*Scroll down to see the Draft AI Project Security and Privacy Measures* 

On December 19, 2024, PAAB hosted a townhall with over 100 representatives across the pharmaceutical industry to share their initial thoughts on PAAB’s approach to the integration of AI tools on our client-facing submissions platform. The overview slides of PAAB’s phased approach are provided in the attached document.

 

Phase 1 Overview
Work completed to date in phase 1 has exclusively leveraged PAAB’s internal knowledge base, ensuring that no current or past submission content has been utilized. This approach reflects our commitment to confidentiality and data security.

 

Looking Ahead: Future Phases
In future phases (i.e., phase 2 and beyond), the model could potentially benefit from actual examples/submissions. Please note that PAAB’s models will use actual submissions only with explicit opt-in consent from senior staff at the corresponding manufacturer who are authorized to provide such consent.

If you would prefer not to wait for us to contact you, we invite you to reach out to us at info@paab.ca to schedule an initial debrief meeting and learn more about the opt-in process.

 

Opt-In Options
Manufacturers who do not opt in can rest assured that they will still benefit from automation derived from PAAB’s proprietary knowledge base. Those who choose to opt in will gain access to enhanced automation. We plan to offer the following opt-in levels:

1. Option A: Allow the contribution of your de-identified submission data to a broader data set comprised of data from all those who opt in to Option A. By contributing to a broader dataset, this option will unlock access to more advanced automation features and insights, as the model can leverage a richer pool of data to develop nuanced contextual understanding and improved accuracy of compliance.  
2. Option B: Allow the use of your submissions solely for the assessment of your own ads. This option will leverage the PAAB internal knowledgebase and submissions from across your organization alone. While it will provide some efficiencies, it will be limited by a smaller data set.

NEW: Draft AI Project Security and Privacy Measures

1. Project scope

All aspects of all submission types will eventually be incorporated into this initiative.  While we are beginning with specific areas/topics (e.g., APS format issues, patient information, specific segments of HCP review), the scope will gradually grow to encompass the foundational aspects of the pre-clearance regulatory review.

2. Timeline

We hope to have some client-facing aspects of the project available to clients who elected to participate near the mid-point of the current calendar year. We aim to automate all aspects of the review through a 4-phase journey intended to be completed within 2 years.

3. How the submissions are utilized

The submissions would be used in a way that mirrors how they are used when humans perform and review and when humans are trained to review. i.e.,

• Entire APS is used as reference material for the model during review: Those historic submissions identified by the submitter as containing previously approved claims are crawled by the model for awareness of which content had been previously approved.
• Segments of APS (potentially anonymized) are used for model training:  Segments of submissions are stored in a secured database (see below) in association with the relevant segment of internal guidance documents that outline why the APS segment was questioned or approved (i.e., as examples showcasing real-world application of the internal guidance).

Importantly, the model is designed to identify perceived contraventions of the regulations for human PAAB reviewer assessment  (i.e., at no point will a manufacturer be exposed to another manufacturer’s submissions).

4. Data security & Risk Management

A firewall surrounds all of the infrastructure components of eFiles. Additionally, all traffic is logged. It is also important to note that the model, its corresponding databases and file storage are housed within that same firewalled infrastructure (i.e., the use of the data by the model does not introduce the data into a new environment that it has not already resided in).

The data is backed up in different availability zones (as is best practice). But, these availability zones are also within the same firewalled environment. 

All data is encrypted in transit (SSL) to the web application server.  The data is encrypted at rest on the database server and in file storage.  There is no public access to the backend database server or the file storage.

In addition to the security design described above, we conduct targeted penetration tests on all apps that we host including the eFiles system. 

Be among the first to benefit Learn more! Reach out to us at info@paab.ca to schedule an initial debrief meeting and learn more about the opt-in process. We encourage those interested to reach out early to ensure that your organization is among the first to be included in this initiative and gain access to enhanced automation as soon as it becomes available.

 

For more information on the Terms of Use relating to use of PAAB assets such as PAAB correspondence, guidance documents, advisories, training materials and Q&A in the training of AI models and related purposes, please click here.

 

Thank you to all who participated in the townhall for your thoughtful questions and valuable insights.