Agnes Shanley is senior editor of Pharmaceutical Technology.

Agnes Shanley

	Like many crucial regulations, good laboratory practices (GLPs) were enacted in 1979 after FDA observers found serious problems in documentation, training, and data integrity at a number of US research labs (1). Decades later, regulators still find deficiencies in the way that some companies’ labs approach data integrity, training, and standard operating procedures (SOPs). Another major problem is reproducibility. According to the Global Biological Standards Institute (GBSI), 50% of published preclinical research cannot be reproduced, a problem that results in product development delays and wastes $28 billion/year in the US alone (2). Culprits were found to be biological reagents and reference materials, study data, and lab protocols.

	A number of tools are being developed to help lab scientists capture and use more data, for example, LabStep, an interactive digital platform designed to help scientists get around some of the deficiencies of electronic lab notebooks (ELNs) and refer directly to protocols, SOPs, and other important data (3).

	LabTwin introduced a new voice-activated lab assistant at BIO 2019 in Philadelphia. Combining artificial intelligence, voice recognition, and machine language, the hands-free device allows researchers to document steps taken and save explicit details that cannot currently be saved in ELNs (4).

	Ultimately, compliance depends on following best practices. Stuart Jones, regulatory quality assurance professional in good laboratory practice (RQAP-GLP) and director of quality assurance at PPD Laboratories’ Bioanalytical Laboratory shared recommendations with 

: Because we work in such a regulated environment, a seemingly minor matter can have a significant impact on quality. As such, training is an important best practice, from the time of hire, to retraining when a deviation occurs. Annual refresher training as well as specific group remedial training also should be provided when needed. Meanwhile, the use of automated or electronic systems, such as ELNs, can be especially beneficial in maintaining the most accurate documentation.

: How do you recommend that companies tackle training?

: Initial training, especially with newer employees, can be done through reading, lecture, and/or some type of knowledge or learning assessment, but the best results occur when that theoretical work is followed up and supplemented by hands-on training. This is accomplished most effectively by teaming new employees with experienced staff using training goals established within a predetermined curriculum. Some measure of refresher training should be required on at least an annual basis and it should be consistent across all experience levels. Metrics around unplanned protocol and SOP deviations, as well as human error, should be used to gauge the effectiveness of training plans.

: What best practices do you recommend to make data less siloed and more accessible to those who may need it (e.g., on cross functional teams working at the same company or facility?)

: One of the best ways to establish a more cross-functional approach and enhance data accessibility is to use one system across all sites. If one across-the-board system is not a possibility, then the multiple systems must be able to work in tandem. Data portals and SharePoint sites also can be utilized to securely share information on a real-time basis.

: Reproducibility is a problem for preclinical research. Is that also the case for quality control labs? What do you recommend?

: We have found that, after research and development of the method by our scientists, it is important to involve the sample analysis team in performing some, if not all, of the validation experiments, with technical assistance provided, as needed, by the R&D scientists who developed the method. This approach allows for a shared collaboration between the research and production teams and continues into sample analysis to ensure reproducible results from the developed and validated method. Best practices include following the proper bioanalytical method validation guidances, the bridging of critical reagents, analyst method qualification, and scientific expertise/knowledge of the assay, as well as the use of incurred sample reproducibility testing.

 28(10) pp 14-21 (2015).
	3. Lab Twin, “Voice Activated Laboratory Assistant to Launch at BIO 2019,” Press Release, May 21, 2019.
	4 J. Gould, “How Technology Can Help Solve Science’s Reproducibility Crisis,” Podcast, nature.com, April 25, 2019.


	Supplement: Outsourcing Resources
	August 2019
	Page: s14

	When referring to this article, please cite it as A. Shanley, “GLPs: Better Data Access Needed to Improve Compliance," 

Pharmaceutical Technology Outsourcing Resources

Articles

Tools help lab scientists capture and use more data and work to ensure data integrity.

GLPs: Better Data Access Needed to Improve Compliance

	Research studies suggest that artificial intelligence (AI) is becoming more widely used within the pharmaceutical industry, and that senior executive support will be crucial to piloting and implementing AI-based approaches to improve R&D efficiency. Industry interviews and an extensive survey (1) by the Tufts Center for the Study of Drug Development (TCSDD), the Drug Information Association (DIA), and eight bio/pharmaceutical manufacturers, has found that pharmaceutical and biopharmaceutical companies are adopting artificial intelligence to help improve functions ranging from discovery and development to risk assessment, safety monitoring, and manufacturing. The research involved interviews with AI experts within the industry, as well as a survey of 402 professionals at pharmaceutical and biopharmaceutical organizations, to assess the current state of AI use and implications for the future.

	Survey results found that 70% of respondents use AI in some form, with planning and piloting efforts focused mainly on clinical trials (i.e., patient selection and clinical study recruitment) and identification of data gathering for medicinal products.  Research found that staff skills (55%), data structure (52%), and budgets (49%) are currently the greatest obstacles to increased use of AI in pharma. Nearly 60% of survey respondents said that their companies plan to increase staffing over the next two years, to support AI use or implementation. 

	In addition, TCSDD reports, the survey found that:

		The clinical operations function makes the highest use of AI (61%), followed by pharmacovigilance/safety/risk management (57%), and information technology (IT) (55%).

		42% of respondents reported that AI implementation is not centrally managed at their companies, while 20% indicated that it is managed by R&D and 12% said that it is overseen by the chief information officer.

		59% of respondents plan to expand AI staff through 2020, with the largest staffing increases slated for data scientists, computer scientists, IT specialists, and AI architects.

	“Pharmaceutical and biotechnology companies as well as service providers now rely on AI technologies across all therapeutic areas,” said Mary Jo Lamberti, research assistant professor and associate director of sponsored research at TCSDD. She expects precision medicine and demand for new rare diseases treatments to drive potentially exponential growth in the industry’s use of AI as regulators and industries develop standard policies and regulations to address ethical use, bias, and validation.

