Emerging Programs

AI Workgroup Overview

The Emerging Programs AI workgroup, led by CTTI, focused on exploring the potential of Artificial Intelligence (AI) and Machine Learning (ML) to improve the quality of clinical trials.

Through a series of collaborative meetings, the workgroup, comprised of individuals grounded in knowledge about AI and ML, aimed to identify opportunities and challenges in integrating these tools into clinical trials. The discussions provided insights into the evolving landscape of AI and ML in clinical trials and helped keep a pulse on future directions. By fostering continued collaboration, CTTI aims to accelerate the responsible adoption of AI and ML in clinical trials, making them more effective, inclusive, and aligned with the needs of both patients and researchers.

Questions

Contact Lindsay Kehoe at Lindsay.kehoe@duke.edu

Meeting Materials Workgroup Members Special Guests Collective Summary

Meeting Materials

Topic: Selection and Recruitment of Trial Participants
- Summary
Topic: Site Selection and Participant Adherence and Retention
- Summary
Topic: Dosing/Dose Regimen Optimization
- Summary
Topic: AI Applications in Clinical Data Collection, Management, and Endpoint Assessment
- Summary
Topic: Trial Design, Planning, & Configuration
- Summary

Workgroup Members

Members

Sal Campo (tenXclinical)
Dominic DeBellis (Merck)
Luca Emili (InsilicoTrials)
Hussein Ezzeldin (FDA/CBER)
Rachael Fleurence (NIH)
Amy Froment (Regeneron)
Nijat Hasanli (Lindus Health)
Lara Jehi (Cleveland Clinic)
Ece Kalvalci (Lindus Health)
Rohan Khera (Yale)
Jane Knochel (InsilicoTrials)
Federica Lucivero (Oxford)
Oleg Milberg (Regeneron)
Michael Pencina (Duke)
James Riddle (Advarra)
Marsha Samson (FDA/CDER)
Mario Torcio (InsilicoTrials)
Henry Wei (Regeneron)

Special Guests

Collective Summary

CTTI Workgroup Summary of Barriers and Gaps in Use of Artificial Intelligence and Machine Learning for Clinical Research

For the past 15 years, CTTI has been committed to developing and driving the adoption of practices that improve the quality and efficiency of clinical trials. CTTI recognizes that Artificial Intelligence (AI) and Machine Learning (ML) offer tremendous potential to help us achieve these goals, opening up new opportunities to address longstanding challenges in clinical trials.

To follow this evolving technology, CTTI brought together a diverse group of stakeholders to discuss AI and Machine Learning applications in clinical trials and the opportunities to address ongoing inefficiencies and quality issues. Below is a summary of the five Emerging Programs AI Workgroup meetings that met between August 2024 and July 2025. Topics for the first four meetings were taken from Section C of the Food and Drug Administration’s discussion paper entitled Using Artificial Intelligence & Machine Learning in the Development of Drug & Biological Products. This chart reflects what was discussed during those meetings and is not meant to cover all challenges, applications, or barriers.

