ISPE 2024 ISPE Biotechnology Conference, June 18 , Extended Notes

ISPE Boston Area Chapter recently hosted the 2024 ISPE Biotechnology Conference on June 17th and 18th.

The synopsis below is an extension of a prior LinkedIn post, as well as my notes and experiences from the conference.*

💡Regulatory Challenges in Machine Learning Solutions 💡

Danilo Neri spoke about LLM use cases, including data integrity and data quality concerns surrounding the use of certainmodels. Overall, he reverberated concerns and controls that can be put in place around models. Things to think about here include mailing data quality and ensuring data integrity is maintained and used for the development of the model. Additionally, it touched on the notion that expectations on models have been released but not exact guidance or legislation within current regulatory frameworks. Something to look for will be detailed requirements on System Development Life Cycle documentation.

🛠️ AI in Document Generation for cGMP 🛠️

David O’Connell gave a case study in which AI was used for initial document generation, the nuances around this, and the efficiency and cost savings brought onto a specific project. He noted several statistics here, including a budget of $11.4M for 6,500 documents across 22 buildings. He noted a roughly 7–7.5% savings on the project(roughly $.71M). Overall, using AI for CQV implementation was able to reduce the strain of initial document drafts, and the production of documents was able to increase to around 100 per day, far more than the receiving party could process per day(noted at 15). We can note the return on such a system to be achieved in less than 6 months and reduced staff savings per project; overall efficiency was noted to be around 1–2%.

This granular presentation and breakdown were fantastic in understanding how far modern tech has taken us for document uploads and population.

It segments well into Peter Blennerhassett’s analytical approach to Unified Knowledge Systems(UKS), addressing how to find and use critical information efficiently. A case study was presented, in which he was able to reflect data from interviews about using a UKS in a technology transfer. Overall, the purpose of a UKS is to reflect a management system such as Building Information Modeling(BIM). A flagship system across industries, particularly in design construction and operation to manage projects. With so many documents for any particular project, moving towards unified systems and knowledge platforms makes the utmost sense. Additionally, as mentioned in the prior speech, AI will increasingly be able to automate and manage documents, leading to a powerful synergy of document management on the backend of a system, with a true framework for collaboration among parties on the front end.

Additionally, Ted Webb AProduct Development and Lifecycle Management (PDLM) Tools tools can better enable technology transfers and replace prior documentation methods. One strategy was implementing a Digital PDLM system along with the traditional tech transfer process. Following this, a Digital Product Profile(DPP) will be used to keep all advanced process controls and data. The benefit of early implementation here is your manufacturing environment will be digitally enabled and controlled, and the reliance on traditional and convoluted document processes can fail.

*Note: All notes, stats, and figures are a reflection of my notes from speakers and not a direct claim or reflection by the speaker.