	Deep Knowledge Analytics released a report that looked at practices at 50 pharmaceutical companies to identify leaders who are most actively driving the use of AI in their organizations (2). The goal is to benchmark the impact of management support on the use of AI, and overall efforts to increase R&D efficiency, according to Margaretta Colangelo, managing partner at Deep Knowledge Ventures. She expects to find a correlation between level of commitment to AI, and market capitalization growth.

, e-publication, June 24, 2019.
	2. Deep Knowledge Analytics, “Pharmaceutical Division Report, AI-Friendly CEOs and Board Members of Pharmaceutical and Technology Corporations,” 


	Supplement: Outsourcing Resources
	August 2019
	Page: s15

	When referring to this article, please cite it as A. Shanley, “Use of AI is Growing in Pharma,"

Bio/pharmaceutical companies are adopting artificial intelligence for discovery, development, risk assessment, safety monitoring, and manufacturing.

Use of AI is Growing in Pharma

Like many crucial regulations, good laboratory practices (GLPs) were enacted in 1979 after FDA observers found serious problems in documentation, training, and data integrity at a number of research labs (1). Global GLPs establish guidelines for standard operating procedures (SOPs), which explain how to carry out specific tests and how to use and maintain laboratory equipment. GLPs also set requirements for equipment calibration and maintenance, data collection, and investigation and documentation of out-of-specification (OOS) findings.

	Decades later, global regulators still find deficiencies in the way that some companies’ preclinical and quality-control labs approach data integrity, training, and SOPs. Incorrect approaches to investigating OOS conditions are also frequently found in regulatory citations, for GLPs as well as for good manufacturing practices (GMPs).

	Another major problem that can be traced to inadequate compliance with GLPs is lack of reproducibility (i.e., a situation in which other laboratories cannot replicate the results published in original research conducted in an innovator laboratory). In 2015, the Global Biological Standards Institute (GBSI) found that 50% of published preclinical research could not be reproduced, a problem that continues to delay the development of new products and wastes $28 billion/year in the United States alone (2).

	In some cases, researchers found, equipment was not being used efficiently based on vendor guidelines; in others, biological reagents and reference materials, cell lines, antibodies, and reagents were not properly validated (3). Also faulted were incomplete study data and lab protocols.

	Industry executives have pointed to cross-functional data access and utilization as a major challenge for the industry. This challenge affects the lab as well as the plant. IDC Health Insights surveyed 126 biopharmaceutical and pharmaceutical executives in the United Kingdom and the US, and found a significant gap between their need and their strategies for harnessing data (4). More than 98% of respondents said that cross-functional data access was important or very important to their business strategies, and 94% described the ability to apply advanced analytics and/or artificial intelligence the same way. However, 51% of those surveyed said that they did not have a clear strategy in place to help them reach either of those goals.

	A number of digital tools are available to help lab scientists capture and use more data, to improve laboratory efficiency and reproducibility. These tools allow data to be extracted from various software systems, and some of them feature use of machine language and elements of artificial intelligence. One example is LabStep, an interactive digital platform designed to help scientists get around some of the deficiencies of electronic lab notebooks (ELNs). The platform allows users to refer directly to the most relevant protocols, SOPs, and other important data (5) as they work.

	LabTwin is also active in this area, and will introduce a new voice-activated lab assistant at BIO 2019 in June, in Philadelphia. Combining artificial intelligence, voice recognition, and machine language, the hands-free device allows researchers to document steps they have taken and to save explicit details that cannot currently be saved in ELNs (6). Labfolder (7) is yet another platform that has been designed to enable more laboratory data to be saved, accessed, and used throughout any organization.

	Ultimately, compliance with GLPs depends on following best practices. Stuart Jones, regulatory quality assurance professional in good laboratory practice (RQAP-GLP) and director of quality assurance at PPD Laboratories’ Bioanalytical Laboratory shared recommendations with 

What GLP problems do you frequently encounter at pharma and biopharma companies, and how can they be prevented?

: Because we work in such a regulated environment, a seemingly minor matter can have a significant impact on quality. As such, training is an important best practice, from the time of hire, to retraining when a deviation occurs. 

	Annual refresher training as well as specific group remedial training should also be provided whenever needed. Meanwhile, the use of automated or electronic systems, such as laboratory e-notebooks, can be especially beneficial in maintaining the most accurate documentation.

How do you recommend that biopharmaceutical companies tackle training?

Initial training, especially with newer employees, can be done through reading, lecture, and/or some type of knowledge or learning assessment, but the best results occur when that theoretical work is followed up and supplemented by hands-on training. This is accomplished most effectively by teaming new employees with experienced staffers, using training goals established within a predetermined curriculum.

	Some measure of refresher training should be required on at least an annual basis and it should be consistent across all experience levels. Metrics generated around unplanned protocol and SOP deviations, as well as human error, should be used as indicators in determining the course and effectiveness of current training plans.

What best practices do you recommend to make data less siloed and more accessible to those who may need it (on cross functional teams?)

One of the best ways to establish a more cross-functional approach and enhance data accessibility is to use one system across all sites. If one across-the-board system is not possible, then the multiple systems must be able to work in tandem. Data portals and SharePoint sites also can be utilized to securely share information on a real-time basis.

Reproducibility is a major problem for biopharmaceutical preclinical research. Is this problem also seen in pharma quality control labs? What best practices do you recommend regarding the validation of materials and methods?

We have found that, after research and development of the method by our R&D scientists, it is important to involve the sample analysis team in performing some, if not all, of the validation experiments, with technical assistance provided, as needed, by the R&D scientists who developed the method. This approach allows for a shared collaboration between the research and production teams, and continues into sample analysis to ensure reproducible results from the developed and validated method.

	Best practices include following the proper bioanalytical method validation guidances; the bridging of critical reagents; analyst method qualification and scientific expertise/knowledge of the assay; as well as the use of incurred sample reproducibility testing as one of the ways that the bioanalytical lab can prove that the method can be reproduced.

28(10) pp 14-21 (2015).
	3. M. Williams, 

The Case for Connectivity in the Life Sciences

, May 22, 2019.
	5. Lab Twin, Press Release, “Voice Activated Laboratory Assistant to Launch at BIO 2019,” May 21, 2019.
	6. J. Gould, “

How Technology Can Help Solve Science’s Reproducibility Crisis

The existence of tools today that were not available in the past can help ensure successful product quality and compliance with regulations.

