US FDA’s Knowledge-aided Assessment & Structured Application (KASA): revolutionizing drug product review using Artificial Intelligence (AI)?

(originally published on LinkedIn)

I have been trying to understand KASA and wanted to share my thoughts with you as it could potentially bring some major disruptions that will require skill and mindset change in the pharmaceutical manufacturing space. Personally it is a very interesting topic, but reading what’s out there, the scope of the required changes that will arrive with the implementation of KASA goes beyond QA. It will require management level decisions and will have company wide implications. The synopsis is presented below. Please feel free to add in through the comments section for the benefit of our pharma colleagues.

It seems US FDA wants to bring in big data and algorithms in a big way to automate a large part of the assessment process for new drug product/drug substance submissions/approvals. The aim, however, may not only be to reduce the number of expensive on-site inspections, but when required, enable inspectors to focus more efficiently on the risks and red flags generated by the KASA platform (which would have been fed by submissions for a particular drug product from originator (for NDAs), generic competitor’s submissions (for ANDAs), past inspection data of manufacturing facilities, and other historical data).

From a generic drug product’s perspective, the assessment can bring in a new rigour as the inherent risks can be pre-assessed by (let’s say) benchmarking a new eCTD submission or any other type of submission [mostly relating to Pharmaceutical Quality/Chemistry, Manufacturing, and Controls (PQ/CMC) made using a structured application that communicates with the KASA interface] for an ANDA against the originators and other generic manufacturers eCTDs (in not first-to-file instances), necessitating a pre-approval inspection when needed. FDA already undertook a project to identify and prioritise PQ/CMC information that would benefit from a structured submission approach. 

By extension of the methodology based on structured data and Machine Learning algorithms that would power the new assessment system, FDA should be able to arm its inspectors beforehand with a fair knowledge of the inherent strengths and weaknesses of a manufacturer, and also what to look for when on the ground (during an on-site inspection). On the other hand, when the new assessment system has been validated and is up and running, it will help remove ambiguity from the interpretation of the regulations or advise regulators in framing better regulations (where fresh outlook is needed).  

Also as the EMA and USFDA have put in place a mutual recognition agreement (MRA) on good manufacturing practice (GMP) inspections, it is quite possible that data on observations are shared between the two authorities in developing a more robust risk assessment system.  

I think we have to get ready for AI (Artificial Intelligence) powered decision making in drug approvals and drug safety assessments as well. 

However, what it means in the short term (maybe over the next 5 years) is a lot of chaos and additional work for companies while FDA is still trying to acquire structured data (as at the moment majority of the information is unstructured and free-form). So FDA will probably begin by starting enforcing more structured e-submissions of every type of information including attachments and all supplementary documentations with eCTDs.

So far eCTDs have been used for document metadata management, providing a structured way for the inspectors to access all documents as part of each application. However, the next steps would require to structure the content of documents itself so that the entire application can be queried and mined by algorithms (Natural Language Processing, decision tree based classification, etc.). Applications could see getting rejected purely on the basis on submissions not being in line with the format requirements. This will translate into higher QA costs (more manpower, training and resources will have to be dedicated, in addition to better time scheduling) over the next few years.  

Hope this helps in kickstarting some level of preparations to deal with the future of Regulatory Affairs.  

Leave a Reply

Your email address will not be published.