(T1330-11-63) Developing a Risk Score for Critical Suppliers Using Artificial Intelligence (AI)

Purpose: Pharmaceutical companies are increasingly relying on 3rd party manufacturing for both active pharmaceutical ingredients (APIs) and finished dosage form (FDF) products. Despite this trend towards outsourcing, attempts to improve risk-based management of 3rd parties have been hampered by a lack of clean regulatory data, disparate systems, and privacy concerns. Unique site IDs have been developed that are regulatory agency independent. Using data enrichment facilitated by AI and machine learning (ML) tools, site scores are calculated that represent the GMP health of individual manufacturing sites and are mapped to unique site IDs. The scores are reliable, accurate, explicable, and repeatable, and rank suppliers into categories ranging from very good to very poor based on regulatory and inspection performance. The resulting risk scores are useful for vendor due diligence, management of vendors, and inspection readiness.

Methods: Data extraction from Internet sites, repositories, and online databases and enrichment using Natural Language Processing (NLP) tools and other Artificial Intelligence (AI) and Machine Learning (ML) tools, developing ontologies and scoring systems, and using those systems to score the GMP health and compliance of manufacturing sites based on the extracted and transformed/enriched data.

Results: Thousands of pharma GMP manufacturers have been ranked into categories ranging from very good to very poor based on regulatory and inspection performance, with data that is reliable, accurate, explicable, and repeatable. Valid comparisons between companies can be accomplished using these results.

Conclusion: It is possible to unambiguously identify drug manufacturing sites and rank the potential risk inherent with each using data enrichment facilitated by AI and machine learning (ML) tools to calculate site scores that represent the GMP health of individual manufacturing sites. The scores are reliable, accurate, explicable, and repeatable, and rank suppliers into categories ranging from very good to very poor based on regulatory and inspection performance. The resulting risk scores are useful for vendor due diligence, management of vendors, and inspection readiness.


Data Analytics and Enrichment Model


Human Drug GMP Quality System Labels


Example Manufacturing Site Profile