AI news ·
Weill Cornell Medicine researchers develop artificial intelligence system to speed clinical trial design using real-world patient data
Weill Cornell's EmulatRx uses five AI agents to design clinical trials with real-world patient data. The system simulates protocols to flag pitfalls before enrollment begins.
A new AI system that operates as a collaborative team of five specialized agents could accelerate the design of randomized clinical trials, one of the most expensive and time-consuming stages of drug development. The system, called EmulatRx, was developed by Weill Cornell Medicine researchers and described in a study published in Nature Communications. It uses real-world patient data to simulate trial protocols, flag potential pitfalls, and recommend refinements before a single patient is enrolled.
Clinical trial design is a bottleneck for new therapies, requiring months of painstaking coordination among clinicians, statisticians, and data specialists. "Randomized clinical trials are the gold standard for determining whether new treatments are safe and effective," said senior author Dr. Fei Wang, associate dean for AI and data science at Weill Cornell. "But designing them is a slow, expensive and highly complex process."
A virtual research team
EmulatRx is built around five large language model (LLM)-powered agents that mirror a real scientific team. A central Supervisor agent coordinates the workflow, while a Trialist, Informatician, Clinician, and Statistician each handle distinct parts of the design process. "Because each EmulatRx agent is empowered by a large language model, they can exchange information in natural language and work together much as human experts do," Wang said.
The Trialist extracts key elements from past studies-eligibility criteria, treatment arms, endpoints-to outline a trial structure. The Informatician translates those requirements into database queries that can pull appropriate patient records from electronic health records. The Clinician checks that the design aligns with medical knowledge and published evidence, and the Statistician uses real-world data to estimate treatment effects and sample sizes.
Learning from real-world patients
To evaluate the system, the team used de-identified electronic health records from large clinical databases covering both acute conditions (heart failure, septic shock, kidney injury) and chronic diseases (Alzheimer's and Parkinson's). EmulatRx applied a technique called target trial emulation, which applies the causal framework of a randomized trial to observational data, to identify eligible patient groups and compare outcomes.
The system reproduced many treatment effects reported in historical trials, suggesting it could help researchers assess trial designs before committing to costly studies. It also surfaced subgroup differences-for instance, when a treatment helped one population but increased risk in another-that could inform more precise and safer eligibility criteria. This approach draws on AI for Healthcare methods that combine structured diagnosis codes, lab results, and free-text clinical notes to build a richer picture of patient populations.
Human in the loop
Rather than running autonomously, EmulatRx is designed for researcher oversight. Experts can monitor the agents' exchanges, pause the process, and correct missteps, with the system learning from those interventions. "It is important to keep a human in the loop, so the system does not go in an unreasonable direction," Wang said.
The system is not yet ready for clinical deployment; it requires broader validation across other health systems and data types. The researchers are working on commercial development and hope to make it available for investigator-initiated trials at universities. Wang stressed that randomized controlled trials remain essential: "The question is how to design them so they can be conducted more efficiently and have a higher chance of success."
Why this matters for Science and Research
For researchers who design or conduct clinical trials, EmulatRx points to a future where AI agents can pre-test protocols, identify hidden biases, and optimize inclusion criteria before ethical and financial commitments are made. This kind of system-applying AI for Science & Research to real-world evidence-does not replace human judgment but can dramatically reduce the iterative, manual work that slows trial launches. The takeaway: integrating collaborative AI into study design workflows could help research teams allocate resources to trials with the highest likelihood of producing actionable results.