The Need to Harmonize Good Laboratory Practices

Over the past few years, after some notable failures at a number of companies (1), global regulators have been paying much closer attention to how pharmaÂ­ceutical manufacturers safeguard the integrity of their data, to prevent accidental manipulation as well as fraud. Data integrity is the foundation of the current good manufacturÂ­ing practices (cGMPs). According to FDA, between January 2015 and May 15, 2016, 21 out of the 28 warning letters issued to pharmaceutical manufacturers centered around problems with data integrity (2).

	As Orlando López, Kansas-based senior site automation specialist for a Big Pharma company, has noted (3), conÂ­nections between different data points (i.e., initial data capÂ­ture, metadata, and records) cannot be weakened or broken because they provide the only objective evidence that operaÂ­tions meet regulatory requirements and are being managed responsibly. These intact connections are also needed for process validation and continuous process verification.

	FDA’s 2016 draft guidance clarified a number of best practices, such as the need for paper and electronic recordkeeping to use the same basic practices, and for an administrative role for managing data, separate from funcÂ­tions that are generating and recording that data. FDA also introduced the ALCOA acronym to emphasize the fact that data must be “attributable, legible, contemporaneous, original (or a true copy), and accurate.”

	In December 2018, FDA updated this guidance (4). “The new guidance serves to clarify data integrity requirements that have consistently challenged organizations. It firms up expectations around data/metadata and the data lifecycle within a risk-based structure of systems and design controls,” says Kir Henrici, CEO of the Henrici Group, a consultant for the Parenteral Drug Association (PDA), which pubÂ­lished best practices on laboratory data integrity in 2018 and plans to publish manufacturing data integrity guidance by the end of 2019.  FDA has emphasized the need for careful risk manageÂ­ment, the importance of audit trails and access controls, and the need to save all records that could be relevant to cGMP compliance.

	FDA’s latest data integrity guidance also underscores the importance of senior management support and involvement in data integrity efforts. “Leadership and culture are key, and, if underdevelÂ­oped, are the starting points for comÂ­pliance risks and gaps,” notes Henrici.

	Training and ongoing communicaÂ­tion with senior management are crucial to establishing and maintaining this support, says López. Expressing potenÂ­tial losses or liabilities in dollars and cents terms can be especially effective, he says, noting a 2017 hacking inciÂ­dent at Merck & Co., which cost $105 million to fix (5) and might have been prevented if investment had been made in data security. He also recommends keeping senior management aware of regulatory citations in a way that allows them to connect data issues to lost batches, product recalls, incorrect labels, or formulation changes. “Senior managÂ­ers needs to hear about money. It gets their attention immediately,” he says.

	It is also important to take a systematic, risk-based approach to data integrity proÂ­grams, says Henrici. Some organizations may roll out data integrity programs in fragments, or as a reaction to a specific event, but then allow efforts to trail off, she says. “Pharmaceutical companies need to structure a systematic, risk-based approach that integrates the quality manÂ­agement system (QMS), which assures the efficacy and sustainability of the entire data integrity initiative.”

	Process mapping is especially important in ensuring that the right documentation is easily accesÂ­sible to staff, but also to regulators. “Companies should track their ‘qualÂ­ity decisions’ to identify where non-cGMP records (e.g., emails) will be required to demonstrate cGMP comÂ­pliance,” says Henrici.

	“You need to understand each speÂ­cific process. Then, if you do a good risk assessment, to record which systems and which data you will focus on, you will be in a much better position to demonstrate full cGMP controls to the authorities,” says López. “If you have done both the risk assessment and the process mapping correctly, there should be no doubt that you are implementing controls to the correct records,” he says. Decisionmaking processes must be documented, says Henrici-not only reported data but any data that support quality decisions-and they need to be justified, retrievable, and reviewable, she says. Maintaining a clear audit trail and using the best approaches to computer system validation will be crucial and will require new training and staffing practices, says Henrici. “Next-generation quality assurance will consist of multiple disciplines or working teams to keep pace with data integrity requireÂ­ments in the era of smart manufacturing, big data management, and artificial intelÂ­ligence,” she says.

	In a sense, however, data integrity is only the first step in a continuum. “Pharmaceutical industry guidelines often refer to data reliability, which is a step beyond data integrity,” says López. “But, ultimately, the goal for our industry is data quality, which is a step beyond reliability. We canÂ­not reach that point with the guideÂ­lines that we have now,” he says. “And cGMP requirements alone will not suffice.” (6) Another challenge is the fact that existing pharmaceutical industry data guidelines don’t go deep enough, to the engineers and IT proÂ­fessionals who are implementing the systems, says López. “ALCOA may give them the ‘why,’ but it doesn’t give them the ‘how,’” he says.

	Annex 11, first published in 1986, addresses data integrity in a more explicit way than the US 21 Code of Federal Regulations (CFR) Part 11, he notes. López recalls one Big Pharma company that correlated Annex 11 with Part 11 computer system requireÂ­ments, which made things easier for engineers and IT professionals impleÂ­menting systems.

	Currently, there are a number of different data integrity guidelines. In pharma, these include regional ones as well as those developed by proÂ­fessional societies such as PDA and the European Compliance Academy (ECA). In addition, the software industry has codes of its own, as does the manufacturing industry in genÂ­eral. The focus of each of the relevant pharma guidelines is different, López says, and the definition of data integÂ­rity in software engineering standards (i.e., as mainly about data security and not changing the records), differs somewhat from its definition in the pharmaceutical industry. The result can be confusing to IT professionals, at least initially, and take time to sort out.

	“I would love to see integraÂ­tion between pharma standards and the software engineering standards, because currently the industry is not speaking the same language as softÂ­ware development academics. For example, for the pharma industry, valiÂ­dation is a process associated with the system lifecycle (SLC), which conÂ­tinues from the start of a project until decommissioning. For software develÂ­opment, validation is just the testing phase of the SLC. So when you hire a new graduate, he or she won’t underÂ­stand the difference,” says López.

