The Impact of Artificial Intelligence on CSV
AI – Advance or darkest Threat?
Artificial Intelligence (AI) has the potential to become humanity’s most significant advance or darkest threat. As we develop our relationship with AI, it is likely to reveal itself as both an attractive and sinister character.
AI already has many friends including Big Data, Cloud computing, The Internet of Things (IoT), Advanced Robotics, 3D printing, and us, its human creators. AI will soon gain more influence over our lives.
Sadly, like the internet, and other technological achievements throughout the history of mankind, AI also will have the ability to serve the powerful but enslave the powerless.
AI will dominate Manufacturing Processes
AI will soon be embraced by most industries as it starts to deliver real-life business benefits. For example, reduced employee costs when a “smart” approach to manufacturing is implemented with virtual factories controlled by one or two human operators. These virtual factories obviously have the potential to replace the physical presence of many operators performing routine, repetitive tasks in the manufacturing facility.
Therefore, we should not be surprised when we see the tangible effects of AI upon the healthcare and life science industries. One effect will be a shift away from producing medicines with fixed specifications to an AI driven real-time monitoring, simulation, integration and predictive analysis of manufacturing processes.
Neither should we be surprised if we witness a shift from the conventional use of computerized systems for the physical manufacture of curative medicines to a more digitized, personalized and preventive approach to healthcare.
Predictive analysis is currently available in validated, commercial software tools used for ongoing process verification and it seems that some more mature AI tools may have made that “personalized” shift already. But now, we are entering the unvalidated realm of cognitive assessment, where AI is programmed not to rely solely on statistical projections but also to mimic human thought and decision-making.
AI will disrupt Ethics, Legalities and Regulation
The study of AI ethics, legalities and regulation is complex and fascinating; and the increasing acceptance of AI applications in the life science industries is offering challenges and opportunities to experts in medical ethics, lawyers and regulators alike. Some expressions of the ethical, legal and regulatory issues associated with the use of AI are already familiar to us with the advance of Personalized Medicines.
Who will be responsible for AI-associated Treatment Events?
There is an interesting example of a start-up company that is pioneering the use of AI to design personalized cancer-treatment regimes. The company’s technology mimics cell biology at a molecular level and seeks to identify the best drugs to use for specific tumours. This AI-enhanced technology searches for combination therapies by performing millions of simulated experiments each day. A leading pharmaceutical company has already awarded the start-up a grant to continue its research.
Using this example, there are open questions regarding:
- The medical ethics, or the set of moral principles that professionals can refer to, if following the technology`s regulatory approval, there are treatment-associated adverse events and patient mortalities.
- The legal position as to with whom the liability lies if there is no objective evidence how machine learning allowed algorithms to choose a catastrophic drug combination.
- The compliance expectations of regulators regarding the lifecycle management of computerized systems with neural networks and machine perception.
AI Governance – Setting the Standard
We are still waiting for the legal, international framework for AI-governance to be issued even though a new healthcare strategy for AI governance is urgently required. For sure, current CSV regulations will have to change.
The FDA is only now “actively developing a new regulatory framework to promote innovation” in the AI space even though at least fourteen AI-embedded medical devices have been approved since January 2017. Also, ISO, the International Organization for Standardization, currently has no approved specific AI-related standards and has only recently started to develop seven standards relating to the management of AI systems.
The AI compliance context could become radically and dangerously impacted. Without enhanced regulation and standards, auditors who assess the compliance of AI vendors could easily be faced with increasing intellectual property restrictions on accessibility to the vendor’s information and data. This would mean that their audit evidence becomes less scientifically-sound and meaningful and only an elite group of experts maintain a technical understanding of how the mysterious AI actually works.
A New Era for Computerized System Validation
It is likely that the world of CSV will be dramatically disrupted as “Industry 4.0” is complemented by “Pharma 4.0” and “Quality 4.0”. Life science companies with strategic plans to survive and thrive in the “Industry 4.0” revolution will no longer be able to wait for the approval of manually-written validation reports. The era of waiting days or weeks to approve documents before companies can put their AI application to real-life business benefit is over.