Topic	Ongoing Clinical Trial Enterprise Challenges	AI Opportunities/Applications	Examples
Participant Recruitment & Selection	overly restrictive eligibility slow recruitment methods difficulties identifying eligible patients and matching to trials lack of representativeness	Use of trial emulation, machine learning, and real-time insights into patient populations to: automate eligible patient identification improve matching accuracy simplify consent forms foster more data driven and inclusive trial designs	Evaluating eligibility criteria of oncology trials using real-world data and AI, Nature, 2021 The role of artificial intelligence in hastening time to recruitment in clinical trials, BJR, 2023
Participant Adherence & Retention	maintaining retention and adherence throughout a trial underuse of drug, device, or trial data	Use of Natural Language Processing (NLP) to collate unstructured data formats, such as clinical records, social media, and patient interactions to: better appreciate the social characteristics when assessing behaviors and factors (e.g. word of mouth, dose, regimen, adverse event) influencing retention and adherence understand the performance of medications and devices in the real world	A Patient-Centered Conceptual Model of Symptoms and Their Impact in Early Parkinson's Disease: A Qualitative Study, J Parkinsons Dis, 2022 How One Health System Uses AI-based Technologies to Identify Patients’ Social Determinants of Health, Managed Healthcare Executive (blog), 2023
Site Selection	labor-intensive, often relying on expert intuition or small, structured datasets	Use of Natural Language Processing (NLP) to collate data, such as clinicaltrials.gov, to: identify trial sites with the right capabilities	Application of an automated natural language processing (NLP) workflow to enable federated search of external biomedical content in drug discovery and development, Drug Discov Today, 2016
Dosing/Dose Regimen Optimization	relies heavily on expert-guided pharmacometric models, which can be complex, labor-intensive, and limited in scalability	Use of data-driven modeling, causal inference, and simulation-based scenario testing to: improve efficiency, accuracy, & personalization in dose selection	Deep learning prediction of patient response time course from early data via neural-pharmacokinetic/pharmacodynamic modelling, Nat Mach Intell, 2021 Machine Learning-Based Quantification of Patient Factors Impacting Remission in Patients With Ulcerative Colitis: Insights from Etrolizumab Phase III Clinical Trials, Clin Pharmacol Ther, 2024
Data Collection, Management, & Endpoint Assessment	agreement over meaningful and acceptable endpoints use and reliability of RWD inefficiency and inconsistency in data collection and management	Use of machine learning and GenAI to support: eCRF automation, audit trail review, image analysis, data reconciliation, and medical coding use of synthetic data to predict outcomes, serve as control arms, and improve trial design	Continuous Wearable Monitoring Analytics Predict Heart Failure Hospitalization: The LINK-HF Multicenter Study, Circ Heart Fail, 2020 Predicting Missing Values in Medical Data via XGBoost Regression, J Healthc Inform Res, 2020
Trial Design, Planning, & Configuration	overly complex, burdensome or not adequate and well controlled trial design slow study start-up	Use of application-specific language models (ASLMs), deterministic logic and probabilistic large language models (LLMs) to: interpret protocols, generate form logic, and accelerate system buildout reduce complexity and enable rapid study startup, particularly for protocol configurations that repeat across trials	A policy framework for leveraging generative AI to address enduring challenges in clinical trials, npj Digital Medicine, 2025 Clinical Trials Protocol Authoring using LLMs, arXiv:2404.05044, 2024

AI and ML Implementation Barriers

General Barriers	Topic Specific Barriers
Integrating AI with clinical practices Ensuring data quality Domain-specific expertise to fine tune AI models Continuous data validation Data privacy and transparency in decision-making processes Costs and operational challenges Lack of familiarity, confidence, and/or understanding of ROI Regulatory acceptance uncertainty	Participant Selection and Recruitment: a lack of real-time data sharing Participant Adherence & Retention: reporter bias Site Selection: continuous data validation Dosing/Dose Regimen Optimization: patient and data variability, exposure-response interpretation/model interpretability, and complex biological responses that challenge simple model and causality assumptions Data Collection, Management, & Endpoint Assessment: lack of consistent standards and governance for use of real-world data and for data reuse; need to differentiate between synthetic control arms, external controls, and digital twins as each has distinct regulatory and ethical implications; lack of metrics and key performance indicators like eCRF build time or coding throughput Trial Design, Planning, & Configuration: current validation (e.g., computer system validation and audit frameworks) are not designed for AI-assisted tools, creating risk in configuration accuracy

General Barriers

Topic Specific Barriers

Integrating AI with clinical practices
Ensuring data quality
Domain-specific expertise to fine tune AI models
Continuous data validation
Data privacy and transparency in decision-making processes
Costs and operational challenges
Lack of familiarity, confidence, and/or understanding of ROI
Regulatory acceptance uncertainty

Participant Selection and Recruitment: a lack of real-time data sharing
Participant Adherence & Retention: reporter bias
Site Selection: continuous data validation
Dosing/Dose Regimen Optimization: patient and data variability, exposure-response interpretation/model interpretability, and complex biological responses that challenge simple model and causality assumptions
Data Collection, Management, & Endpoint Assessment: lack of consistent standards and governance for use of real-world data and for data reuse; need to differentiate between synthetic control arms, external controls, and digital twins as each has distinct regulatory and ethical implications; lack of metrics and key performance indicators like eCRF build time or coding throughput
Trial Design, Planning, & Configuration: current validation (e.g., computer system validation and audit frameworks) are not designed for AI-assisted tools, creating risk in configuration accuracy