	Without understanding the pharma validation concept and the relationÂ­ship between software engineering and software quality, people in pharma’s IT and engineering departments will have different understandings and the resulting programs will be incomplete, he says. López recalls seeing such a situation when asked to consult for one mid-sized pharmaceutical comÂ­pany several years ago. Data security had been made part of the system’s user requirements, but, as one looked at the different processes throughout the product lifecycle, there was one point where the computer system was missing provisions for security, so that anyone could make changes to the data. “There was testing, but no transparency, and in the design document, there had been no steps taken to ensure security throughout the lifecycle,” he says.

	Common standards and definiÂ­tions will be the key to moving the industry beyond data integrity to data quality, López says, explaining that the origin of the International Society for Pharmaceutical Engineers’ (ISPE’s) GAMP 5 guidance (7) is the International Standards Organization’s (ISO’s) 9000-3, which is also the founÂ­dation of the pharmaceutical industry’s guidelines. “Why are we speaking difÂ­ferent languages?” he asks. “We need to synchronize all the different stanÂ­dards (e.g., ISO, ISPE, The Institute of Electrical and Electronics Engineers [IEEE]), and pharma’s so that we are speaking one common language and so that we will understand each other,” says López. Without that common understanding, gaps will persist, some worse than others, he says.

	The pharmaceutical industry’s approach to data integrity and quality will need to evolve as the industry moves to a Pharma 4.0 model and adopts data analytics and artificial intelligence. At INTERPHEX 2019 in April, Henrici spoke of the need for a “Data Integrity 4.0” strategy, a framework for matching IT and data integrity rules. She also noted that FDA approved 12 artificial intelligence algorithms in 2018.

	If the industry is to reap the benÂ­efits of using algorithms and artificial intelligence, she said, companies will need to create multifunctional data governance teams to bring different perspectives to this effort, and facilitate communication between industry and regulators, and data will need to be safe, secure, and relevant (8).

	In the end, the required knowlÂ­edge is already there for implementing new IT approaches, says López, who developed an analytics and business intelligence platform at a Big Pharma company several years ago, before the term “Big Data” had even become a buzz phrase. He advises breaking the problem down to its simplest level: the data wave and input/output (I/O). “With a Big Data project, instead of one, you might have 20 data waves and I/Os. The pharmaceutical industry has been working on all the technoloÂ­gies that industries are exploring (e.g., wireless, industrial internet of things, and Big Data). In the end, each type involves software and hardware, input and output, and there are certain tools best suited for each situation,” he says. “On the fundamental level, we need to understand the relationship between the system, process mapping, data wave sets, and intended use, and simply apply what we have learned during the past 30 years to the new technologies,” adds López (9).

	The industry has come a long way in improving data integrity since the Barr decision in 2005 (1), a landÂ­mark case that resulted in mandates for recordkeeping and the investigaÂ­tion of out-of-specification conditions. New guidance and better integration of existing best practices-not only those designed for pharma’s end users, but for implementation specialists in IT and engineering-promise to push pharmaceutical manufacturers beyond data integrity in the future.

Data Integrity: Detecting and Mitigating Risk

,” presented at the Life Sciences Trainers and Educators Network annual conference (2017).
	2. S. Barkow, “

Current Expectations and Guidance Including Data Integrity and Compliance With cGMPs

,” presented at the International Society for Pharmaceutical Engineers Data Integrity Workshop (Bethesda, MD, June 5, 2016).
	3.  O. López, “

Defining and Managing Raw Manufacturing Data

Data Integrity and Compliance with cGMP, Q&A–Guidance for Industry

 (Silver Spring, MD, December 2018).
	5. C. Souza, 

 38 (12) 28–29, 38 (2018).
	6.  O. Lopez, “Comparison of Health Authorities’ Data Integrity Expectations,” presented at the 4th Annual Data Integrity Validation Conference (Boston, MA, August 2018).
	7.  C. Plagiannos, “

What is GAMP 5 and How Do I Use It Effectively

, montrium.com, Nov. 30, 2015.
	8. A. Shanley, “

, April 15, 2019.
	9. O. López, “Electronic Records Integrity in Data Warehouse and Business Intelligence” in 

Data Integrity in Pharmaceutical and Medical Devices Regulation Operations

 (CRC Press, Boca Raton, FL, 1st ed., 2017), 341–351.

Vol. 43, No. 7
	July 2019
	Pages: 48–50

	When referring to this article, please cite it as A. Shanley, “Removing Gaps in Data Integrity," 

FDA guidance is expected to improve industry practices, but work is also needed to bridge disparate industry and software engineering standards. 

Removing Gaps in Data Integrity

In the pharmaceutical industry, it often takes tragedy to lead to the passage of important laws. The Food, Drug, and Cosmetic (FD&C) Act of 1938, which established FDA’s authority over questions of drug safety in the United States, was passed after more than 100 people in 15 states died. The victims had ingested the first liquid form of sulfanilamide antibiotic, dissolved in a toxic-and at that point untested-solvent, diethylene glycol (1).

	Current good manufacturing practices (cGMPs)-which set firm requirements for the safe manufacturing of finished drug products, and for plant and process design and operation-were passed in 1963 to update the FD&C Act. They came as a response to deaths caused by Winthrop Pharmaceutical’s failure to prevent cross-product batch contamination. The company had used two tableting machines within the same room, interchangeably, to manufacture antibiotic and anticonvulsant tablets. Lax controls led to the release of antibiotic tablets that had been contaminated by the anticonvulsant at two to three times the non-lethal dose; hundreds of people died or were hurt after taking the contaminated antibiotic (2–4).

) Parts 210 and 211, and 212 for some radiopharmaceuticals, cGMPs aim to ensure that pharmaceutical processes and facilities are designed, monitored, and controlled properly, says Michael Kopcha, director of the Office of Pharmaceutical Quality (OPQ), at FDA’s Center for Drug Evaluation and Research (CDER). Unlike voluntary guidances and best practices, cGMPs are the law and must be followed by drug manufacturers. Recently, FDA has extended cGMPs into dietary supplements and compounding.

