{"id":650,"date":"2024-12-23T21:58:00","date_gmt":"2024-12-23T20:58:00","guid":{"rendered":"https:\/\/www.darqa.org\/?p=650"},"modified":"2026-08-11T12:55:51","modified_gmt":"2026-08-11T10:55:51","slug":"report-darqa-glp-theme-day-29-november-2024","status":"publish","type":"post","link":"https:\/\/www.darqa.org\/index.php\/language\/en\/2024\/12\/23\/report-darqa-glp-theme-day-29-november-2024\/","title":{"rendered":"Report on the DARQA GLP Theme Day, 29 November 2024"},"content":{"rendered":"

Machine translation: this English text was automatically translated from the original Dutch article. If anything is unclear or differs, please refer to the Dutch original.<\/em><\/p>\n

Event language:<\/strong> This event was held in Dutch.<\/p>\n

From Good to Great \u2013 GLP, Quality Assurance and Quality Improvement Tools<\/strong> <\/p>\n

On Friday 29 November 2024, around 50 participants gathered at De Eenhoorn in Amersfoort for a GLP Theme Day about recent OECD GLP documents.<\/p>\n

Frans Brouwer, chair of the GLP Committee, set the scene with a roadmap that took us from the introduction of the OECD Principles of GLP in 1981 through several important milestones. These included the consensus document on the validation of computerised systems in 1995, which was updated as Advisory Document No. 17 in 2016. The consensus document on applying GLP to the management of multi-site studies, No. 13, appeared in 2002. In 2022, Advisory Document No. 23 on QA and GLP was published, together with Position Paper No. 24 on quality-improvement tools and GLP. The question was raised: where are we heading in the future, and what role will AI play in our daily lives and in life-sciences research?<\/p>\n

<\/p>\n

Erna Hissink, a GLP inspector at the Dutch GLP Monitoring Authority, the Health and Youth Care Inspectorate (IGJ), presented the key elements of Advisory Document No. 23. These included QA\u2019s responsibilities and the importance of QA having access to the Master Schedule. She said that Dutch inspectors will not routinely request access to QA reports, although some other countries may ask to see them. She explained that, even in a global company, Dutch inspectors expect QA to report to Test Facility Management at the local site. QA does not have to be an expert in every procedure, but it must understand the content of the relevant SOPs. It is also important to define QA\u2019s independence properly. QA must provide appropriate oversight of studies, and critical activities for inspection must be defined in the QA SOPs. During process-based inspections, it may not be possible to inspect certain studies. The frequency and rationale of process-based inspections must be included in the QA SOPs. The Test Facility Management function must also be covered as part of facility inspections. QA may have other roles too, such as training personnel in GLP, managing SOPs and inspecting third parties. Test Facility Management must be aware of QA activities and ensure that the QA programme is carried out as planned. It is up to QA to decide whether to implement a risk-based programme. Risks should be determined on the basis of past experience, inspections, changes in the organisation of the test facility, and so on. QA itself must also be assessed. This can be done by someone from another department or by an external QA professional.<\/p>\n

Mirjam Smeets, also a GLP inspector at the Dutch GLP Monitoring Authority, discussed OECD GLP Position Paper No. 24 on quality-improvement tools. She explained that this is a position paper and that the IGJ therefore cannot enforce it. Test Facility Management is responsible for improving the facility. A mechanism must be in place to collect problems and perform a Root Cause Analysis before deciding on corrective and preventive actions. The effectiveness of the actions taken must also be monitored. Input for improvement may come from identifying problems, non-compliances, deviations, risk assessments, system performance and monitoring, and feedback from employees. Internal and external audits and government inspections can also provide input for improvement. When conducting a Root Cause Analysis, it is essential to identify the true root causes.<\/p>\n

Many tools can be used, including the Five Whys, fishbone diagrams, Pareto charts (the 80\/20 rule) and FMEA.<\/p>\n

The three most common mistakes are jumping to conclusions, treating symptoms and assigning blame. Jumping to conclusions may provide a temporary fix, but it is usually not the long-term solution. When CAPA plans are drawn up, it is essential that the solutions genuinely address the Root Cause. Think of it as a cockpit dashboard<\/em>: when a warning light comes on, you need to know what the problem is. Bear in mind that there may be more than one problem.<\/p>\n

Maaike van Dooren gave a comprehensive overview of how Charles River Laboratories in Den Bosch implemented a risk-based process audit programme. The first step is to determine the risk class by considering the impact and likelihood of something going wrong in a study phase. The risk class is then assessed against the ability to detect the issue, producing a risk category. Finally, the risk category is considered in relation to the frequency with which the process is performed: monthly or less often, weekly or more than twice a month, or daily or more than twice a week.<\/p>\n

Maaike explained the following steps taken to establish a successful risk-based process audit programme: <\/p>\n

1. Understand the study type. <\/p>\n

2. Divide the study into processes. <\/p>\n

3. Classify which processes are routine and repetitive. <\/p>\n

4. Determine the risk class. <\/p>\n

5. Ask experts for assistance. <\/p>\n

6. Review the risk categories. <\/p>\n

7. Ask experts for assistance. <\/p>\n

8. Determine the process-audit frequency. <\/p>\n

9. Conduct an annual review and reassess every quarter. <\/p>\n

Frans Boeijen then gave a presentation on AI in Quality Land<\/em>. He began with the history of AI, going back to the first computer checkers programme in 1950. The first ELIZA chatbot was developed in 1965, and major advances in AI and machine learning followed in the 1990s. By 2000 we had smart toys, and in 2020 OpenAI was available on the internet. AI encompasses machine learning, which in turn includes deep learning. Reinforcement learning can take place through trial and error and through feedback on successes and failures. AI systems are not error-free and must be assessed carefully. Considerable progress has been made in research using data from wearable devices such as smart watches, rings and phones. It is essential to use good, reliable data for AI.<\/p>\n

The quality of the input determines the quality of the output.<\/p>\n

Some of the risks of AI and machine learning are: <\/p>\n