online]. Available: http://support.microsoft.com/kb/291320. [Accessed: 01-Nov- 2011]. [24] E. D. Consulting, "Fill a ListBox-control with values from another workbook using VBA in Microsoft Excel," ExcelTip.com, 2003. [Online]. Available: http://www.exceltip.com/st/Fill_a_ListBoxcontrolwithvaluesfromanotherworkbook usingVBAinMicrosoftExcel/41 0.html. [Accessed: 01-Aug-201 1]. [25] C_A_P, "Microsoft Excel - Find last Cell Reference with specific Data in." [Online]. Available: http://www.eggheadcafe.com/community/excel/66/10074218/find-last-cell-reference-withspecific-data-in.aspx. [Accessed: 01-Aug-20 11]. [26] Datsmart, "Macro to Open Workbook then use supplied name elsewhere in code," Mr. Excel, 2010. [Online]. Available: http://www.mrexcel.com/forum/showthread.php?t=497348. [Accessed: 01-Aug-201 1]. [27] Yahoo!, "Amgen Inc., (AMGN)," Finance. [Online]. Available: http://finance.yahoo.com/q?s=AMGN. [Accessed: 09-Feb-2012]. 62 7 7.1 Appendix High Density Particle Detection Module Performance Table 5 and Figure 27 show a sample of the performance of the High Density Particle Detection Module's performance. This data was taken manually by running a randomly selected subset of spiked vials several times. The scores reported are outputs of the Maximum Image Projection algorithm built by the DPE ASE team. Higher scores correspond to more detected motion and contrast of the contained particle. Table 5 - Detection scores of HDPD module for glass beads Vial No. Defect (um) Run 2 Run 1 Run 3 0 0 1 0 0 0 0 0 0 0 1 No. Detected 000 0 0 0 13 14 0 0 18 19 0 21 23 0 0 0 1 113 114- 70 35 91 89 117 4 70 62 113 44 46 4 349 200 119 202 292 380 152 303 111 4 127 215 175 4 4 282 396 119 101 4 4 103 254 261 243 4 4 0 0 Run 4 _ 0 21 _ 0 0 0 0 0 0 0 0 116 119 121 70 70 155 483 70 124 122 70 294 204 205 183 40 206 207 208 100 100 100 100 100 301 286 271 221 285 282 355 223 280 240 4 __210 100 297 315 115 169 4 303 304 400 803 1299 927 1454 4 400 1115 1089 1337 1527 4 307 308 400 400 400 400 1285 1736 1281 1470 4 1278 1257 1175 1072 4 1180 1655 1890 1732 4 1652 1684 1645 1467 4 310 312 63 Detection of High Density Particles 10000 1000 0 0 *Run 1 100 tRun 2 ARun 3 10 XRun4 1 100 0 200 300 Particle Size (um) 400 500 Figure 27 - Detection scores of high density particles 7.2 FIN Model Data Aggregation and Distillation Code This section contains the core excel and visual basic functions and code that generate the FIN model. The Excel functions pull from a central worksheet with columns for each attribute and forecast year. The general approach is to find a drug product by SKU and aggregate forecasts for that SKU by year and requested attribute. The Visual Basic code below was developed so that the FIN model would be automatically updateable and provide the user with a GUI for loading in new forecasts. 7.2.1 Commercial and pipeline drug product import: =IF($B2="Commercial",VLOOKUP($D2,INDIRECT(""' & 'ControlPanel'!$C$3 & "'!$F$4:$U$300"),MATCH(""& V$1& "",INDIRECT(""' & 'ControlPanel'!$C$3 & "'!$F$3:$T$3 "),0),FALSE),SUMIFS('PipelineData'!D:D,'PipelineData'!$C:$C,$E2,'Pipeline Data'!$B:$B,$G2)/COUNTIFS($E:$E, $E2,$G:$G, $G2) *1000)/1000000 7.2.2 Breakdown by protein =SUMIF('DrugProdProperties& Forecast'!$F:$F, $A2,'DrugProdProperties& Forecast'!V. V) 7.2.3 Aggregation by presentation 64 =SUMIFS('DrugProdProperties& Forecast'!V V, 'DrugProdProperties& Forecast'!$G.$G,Presentation!$A3,'DrugProdProperties& Forecast'!$I$IPresentation!$C3,'Drug ProdProperties& Forecast'!$J:$J,Presentation!$B3) 7.2.4 Aggregation by viscosity =SUMIFS('DrugProdProperties& Forecast'!V:V, 'DrugProdProperties& Forecast'!$0.:$, "> "&'Control Panel'!$G20,'DrugProdProperties& Forecast'!$0.$, "< "&'Control Panel'!$G21) 7.2.5 Aggregation by protein aggregates =SUMIFS('DrugProdProperties& Forecast'!V:V'Drug ProdProperties& Forecast'!$Q:$Q,"N",'Drug ProdProperties& Forecast'!$R:$R,">2") 7.2.6 Inspectability By viscosity: =SUMIFS('DrugProdProperties & Forecast'!V. V, 'DrugProd Properties& Forecast'!