	FDA last formally revised pharmaceutical cGMPs in 1977; since that time, the agency has published new guidance documents  meant to improve manufacturers’ approaches to risk assessment and process understanding and control, with the 21st Century cGMPs (5) and process analytical technologies (6) guidances in 2004, and the revised process validation guidance (7) in 2011.

	The industry has undergone radical changes, however, since 1977. Instead of focusing on large-volume “blockbuster” drug markets, pharmaceutical manufacturers now juggle larger product portfolios with smaller, globally diverse markets. In addition, they source most APIs from India and China, and outsource more strategic operations to contract research organizations (CROs) and contract development and manufacturing organizations (CDMOs).

	At the same time, the industry has become far more complex with the growth of biopharmaceuticals, the shift from branded to generic pharmaceuticals, and the move to biosimilars and personalized medicine. “When cGMP came into effect, pharma companies were vertically integrated,” says Hedley Rees, managing consultant of PharmaFlow Ltd., based in the United Kingdom. “Quality systems were under the ownership of the companies developing and manufacturing the products, with a shared set of standard operating procedures and business objectives. The people working in the pharma companies all had skin in the game; they shared the pain and the rewards in the journey to regulatory approval,” he adds. The past four decades have seen supply chain complexity increase by an order of magnitude, Rees says.

	Should cGMPs be revisited and updated, given the fundamental changes that have taken place within the industry since they were published? Opinion on the subject is divided. At this point, FDA has no plans for any major revisions of US cGMPs. “We intend to continue modernizing or clarifying the regulations as needed to harmonize them with other FDA regulations and international cGMP standards, and to use guidance for industry as a primary way of developing and explaining our recommendations for complying with the statute and regulations,” says FDA’s OPQ Director Kopcha.

	However, some within the industry believe that cGMPs need an overhaul. “The current CGMP regulations clearly need updating to reflect the current state of complexity in pharmaceuticals materials, processes, products, and supply chain,” says Ajaz Hussain, consultant, head of the National Institute of Pharmaceutical Technology and Education (NIPTE), and formerly head of FDA’s Office of Pharmaceutical Sciences at CDER.

	“FDA’s periodic updates have been a bit like patches on software or add-ons to an existing building. It may take years, but cGMPs need to be modernized,” says consultant Emil Ciurczak, an early adopter of process analytical technology (PAT) and president of DoraMax Consultants.

	The task would be herculean, notes Hussain. “We have done all that we can to avoid having to do the update, with guidelines and guidances. That has been the entire basis of the 21st century initiative,” he says.

	One challenge is that FDA’s guidance documents, whether on PAT or process validation, only present suggestions and companies can-and often do-bypass them, as long as they meet legal product quality requirements. Some observers ask whether some suggested practices might need to become required ones if the industry is to develop a more scientific approach to manufacturing and quality control.

	Industry observers pinpoint the need for more cGMP clarity in:

		Training and employee’s personal development

 211 Part 11 Subpart B, which covers requirements for organization and personnel, should be a high priority target for cGMP clarification and updates. He points to recent FDA warning letters that refer to 

 211, specifying deficiencies in production record review, batch record and maintenance data practices, training, and investigation of batch failures or other discrepancies. 

	Even though regulators may not explicitly refer to Subpart B in their citations, Hussain suggests, companies’ failure to follow its requirements is often the root cause of many, if not most of the cGMP problems that regulators do cite.

	“The cGMPs drafted in 1977 had called for adequate education, training, and experience. But when have we ever defined what ‘adequate’ means for people who work in cGMP, for the inspectors at FDA, for the reviewers at FDA? We have this fundamental flaw in the system,” Hussain says. “So how can we expect people who are not qualified for their jobs to do anything better? “Our attention needs to be focused on being objective,” he adds. Hussain sees professional development as the key to developing the type of managers needed to instill a “culture of quality” within the industry.

	Inadequate management training and development may also have implications for the industry’s use of technology, says consultant Gary Ritchie. He uses digital records as an example. Like paper-based records, digital records are subject to the predicate rule; they must be retained for a set period of time after the drug expires. But digital records offer an immediate transparency that paper does not. “In some cases, companies are afraid to be stuck with electronic records that actually show how management made decisions when they processed something. You cannot erase the record,” he notes.

	“The reality is that people are still working on paper in the 21st century digital world, because it is more flexible,” says Ritchie. “FDA has suggested that, in the future, it may not allow hybrid (i.e., paper-on-glass) systems, but until the science of manufacturing evolves, change won’t take place,” he says. “FDA wants to hold up as an example those companies that have moved to digital records, but there are still so many more facilities that use paper and whose practices remain stuck in the 1900s.”

	Another area where cGMPs could be shored up, observers say, is in process validation and delineating the expectation of statistically valid sampling and statistical process control. “That was the stated principle behind FDA’s process validation guidance in 2011,” says Hussain.  As he notes, only a handful of Big Pharma companies appear to be using modern process validation principles correctly. “The rest are struggling, and haven’t even begun to use the concept for new, let alone legacy, products,” he says.

	One problem pointed out by FDA compliance officer Grace McNally in a 2011 presentation is a failure to use the best statistical methods and validation principles. As she noted, “At times, compendial standards take on the character of statistical procedures … but, in all cases, statements about whether the compendial standard is met apply only to the units being tested” (8). Optimal production calls for meaningful in-process tests and for a statistically significant number of tests based on batch size of final dosage forms, Ciurczak says; companies may be following traditional practices, yet failing to meet the requirements set by cGMPs. “Let’s say that a company tests 20 tablets out of two million. Even someone without a math background knows that’s not statistically significant,” Ciurczak says.  He uses military standards as a contrasting example. “If you are manufacturing more than a half million units, the US Department of Defense says that you must run 25,000 tests. If a pharma company had to run 25,000 HPLC [high-performance liquid chromatography] tests, it would cost way too much and it wouldn’t show what was happening in the other 1.75 million.” Finally, there are the tests themselves, and whether they make sense given today’s manufacturing realities. “Friability and disintegration tests are not meaningful, and they are simply not timely anymore,” says Ciurczak. “They may have been timely when you had a single-punch machine that was putting out 100 tablets an hour. They are not timely when you have a 36- or 40-punch machine that is putting out 100,000 or 200,000 tablets an hour,” he says.

