SeerPharma engaged with students from The University of Melbourne's Master of Biotechnology program to thoroughly review the academic literature and hold discussions with industry and regulators to develop a comprehensive report investigating the impact of Large Language Models (LLMs) on Good Manufacturing Practice (GMP) and Quality in the biotechnology sector.
This report investigates the transformative potential of Large Language Models (LLMs) on quality management systems (QMS) within the biotechnology, MedTech, and pharmaceutical industries. The study explores the multifaceted impacts that LLMs can have on various subsystems of the QMS and assesses their implications for industry professionals. The primary focus is on exploring current challenges facing the industry and the potential use of LLMs to solve these. The report also investigated the risks of integrating LLMs and provides a risk assessment and SWOT analysis to aid eventual decision-making for companies looking to use LLMs.
Summary of Findings
The study found that the overall benefits of LLMs lie in optimising processes. LLMs have the potential to improve task efficiency, provide additional insights for data analysis, and assist with more complex activities. Future uses of LLMs can extend into more advanced applications such as predictive maintenance. However, factors such as risk and current acceptance of LLMs suggest that the tool may not be as beneficial for external auditing and customer complaints compared to the other subsystems of the QMS explored.
Table 1: Summary of General Findings
| Validation / Data Integrity | • Incorrect outputs are a significant deterrent to the integration of LLMs in regulated sectors. • Human review is proposed to minimize the risk of using incorrect information generated by LLMs. • Regulatory measures, similar to those in medical device guidance, may be necessary to assess the accuracy and precision of LLMs. • Continuous monitoring and preventive measures are essential as LLMs can change over time, impacting their performance and raising liability concerns. |
| Regulations | • The prevailing attitude towards current regulations is uncertain, especially regarding the governance of AI tools in Australia. • Transparency between organisations and regulators was emphasized. • Regulations in Europe and the United States suggest a growing trend of regulations pertaining to specific use cases, in addition to a general AI act. • The importance of preparing for the introduction of AI policies based on global regulatory trends was highlighted. |
| Risk Assessment | • LLMs have ongoing risks and need potential mitigation strategies to minimise these risks. • LLM training and testing data needs to be continually verified and validated. |
Table 2: Summary of Specific Findings for Each QMS Subsystem
| Auditing | • Current auditing challenges include: technical competence, resource limitations, the time-consuming nature, and a lack of awareness about audit and audit finding significance. • LLMs can optimise risk-based audit schedules, enhancing preparation and management. • LLMs can educate auditors on standards, providing in-depth knowledge and overcoming rotational audit roster limitations. • LLMs support on-site audits with checklists, automated reports, and facilitating discussions with operators. • LLMs may evaluate inspection readiness and improve external auditing and regulatory inspections. |
| Customer Complaints | • LLMs can streamline the handling of customer complaints, aiding in data analysis to identify trends and suggest solutions. • LLMs can assist in identifying, categorizing, and flagging complaints, automating communication and improving efficiency. • LLMs, especially in the form of chatbots, can guide customers through the complaint process, minimizing the need for human checks. • Human review is sll currently deemed necessary for complaints despite the potential for LLM automation. • The cost-benefit analysis may limit LLM use, especially for companies with few annual complaints or specialized product markets. |
| Calibration & Maintenance | • Most companies currently rely on supplier advice for equipment calibration scheduling. • Manual processes with paper-based data recording and subsequent digital transfer pose challenges. • LLMs can make schedules data-driven and risk-based, analysing historical trends and suggesting changes for optimised efficiency. • LLMs can analyse multivariate time series sensor data for real-me monitoring and machine-learning algorithms to identify abnormalities and degradation patterns. • LLM can also enables predictive maintenance, minimizing costs associated with frequent preventive maintenance. |
| CAPA | • Several CAPA challenges currently exist: tight timeframes, engaging individuals due to competing priorities, issues locating desired documents, filling out CAPA forms and composing actions due to the subjective nature and inconsistencies. |
Submit the form below to access the full report Impacts of Large Language Models in Quality Management Systems within Biotechnology Industries.
Initiatives such as this are just another way we look to “Advance Quality and GMP Best-Practices”.