$0.$, "> "&$A8,'Drug ProdProperties& Forecast'!$G.$G,$C8,'DrugProd Properties& $D8, 'DrugProdProperties& Forecast'!$J$J,$B8) Forecast'!$I$I, By presentation: =SUMIFS('DrugProdProperties & Forecast'!V V'Drug ProdProperties & Forecast'!$G.$G,$C16, 'DrugProdProperties& Forecast'!$I:$I,$D16,'DrugProdProperties& Forecast'!$J.$J,"> "&$A 16, 'DrugProdProperties& Forecast'!$J:$J,$B16) By protein aggregation: =SUMIFS('DrugProdProperties& Forecast'!V V'Drug ProdProperties& Forecast'!$G:$G,$C28,'DrugProdProperties& Forecast'!$SS,"Manual",'DrugProdProperties& ProdProperties& Forecast'$J:$JInspectability!$B28) Forecast'!$I:$I,Inspectability!$D28,'Drug 7.2.7 Visual Basic Macro for commercial product forecast import Private Sub GetNewData Initialize() Me.BLoadData.Enabled = False End Sub Private Sub BCloseClick() Unload GetNewData End Sub Sub B LoadDataClick() Dim i As Integer Dim sht As String Dim Source As Workbook For i = 0 To LB SelectWorksheet.ListCount - 1 If LBSelectWorksheet.Selected(i) = True Then sht = LBSelectWorksheet.List(i) 65 End If Next i With Me.TxtSourceFile Dim szToday As String Dim DataSheet As String Dim DataCell As String Dim UpdatedCell As String 'set the control panel cell where the source data sheet is referenced DataCell = "C3" UpdatedCell = "C4" 'create date stamp for new data szToday = Format(Date, "mmm-dd-yyyy") 'sets the source file location from the user selection Set Source = Workbooks.Open(Me.TxtSourceFile.Text) 'copies the selected DP data worksheet to the main spreadsheet Source.Sheets(sht).Copy ThisWorkbook.Sheets(3) 'close source workbook without saving changes Source.Close False 'determine what data is being referenced currently ThisWorkbook.Sheets(3).Activate 'move old data spreadsheet to end of workbook DataSheet = Sheets("Control Panel").Range(DataCell) ActiveWorkbook.Sheets(DataSheet).Unprotect ActiveWorkbook.Sheets (DataSheet) .Move After:=ActiveWorkbook.Sheets ("Old Data--- ThisWorkbook.Sheets("Control Panel").Range(UpdatedCell) = szToday 'if the user loads data from the same day more than once, throw error On Error GoTo ErrHander: Sheets(DataSheet).Name = Sheets(DataSheet).Name & " Old" 'Name the new data sheet by date and set reference in the control panel ThisWorkbook.Sheets(3).Name = "DP Data " & szToday ThisWorkbook.Sheets("Control Panel").Range(DataCell) = ThisWorkbook.Sheets(3).Name 'set simple protection so the user doesn't change the raw data ThisWorkbook.Sheets(3).Protect MsgBox ("Data Loaded") End With Exit Sub ErrHander: MsgBox "Duplicate date data in spreadsheet, please remove old data" " hNS") Sheets(DataSheet).Name = Sheets(DataSheet).Name & Format(Time(), Old" ThisWorkbook.Sheets(3).Name = "DP Data Sheets ("Control Panel") .Range (DataCell) Resume Next ThisWorkbook.Sheets(3).Protect MsgBox ("Data Loaded") 66 " = & szToday ThisWorkbook.Sheets (3) .Name & End Sub Sub BSelectSourceClick() Dim vFile As Variant 'Showing Excel Open Dialog Form vFile = Application.GetOpenFilename ("Excel "*.xl*", 1, "Select Excel File", "Open", Files (*.xl*)," & False) 'If Cancel then exit If TypeName(vFile) = "Boolean" Then Exit Sub End If TxtSourceFile.Text = vFile 'Open selected file Me.ListWorksheets (vFile) End Sub Sub ListWorksheets(sFile) Dim ListItems As Variant, i As Integer Dim SourceWB As Workbook With Me.LB SelectWorksheet .Clear 'remove existing entries from the listbox ' turn screen updating off, ' prevent the user from seeing the source workbook being opened Application.ScreenUpdating = False ' open the source workbook as ReadOnly Set SourceWB = Workbooks.Open(sFile) Dim ws As Worksheet For Each ws In SourceWB.Worksheets LB SelectWorksheet.AddItem (ws.Name) Next ws 'get the values you want 'close the source workbook without saving changes SourceWB.Close False Set SourceWB = Nothing Application.ScreenUpdating = True End With End Sub Dim Dim Sub DelDups TwoLists() iListCount As Integer iCtr As Integer ' Turn off screen updating to speed up macro. Application.ScreenUpdating = False ' Get count of records to search through (list that will be deleted). iListCount = Sheets ("sheet2") .Range("A1:A100") .Rows.Count ' Loop through the "master" list. For Each x In Sheets ("Sheetl") .Range("Al:A10") ' Loop through all records in the second list. For iCtr = 1 To iListCount Do comparison of next record. To specify a different column, change 1 to the column number. If x.Value = Sheets("Sheet2").Cells(iCtr, 1).Value Then ' If match is true then delete row. Sheets ("Sheet2") .Cells (iCtr, 1) .Delete xlShiftUp ' Increment counter to account for deleted row. 67 iCtr = iCtr + 1 End If Next iCtr Next Application.ScreenUpdating = True MsgBox "Done!" End Sub Private Sub LBSelectWorksheetClick() Me.BLoadData.Enabled = True End Sub Private Sub LBSelectWorksheetDblClick(ByVal Cancel As MSForms.ReturnBoolean) Call BLoadDataClick End Sub Sub ImportCommercial() Dim vFile As Variant 'Showing Excel Open Dialog Form vFile = Application.GetOpenFilename("Excel Files "*.xl*", 1, "Select Excel File", "Open", False) (*.xl*)," & 'If Cancel then exit If TypeName(vFile) = "Boolean" Then Exit Sub End If Modulel.UserFormInitialize End Sub (vFile) Sub MsgBoxAllMySheets() Dim sht As Worksheet For Each sht In Sheets MsgBox sht.Name Next sht End Sub Sub UserFormInitialize(sFile) Dim ListItems As Variant, i As Integer Dim SourceWB As Workbook With Macrol.WorksheetSelect .Clear ' remove existing entries from the listbox turn screen updating off, prevent the user from seeing the source workbook being opened Application.ScreenUpdating = False ' open the source workbook as ReadOnly Set SourceWB = Workbooks.Open(sFile) Dim sht As Worksheet For Each sht In Sheets MsgBox sht.Name Next sht ListItems = Sheets ' get the values you want SourceWB.Close False ' close the source workbook without saving changes Set SourceWB = Nothing ListItems = Application.WorksheetFunction.Transpose(ListItems) ' convert values to a vertical array For i = 1 To UBound(ListItems) .AddItem ListItems(i) ' populate the listbox 68 Next i .ListIndex = -1 ' no items selected, set to 0 to select the first item Application.ScreenUpdating = True End With End Sub Sub BLoadNewData() Load GetNewData GetNewData.Show End Sub Public Function GetLastRowWithData() As Long Dim ExcelLastCell As Object, lRow As Long, lLastDataRow As Long, 1 As Long Set ExcelLastCell = ActiveSheet.Cells.SpecialCells(xlLastCell) lLastDataRow = ExcelLastCell.Row lRow = ExcelLastCell.Row Do While Application.CountA(ActiveSheet.Rows(lRow)) lRow = lRow - 1 Loop = 0 And lRow <> 1 lLastDataRow = lRow GetLastRowWithData = lLastDataRow MsgBox lLastDataRow End Function Many functions and code frameworks adapted from [23-26] 7.3 Net present cost of manual inspection of various capability gaps Post 2020 Discount (WACC Growth Est.3+ Risk) Cause Particulation 5cc 10cc 20cc 50cc Plastics High Viscosity 2014 $ $ $ $ $ $ $ -10/0 5% 10% 20% 30% 30% 35% 02015 495,702 735,764 11,269 287,657 20,747 24,094 $ $ $ $ $ $ $ 14% 14% 19% 19% 14% 14% 24% 0 02016 499,737 804,327 46,901 247,481 5,434 82,329 - $ $ $ $ $ $ $ 477,508 827,742 103,248 326,228 95,707 301,773 153,011 Net Present Cost $ $ $ $ $ $ $ 1,805,400 3,417,959 398,966 1,737,528 1,940,572 6,566,540 1,641,037 0 2017 $ $ $ $ $ $ $ 201 417,408 735,293 142,525 521,348 279,301 603,546 378,304 $ $ $ $ $ $ $ 2018 $ $ $ $ $ $ $ 2012 36,225 37,871 55,325 20,747 132,622 0 $ $ $ $ $ $ $ 2019 02013 208,477 329,316 87,637 0 $ $ $ $ $ $ $ 2020 379,663 652,117 22,870 150,827 20,747 49,397 - 3 537,608 813,303 187,603 903,972 454,456 1,056,205 $ $ $ $ $ $ 333,698 839,314 206,496 607,587 573,009 1,710,047 $ $ $ $ $ $ 132,296 468,648 219,588 624,149 608,617 2,263,297 761,756 $ 1,116,035 $ 1,261,360. Weighted Average Cost of Capital(WACC) calculatedto be 9% based onfinancial metrics as of March 2012, collectedfrom Yahoo! Finance.[27] 69 This page intentionally left blank. 70