	FDA’s 2011 process validation guidance was supposed to mean the end of the “three batch” rule for product release but that concept is still alive and well at many companies, and sampling is still being handled in a rather primitive way at some facilities, says Ciurczak.

	The problem with adherence to this practice is the fact that “any time you take selective samples, you’re guessing,” Ciurczak explains. If one could place hidden cameras in many oral solid-dosage form facilities today, he says, one might often see operators taking material from the top of the last drum, scooping it out and calling it representative of the entire last batch, instead of sampling every drum at different levels using different sample thieves.

	One obvious solution would be to use PAT, but many manufacturers still aren’t, Ciurczak says. “With PAT, you have in-process tests and the potential of 100% analysis. There are online options that can automatically examine 100,000 tablets an hour, and there is zero excuse for not using them.” Ciurczak also sees a need for cGMPs to specify better calibration methods. “GMPs are based on 1950s technology, and they don’t stress raw material quality enough,” he says, at a time when most materials are being sourced from regions where regulatory practices are still evolving. For instance, cGMP practice still accepts the use of compendial testing, even though some older United States Pharmacopeia (USP) tests have proven to be fallible, as was seen during the heparin recall of 2007. “Spot and color tests for heavy metals might have been okay when pharmacists were making things in their back rooms and selling them locally, but now you’re sourcing materials from halfway across the world and the identity tests won’t show whether they are even useable,” says Ciurczak.

	The biggest problem with cGMPs, Ciurczak says, is that they are “one size fits all.” FDA, however, cannot mandate the approaches that companies use to meet requirements. “Measurements, process release, and in-process measurements are not mandatory. PAT methods may reduce variability in commercial product, but how the manufacturer achieves required targets is up to them,” says Ritchie. However, he notes, when something tragic happens, it’s usually lawyers who point out that management knew that other options were available but chose not to use them. This may prove to be grounds for future liability and accusations of negligence, he suggests.

	Some believe the problem is much bigger than cGMPs alone, and requires that drug development and manufacturing, and the chemistry, manufacturing, and controls, and manufacturing functions become more closely connected. “Drug development must undergo wholesale change as product development did in other exemplary companies in other industries in the 1960s,” says Rees. “These companies placed the end user of their products at the center and developed a value chain to deliver on their needs in terms of value for money. They achieved incredible improvements in quality by placing the responsibility for defects on the production operators, rather than a ‘quality function.’ The production operators were given the tools of statistical process control and they used them to great effect, achieving six-sigma quality levels (i.e., 3.4 defects per million opportunities),” he says. Amgen is one of a few biopharmaceutical and pharmaceutical manufacturers that has achieved this level of quality.

	At the same time, Rees explains, forward-thinking manufacturers in other industries began to design for manufacture, as opposed to throwing processes “over the wall.” They adopted a collaborative approach, he says, in which design and production departments worked together to achieve the crucial balance between innovation and production feasibility and economics. “The results were transformational,” says Rees, who has commented on some of these issues in his new book, T

, published in May 2019 (9). Rees sees the answer as going back to a simpler time, and to vertical integration. But one might ask: How can decades of industry change be reversed? For now, it remains to be seen if and how regulators will address the need to clarify industry’s interpretation of cGMPs. Hussain suggests breaking the problem down into smaller bites. “The cGMPs for nutritional supplements, compounding, and tobacco are still evolving. Perhaps the first order of business is to bring some sanity to the definition of adequate education, training, and experience, since the entire system rests on this foundation,” he says.

 83 (25), (June 20, 2005).
	4. K. Stone, Current Good Manufacturing Practices, 

Pharmaceutical cGMPS for the Twenty-First Century: A Risk-Based Approach

Guidance for Industry, PAT-A Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance

,  (Rockville, MD, September 2004).
	7. FDA, 

Guidance for Industry, Process Validation: General Principles and Practices

 (Rockville, MD, January 2011).
	8. G. McNally, “

A Practical Approach to Effective Life Cycle Implementation of Systems and Processes for Pharmaceutical Manufacturing

,” presentation at the Pharmaceutical Quality Systems (ICH Q10) Conference, Arlington, VA, Oct. 4–6, 2011.
	9. H. Rees, 


	Vol. 43, No. 7
	July 2019
	Pages: 16–20

	When referring to this article, please cite it as A. Shanley, “Is it Time to Update cGMPs?" 

Industry practice has changed radically over the past five decades. Can laws published in the 1960s still ensure pharmaceutical quality and safety today?

Is it Time to Update cGMPs?

	On May 6 2019, at St. John’s University in New York City, the 11

annual Charles Jarowski Symposium in Industrial Pharmacy offered a realistic look at the opportunities and challenges posed by continuous pharmaceutical manufacturing. More than 20 products that are now awaiting FDA review involve continuous manufacturing, but there are few compelling economic reasons for generic pharmaceutical manufacturers to invest in the technology, noted Atul Dubey, director of pharmaceutical continuous manufacturing at the United States Pharmacopeia (USP).

	Proponents of continuous manufacturing, and the agility and savings that it might enable, view this as a problem. Currently, 97% of the prescriptions filled in the US today involve generic pharmaceuticals, noted Ajaz Hussain, consultant and director of the National Institute for Pharmaceutical Technology and Education (NIPTE), a consortium of universities dedicated to advancing the concepts of modern pharmaceutical development and manufacturing. Within the past few years, long-standing NIPTE members such as Purdue and Rutgers Universities and the University of Puerto Rico (Mayaguez) have been joined by St. John’s and Long Island Universities, both based in New York City.

	During his plenary lecture, Hussain, who headed FDA’s Office of Pharmaceutical Sciences (OPS) and the agency’s process analytical technology (PAT) team in the early 2000s, and later led Sandoz’s development of the first biosimilar to be submitted for approval by FDA, shared some of his experiences. As he said, continuous manufacturing could offer a solution to improving supply chain reliability, and help prevent shortages of medications. 

