
AI/ML in Regulatory Submissions:
Establishing Model Credibility for Decision-Making
Establish AI/ML model credibility for regulatory decision-making by connecting context of use, model risk, data relevance, performance, uncertainty, and supporting evidence.
Course Overview
How do you determine whether an AI/ML model is sufficiently reliable to support a regulatory decision?
This webinar examines how pharmaceutical and biologics developers can establish model credibility for a defined regulatory purpose. It connects context of use, model risk, data relevance, performance, uncertainty, and supporting scientific evidence to assess whether reliance on a model’s outputs is justified.
Participants will explore a risk-based approach to planning credibility assessments, evaluating evidence, identifying limitations, and documenting the rationale for using AI/ML outputs in regulatory submissions.
Establish model credibility
AI/ML models are increasingly being considered for nonclinical assessments, clinical development, and regulatory evidence generation. However, strong predictive performance alone does not demonstrate that a model is suitable for a particular regulatory use.
The evidence needed depends on the question being addressed, how much the model influences the decision, and the consequences of an incorrect conclusion. A model that performs well during development may still rely on unrepresentative data, insufficiently examined assumptions, or validation conditions that do not reflect its intended application.
This session examines how these weaknesses affect model credibility and how to build a scientifically justified assessment of whether the evidence supports the proposed use.
Determine Whether the Evidence Supports the Intended Use
Examine data relevance, model applicability, uncertainty, and performance limitations in relation to the model’s defined context of use. Understand the justification needed when model outputs contribute to consequential regulatory decisions.
Build a Risk-Based Credibility Assessment
Explore FDA’s proposed credibility assessment framework—from defining the regulatory question and context of use through model risk assessment, credibility planning, evidence generation, and documentation. Practical scenarios illustrate how weaknesses in datasets, validation methods, or model assumptions can affect the adequacy of supporting evidence.
What the session covers
- FDA’s risk-based framework for AI model credibility assessments
- Defining the model’s context of use and the regulatory decisions it will support
- Assessing model risk through its influence on decisions and the consequences of incorrect conclusions
- Evaluating training data relevance, representativeness, and potential bias
- Establishing fit-for-purpose performance criteria and independent validation evidence
- Characterizing uncertainty, limitations, and applicability beyond development datasets
- Connecting model outputs with supporting scientific and regulatory evidence
- Documenting credibility conclusions and unresolved limitations in regulatory submissions
Designed for your role
Meet the faculty

Carolyn Troiano has more than 45 years of experience in computer system validation in the pharmaceutical, medical device, biotechnology, tobacco, and other FDA-regulated industries. She is currently an independent consultant, advising companies on FDA compliance, Computer System Validation (CSV), and large-scale IT system implementation projects. Carolyn participated in the FDA/Industry Partnership to develop 21 CFR Part 11, the FDA’s Guidance for Electronic Records and Electronic Signatures. During her career she has provided consulting and training, including CSV, 21 CFR Part 11, Data Integrity, and many other related compliance topics of interest to the life science industries.