More emphasis in CSV will probably be placed on the “holistic value network” – where a complex computerized system will be treated as a single entity. There will be a focus on the AI-supported business processes (for example, purchasing, software development and complaints handling processes) rather than individual elements such as technical documents, workflows and people.
AI will not wait for Traditional CSV Methodology
It is already acknowledged that current industry norms such as a waterfall approach and GAMP 5 guidance sometimes struggle to keep pace with the rate of software development and release management; we are on the brink of a new era of CSV methodology with AI.
Even the increasingly popular “Agile” approach is likely to be aided by the use of automated testing tools. “Sprints” will have to assume a more nimble, real-time and predictive emphasis. The era of prospective validation using static data in a test environment may be drawing to a close; an intense “real-time” approach in a dynamic, productive environment may be necessary.
AI will Transform CSV Test Strategies
CSV test strategies are likely to be transformed. For example, the use of process and functional risk assessments in determining the scope and type of CSV testing required will probably become a very challenging exercise. It will not be easy to obtain evidence from a tool that is constantly evolving and mimicking human decisions. How will a human CSV professional quickly determine if a challenge test, stress test or penetration test is the most compliant approach?
The use of automated verification tools, including AI’s “self-verification” scripts is likely to soar. These tools and scripts will be needed when validating systems programmed to reveal patterns, trends, and associations within extremely large data sets.
The CSV test approach will have to become more focused upon:
- Data governance including data confidentiality, integrity, availability and accountability.
- Data integration and traceability particularly when systems are processing millions of key-value operations per second.
- Real-time, dynamic and predictive simulation testing of the business processes the computerized system is intended to support.
The era of manual operational qualification testing may also be drawing to a close.
How will we Manage Test Defects, Bug Fixing and Software Releases?
The management of test defects and failed acceptance criteria could become a mine field; systems will be programmed to learn from their errors and instantaneously correct them. The line of demarcation between defect management and corrective and preventive action management may become confusingly blurred.
The nature of bug fixing and software releases will be impacted as AI will not wait for traditional GAMP 5 style system lifecycle release notes. Another aspect of the impact of AI on CSV will probably include retirement management – how will we verify AI has “forgotten” that certain actions are no longer required and erased from a virtual memory?
It is almost certain that CSV compliance, methodologies, test strategies, corrective actions and releases will not resemble the traditional approach we are currently familiar with. This will also necessitate a fundamental and intense change in the role of the CSV professional. We must be prepared for the future – including a willingness to verify our CSV professional credentials when we are asked by the AI system we are attempting to validate to confirm “who we are” by passing a test that we are not a robot.
Impact of AI on CSV Professionals
CSV experts, like many other professionals in the life science industries, may be contemplating the lack of visibility of the future with many emotions ranging from fear to excitement. But CSV professionals can take some actions today to be prepared for tomorrow:
- Engaging in continuing professional development to become familiar with the technologies of Industry 4.0 and Pharma 4.0; and a readiness to engage with Quality 4.0.
- Ensuring our relevant knowledge base is up-to-date with developments in legislation, regulations and standards which will impact the compliance context of CSV.
- Establishing strong collaboration with technical experts in AI and associated technologies so that CSV test strategies are accurate and meaningful.
- Devising relevant and standardized CSV methodologies and test strategies which include the use of automated verification tools and simulated, real-time and predictive testing.
Facing the Future
As we face the future together with AI, we, as CSV professionals, now have an opportunity to direct and influence the regulators. We can devise a scrupulous CSV methodology based on dynamic and objective evidence. But, as ever, we must be guided by the ethical and timeless principles of patient safety, product quality and data integrity.
An Article by HGP Author Sharon Shutler (Director, Quality Management Services)