	Taking note of the Big Pharma companies such as Novartis that have been working on continuous processes for years, Hussain said that the generic pharmaceuticals market is too fragmented for anyone to expect it to take this approach. “If the original product was not continuously manufactured, there is no incentive for generics companies to make the generic continuously. The initial investment for solids is significant,” he said, also acknowledging the need for expertise and technology for data manipulation and archiving as well as process analytical technology (PAT).

	External manufacturing hubs, new prior knowledge

	Hussain predicts that a new business model will develop, in which external manufacturing hubs based at universities, such as the Engineering Research Center for Structured Organic Particulate Systems (C-SOPS) at Rutgers University, and contract development and manufacturing organizations (CDMOs) will handle continuous manufacturing projects for industry, including generics firms.

	He also emphasized the need for pharmaceutical “new prior knowledge (NPK),” or critical information on quality issues and potential failure modes to be made available to generic pharmaceutical manufacturers, a concept advanced in an 

	Currently, generics manufacturers gain knowledge of the off-patent branded pharma molecule, but free access to NPK would allow them to connect the dots between drug substance characteristics and physico-chemical properties, product performance, therapeutic equivalence, and manufacturability. Advocates believe that access to NPK would drive faster development and regulatory review of generics and allow generics firms to sell more products. Currently, FDA data show that over half of approved generics are either never marketed, are sold intermittently, or reach the market only after significant delay, while abbreviated new drug applications (ANDAs) are not submitted for 10% of approved drugs (1).

	Several speakers on the program emphasized the importance of feedforward and feedback control, which, Hussain noted, will be crucial in the commercialization of more advanced therapies.  In fact, pharmaceutical quality by design (QbD) is becoming much more dynamic than what had been outlined in the first FDA guidance on that topic, and is evolving into “quality by control.”

	As pharmaceutical manufacturers become more comfortable with more advanced process control, the focus is shifting from the design space and locking in control strategy for the process during development, to greater use of feedback control so that any variation in input won’t result in a change in output.  The idea, according to Zoltan Nagy, a professor at Purdue University’s School of Engineering, is to make critical quality attributes tunable so that the system can find the conditions at which it needs to operate in order to ensure final product quality. “Focus on final critical quality attributes and let anything else change if it has to change, to leave CQAs constant,” said Nagy, whose presentation focused on continuous crystallization and use of feedback and feedforward control.

	“Unit operations must be considered as a system. In order to manipulate them, you need an array of continuous monitoring systems, but the real power is if we use this information for feedback control, and allow the system to change by manipulating critical process parameters to control critical quality attributes.”

	As Nagy noted, PAT tools have been used for 15–20 years to improve process understanding but until fairly recently they had seldom been used in feedback controls. Software is now being developed, he said, to take these tools and use redundant information for data reconciliation to inferentially control CQAs that may not be directly measurable and maintain them at the desired value.

	As he noted, these approaches can be used whether the process is continuous, batch, or mixed. “A lot of continuous manufacturing platforms are really running on batch processes,” he noted.  AstraZeneca demonstrated how active feedback control might improve crystallization, Nagy said, using measurements to change the temperature profiles within the crystallizers to adjust temperature cycles and optimize the process.

	Representing Big Pharma, experts from two early adopters of continuous manufacturing approaches shared their plans and some of their experiences. Takeda Pharmaceuticals, which acquired Shire Pharma’s continuous manufacturing expertise when it bought the company, has continued continuous research and piloting efforts begun by Shire in 2016, according to Jim Bonner, Takeda’s director of small-molecule drug product, who spoke in the program (2).  The company, along with Thermo Fisher’s Patheon, is exploring the use of continuous manufacturing for encapsulation as well as tableting, focusing on the continuous direct encapsulation of Vyvanse (lisdexamfetamine dimesylate), a treatment for attention deficit/hyperactivity disorder that was approved in 2007. Configuring encapsulation equipment for continuous manufacturing has been challenging, he noted.

	The company has completed trial runs using near-infrared (NIR) and Raman spectroscopy, which involved scanning 30 different blends three times, then using product of least squares (PLS)-based chemometric methods on reference gravimetric rather than high-performance liquid chromatography (HPLC) assay data. Extensive design of experiments work has also been done, and the company is anticipating approval of the continuous process by 2020.  As Bonner pointed out, these efforts have underscored the need for more robust and economic PAT options and simplified model maintenance. In addition, he believes that FDA should be shown that loss-in-weight feedback data are more accurate than data provided by NIR models. 

	Andrew Farrington, principal scientist for oral formulation sciences at Merck in West Point, PA, discussed the company’s pilot project, designed to build upon existing expertise in real-time release testing, raw materials monitoring, and QbD.  The pilot applies continuous-to-direct compression and film coating using tablet transmission NIR and tablet weighing instead of content uniformity, and hardness and disintegration instead of dissolution. The goals are to produce nearly 1 billion tablets per year with less than 90-day lead times from formulation to patient in a two-floor installation that is roughly one-third the size of traditional tableting facilities. 

	The company engaged early in discussions with regulators, and is now working on both a continuous product development line, capable of manufacturing 5–20 kg/h of product, and a commercial line, with capacity of 10–90 kg/h. The advantages of the small-scale unit, he said, included a single point of control, facilitating remote equipment operation and potentially allowing operators to be removed from the processing environment.  The company has developed a predictive model based on material attributes for continuous feeding and has also tested twin screw granulation.  Although the company is optimistic about the role that continuous processes will play in the future, Farrington asserted that a mix of batch and continuous will be better for the industry’s assets and supply chains.

	Scale-up issues: it’s not all about equipment

	Focusing on scale-up issues was Michael Rooney, director of process engineering at Genesis Engineers, who considered some of the challenges that facilities face when using a continuous manufacturing approach to scale up oral solid dosage forms.  As he explained, one cannot simply inject continuous manufacturing into a batch facility. He used two case studies to illustrate the challenges of justifying the cost to convert to continuous. One involved the expansion of a commercial drug using a dry granulation process in an established facility, the other, expansion of a high volume over-the-counter (OTC) product that did not involve API. In the latter case, he said, reducing labor costs was the primary goal.

