Fabio Ynoe de Moraes, Radiation Oncologist and Associate Professor at Queen’s University, shared a post on LinkedIn about a recent paper he co-authored with colleagues published in Nature:
“Clinical Trials Are Entering a New Era — And AI Is the Catalyst
Nature Digital Medicine, Nature Portfolio, Nature Magazine
Co-authors: Aarav Badani, Philipp Vollmuth, Caroline Chung, Alireza Mansouri
For decades, clinical trials have been slow, expensive, exclusionary, and unable to keep pace with scientific innovation. Our new article published today in npj Digital Medicine argues a simple but urgent thesis:
AI will not just optimize clinical trials — it will redesign them. From eligibility → to treatment arms → to safety → to patient communication.
Here’s the future we outline:
1. Eligibility 2.0 — Precision Without Exclusion
AI and real-world data reveal what many suspected:
- Half of today’s eligibility criteria do not change outcomes.
- Broadening criteria can double the number of eligible patients — safely.
We show how ML systems like Trial Pathfinder can eliminate legacy exclusions and build more representative, equitable trials.
2. Adaptive Trials Powered by Reinforcement Learning
Imagine a trial that updates itself:
- Reallocates patients
- Adds or drops arms
- Adjusts dosing
- Learns from interim data in real time
We outline how reinforcement learning plus Bayesian frameworks create trials that are faster, smarter, and more ethical.
3. Digital Twins: From ‘N=1’ to Synthetic Control Arms
Digital twins let us simulate:
- Patient-specific trajectories
- Virtual treatment arms
- In-silico trials
- Early identification of failure
They offer an ethical solution when control arms are problematic, especially in oncology and rare diseases.
4. AI Agents — The New Trial Coordinators
LLM-based multi-agent systems can now:
- Interpret protocols
- Match patients
- Predict outcomes
- Monitor toxicity
- Coordinate trial workflows end-to-end
This goes beyond prediction — it is autonomous execution.
5. Patient-Centered Trials, Powered by LLMs
LLMs can finally make trials understandable:
- Simplified consent documents
- Culturally adapted communication
- EHR → trial-matching pipelines
- Tools to increase public trust
Only 2–3% of eligible patients currently enroll.
We can’t fix trials without fixing communication.
6. Bias, Regulation, and Trust Remain the Hardest Problems
We outline the essential safeguards:
- CONSORT-AI and DECIDE-AI
- Mechanistic + ML hybrid models
- Uncertainty quantification
- Federated learning for global collaboration
- Regulatory sandboxes with FDA/EMA
The message is clear:
AI in clinical trials is not plug-and-play. It demands rigorous science, transparent validation, and ethical guardrails.
The Future: Clinical Trials as Living, Learning Systems
Trials will evolve from static to dynamic, from rigid to responsive, from costly to intelligent, and from exclusive to equitable.
This is not a technological dream.
It is the next chapter in evidence-generation — and it is already happening.
Your turn Where do you see the biggest opportunity (or risk) for AI in clinical trials?”
Title: AI and innovation in clinical trials
Authors: Aarav Badani, Fabio Ynoe de Moraes, Philipp Vollmuth, Caroline Chung, Alireza Mansouri
Read the Full Article in Nature.
More posts featuring Fabio Ynoe de Moraes.
