Comment from CognifAI Solutions

Dear FDA or Regulatory Agency, We appreciate the opportunity to comment on the "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products; Draft Guidance for Industry and Other Interested Parties". Below, we provide specific comments in relation to CoVigilAI, a Generative AI Medical literature Monitoring in Pharmacovigilance, as an example of how AI tools are used within Post Marketing surveillance. Our AI enabled solution designed to enhance Medical Literature Monitoring in Pharmacovigilance - CoVigilAI. It has unique feature like: •Automated ICSR Segmentation - Valid, Invalid, Article of Interest & Potential ICSR •Human in Loop decision making •Automated Key Entity Detection •Auto processing of freely available full text •Causality Assessment •Generative AI based summary of Full-Text •Global Literature Monitoring •Compliant & Audit-Ready •100% Human in loop system 1. AI in Post-Marketing Surveillance •AI-Driven Literature Monitoring in Pharmacovigilance: oCoVigilAI utilizes AI models for real-time monitoring of safety signals post-marketing. By analysing large volumes of data from Online Literature Database, AI can help identify potential adverse events early. FDA Guidance :- The draft guidance mentions the role of AI in identifying safety concerns—clarification is needed on the threshold for regulatory action when AI models detect a potential safety signal. This would be followed by Human In Loop for Final expert decision. •Continuous Learning of AI Models in Post-Marketing Settings: oWe are committed to continuously updating our AI models with new data gathered in real-world settings to improve their predictive power. FDA Guidance :- We propose that regulatory frameworks support dynamic, ongoing validation of AI models post-market to reflect new data and real-world variations in patient safety area. 2. Transparency, Explainability, and Accountability of AI Models •Model Transparency and Interpretability: oCoVigilAI's AI models provide transparent, interpretable outputs to ensure clarity on how decisions regarding patient selection and treatment protocols are made. FDA Guidance :- While AI tools can be complex, we request the FDA to outline the level of explainability required for AI-driven decisions, particularly in cases where models are trained on proprietary datasets. •Accountability for AI-Generated Decisions: oWe fully support the principle that sponsors must maintain accountability for AI-driven decisions, especially in areas such as patient safety and efficacy. FDA Guidance :- It would be helpful if the FDA could provide a framework for ensuring accountability when AI tools are used in regulatory submissions, such as detailing who is responsible when AI outputs lead to incorrect conclusions or actions. •Human Oversight and Regulatory Audits: oWe propose that regulatory oversight ensures that human experts review AI-generated recommendations, especially when decisions have significant implications for patient health. FDA Guidance :- Clarification on how FDA will conduct audits of AI-driven decision-making processes will be valuable. 3. Data Quality, Bias Mitigation, and Equity Considerations •Ensuring High-Quality, Representative Data: oFor CoVigilAI, ensuring that AI models are trained on high-quality and representative data is critical. We have put in place measures to ensure diversity in training datasets. FDA Guidance :- We urge the FDA to provide clear guidance on how data sets should be curated to avoid biases, including specific methods to assess representativeness across key patient subgroups (e.g., racial, ethnic, and gender diversity). •Bias Mitigation in AI Algorithms: oAs part of our AI training pipeline, we have implemented de-biasing algorithms to identify and correct for any disparities. FDA Guidance :- We suggest the FDA develop a framework to assess and monitor the fairness of AI algorithms, including guidelines on how to report and mitigate any detected biases throughout the lifecycle of AI tools. 4. Validation and Regulatory Approval of AI Tools •AI Model Validation: oWe are committed to validating all AI tools used in the development and post-market surveillance of CoVigilAI. Validation includes ensuring that models are robust, reproducible, and yield accurate predictions FDA Guidance :- Further clarification is needed regarding the specific validation protocols FDA requires for AI models used in regulatory submissions. We would appreciate additional guidance on performance metrics, data inputs, and the role of external validation. Thank you for the opportunity to provide feedback. We are committed to ensuring that CoVigilAI meets the highest regulatory standards and ultimately benefits patients. Sincerely, Romesh Sheth Chief technology Officer CognifAI Solutions Pvt Ltd.