	In both cases, he said, using the continuous manufacturing scenario presented a conflict between the flows of people and those of raw materials. “What began as a capacity question ended up as a question of return on investment,” he said. For one thing, with the branded drug company, it was found that the building, which was only 25 feet high, would need to be 60 feet high to accommodate continuous manufacturing. In addition, the shift from batch manufacturing would involve higher operating costs to employ highly skilled technicians who would attend to the PAT technology. The company, which had been trying to justify moving to continuous manufacturing on the back of one major product, is now working on building a continuous platform. “If you develop a platform for groups of product types, rather than try to replace existing batch capacity with continuous, it doesn’t have to run 100%. You can build a platform and a portfolio over time,” Rooney said.

	In the second case, with the OTC product, second-level infrastructure costs hurt financials and there was no subject matter expert on site to develop the PAT required for continuous, Rooney said. In this facility’s case, moving to continuous had no real impact on product cost and the company wound up investing more in its batch process to avoid undue risk. For a more extensive interview with Rooney, please see the June 2019 edition of 

	USP’s Atul Dubey, a Rutgers graduate whose new job title is one year old, discussed broad acceptance issues. “We need to focus on practical challenges of adoption, especially in the generics world,” he said. Starting manufacturing six months ahead of schedule can save a company a billion dollars, but in the end, investment is relative. “The switchover costs for a generic manufacturer are considerable,” he said, and the challenges entail investment and workforce skills.

	The increase in popularity of continuous processes poses challenges for generics companies. “Some may ask ‘If I can use my batch facility, can I meet my continuous product profile?’,” he said. They are concerned about being shut out of the market for these drugs because they will have difficulty making a product that has been approved for use with continuous manufacturing. USP is starting the spade work on standardizing methodologies to help the industry consider continuous processes, including new standard materials and PAT strategy standards, and is also developing a five-day course with Rutgers’ C-SOPs.

	“Continuous may not be USP’s traditional space, but we have to move forward,” Dubey noted.  In the end, he says, there is a need to make it easier and faster for late adopters to shift to continuous. “USP is interested in promoting quality of meds, however they are made…but generics manufacturers need to be able to produce drugs that were approved as continuously manufactured products via batch manufacturing, because the reality is that not all manufacturers will embrace continuous manufacturing,” he said.

	End-to-end continuous: a holy grail or achievable goal?

	Bayan Takizawa, cofounder and chief business officeratContinuus Pharmaceuticals, highlighted pilots and projects underway to help advance end-to-end continuous manufacturing, from API to finished product. There will be a need for deep process understanding, i.e., understanding CPPs, how to identify and monitor them reliably and consistently, and connect them to feedforward and feedback control loops. “A lot of upfront work is required,” he said.

	He also discussed the company’s pilot plant, which is currently operating with a throughput of 25,000 kg/year, using a five-stage continuous reactor with PAT (an infra-red system to monitor cake height), and recovering and reusing solvents in a closed loop.

	As he noted, industry inertia is a barrier to adoption.  He also suggested that equipment vendors need to embrace novel technologies. “Companies balk out of fear that the regulators won’t get new technologies, but FDA has been very supportive of continuous manufacturing,” he said, noting the fact that the agency’s Emerging Technologies Team (ETT) has advocated greater use of continuous.  In addition, he says, the Japanese Pharmaceuticals and Medical Devices Agency (PDMA) has recently set up an organization analogous to the ETT in Japan.

	Continuous filtration work has shown good results so far, said Takizawa. He also pointed to work underway for a large-volume generic drug company in India, which found that using continuous could reduce overall costs by 30–35%; cut the number of unit operations by 80–85%; reduce solvent requirements by more than 60% and footprint by about 90%; and contract product lead time from months to less than 48 hours.

	However, he noted, continuous is not a “slam dunk.”  “Staff expertise is an important issue.  Without strong technical support to handle development, data manipulation, and failure modes and effects analysis, you won’t have database to drive PAT. In addition, continuous would have an impact on existing facilities and traditional work and product flows,” he noted.

	On the academic side, University of Connecticut Professor Diane Burgess described work underway at the University of Connecticut evaluating continuous manufacturing for complex parenterals, in this case, liposomal formulations. The project began in 2013, but the miniaturized equipment was completed around 2018, going from the size of a standard lab bench to that a small refrigerator. Testing involved 11 liposome products, including generic formulations.

	Zoltan Nagy summarized work at Purdue using plug flow crystallizers and oscillatory flow using stirred tanks. “A large amount of fine crystals can create filtration problems, but if we allow spatial cycles, nucleation occurs in the middle of the reactor, and no fines result in the product.” Purdue’s team implemented a particle flow code (PFC) model with encrust formation that constantly de-fouls the system, yielding product with desired quality attributes. “Instead of having a steady state you have periodic steady state and periodically transfer several sections following periodic cycles in a phase diagram. It’s not steady state but it is a controlled state in which amplitude of oscillation is within allowable levels given CQAs.  A crystal size distribution (CSD) feedback controller detects the upper and lower bounds and triggers the point of crystal collection,” he said. Using a traditional control approach in this situation would lead to fouling, he added.

	In short, the program provided a look at the work underway to make continuous manufacturing a reality in more pharmaceutical operations, and how dynamic control will contribute to better batch and continuous processes in the future. 

	1. A. Hussain, K. Morris, V. Gurvich, “

Pharmaceutical NPK: Twenty-First Century Assurance of Therapeutic Equivalence

, April 2019.
	2. J. Bonner, “Continuous Manufacturing Challenges to Commercialization” a presentation made at the 11

annual Charles I. Jarowski Symposium in Industrial Pharmacy,” May 6, 2019.
	3. A. Farrington, “Challenges with Continuous Manufacturing: Integrated Development Strategy for Continuous Manufacturing of Oral Drug Products,” a presentation made at the 11

annual Charles I. Jarowski Symposium in Industrial Pharmacy,” May 6, 2019.
	4. A. Shanley, 

Although continuous approaches offer significant benefits for some products and processes, batch manufacturing is not going away any time soon.