Automated Classification of Post-Approval Changes Using Regulatory Intelligence Models

Post-Approval Changes (PACs) further introduce an immense regulatory burden and sometimes result in long approval times and conflicting categorizations across different countries. We processed 7,500 PACs with 47.7% minor, 48.7% moderate, and 3.7% major changes. For this study, the machine learning algorithms of Gradient Boosting, Random Forest, SVM, and Logistic Regression were trained on 16 regulatory features varying from clinical impact, quality risk, and explanation of quality scores. Gradient Boosting was found to have the highest accuracy of 87.3% for classification, surpassing traditional classification methods and minimizing discrepancies in approval times. For quality scores with high justification, approval times were reduced by 23%, and major changes took an average of 287.5 days as opposed to 156.7 days for minor changes. Such automated platforms have the potential to save 60%–75% of time spent on manual classification methods and allow real-time API connectivity with regulatory intelligence systems. The automated classification results of PACs have a vast potential for scaling up the consistency of regulations, improving approval times, and laying the foundation for global harmonization in post-approval change management.

The machine learning-driven classification of post-approval changes has shown good predictive capability, and the results demonstrate the achievement of reductions in the regulatory approval time. This substantiates the feasibility of the incorporation of regulatory intelligence with the use of AI systems for the achievement of harmonization