Feature | AI and The Changing Landscape of Clinical Trials
Clinical trials are expensive, resource-heavy, time-consuming and insufficiently diverse. And it's not unlikely that the result doesn't answer the question or drive change or improvement in clinical practice or is outdated or no longer relevant.
Artificial intelligence (AI) offers potential solutions to these longstanding challenges, promising to revolutionize how we design, conduct and analyze cardiovascular research. From identifying eligible patients to adjudicating clinical outcomes, AI-augmented tools are beginning to reshape the clinical trials landscape in dramatic and exciting ways.
The current system is ripe for transformation, according to Sneha S. Jain, MD, MBA, a cardiologist and health AI researcher at Stanford University who works with the Stanford Health Care Data Science team to build, deploy and evaluate AI applications in medicine. With pivotal cardiovascular trials costing on average over $35,000 per participant and taking upwards of a decade to complete, the need for innovation has never been more urgent.1
Patient Screening in Seconds
One of the most promising applications of AI in clinical trials is automated patient identification. Traditionally, recruiting participants has been labor-intensive and inefficient, with research coordinators manually reviewing charts to find eligible patients. Now, AI-powered tools can screen thousands of patient records in minutes.
At Stanford, researchers have developed ChatEHR, an AI system that allows clinicians to query patient electronic health records (EHRs) using natural language.
The system could work bidirectionally. There are ongoing initiatives to help clinicians start with a patient and find appropriate trials or start with a trial and identify eligible patients from their entire patient population. It works also for pure clinical assistance: "I can query the chart and ask for the patient's health care journey over the last six months since I last saw them in clinic. Usually that requires over 20 clicks to review their other clinical visits and care. Now it's all summarized in bullet form in seconds," explains Jain.
The RECTIFIER tool, recently validated in the COPILOT-HF trial, demonstrates similar capabilities for clinical trials.2 Using GPT-4 and Retrieval-Augmented Generation, RECTIFIER assessed complex eligibility criteria in potential participants' EHRs with 98-100% agreement with expert clinicians – at a cost of just $0.10 per patient screened. This represents a dramatic improvement in both efficiency and cost-effectiveness compared to traditional manual screening.
Newfangled Trials
Sometimes running a clinical trial is just not feasible. In a recent NEJM AI article, three Harvard trial experts led by Issa J. Dahabreh, MD, ScD, from CAUSALab, review the ins and outs of target trial emulation, simulation and augmentation.5
Target trial emulation is a methodological framework for designing observational studies that mimics a hypothetical randomized controlled trial (RCT). Researchers explicitly specify the ideal trial they would conduct and then use real-world data to emulate it as closely as possible. This approach helps avoid common pitfalls like biases related to timing misalignments.
Target trial simulation extends this framework by using observational data to inform the design of future actual trials. Before committing resources to an RCT, researchers can run simulations exploring how different design choices, such as eligibility criteria or sample size requirements, might affect trial outcomes and feasibility.
Augmentation combines the gold standard of randomization with large-scale observational data. This hybrid approach leverages the strengths of both types of evidence to improve statistical power and generalizability.
While computational tools and AI can automate much of the data extraction and analysis in these methods, they cannot eliminate fundamental issues inherent in the data like unmeasured confounding. Hence, endowing the results of target trial emulations with a causal interpretation rests on assumptions, which are untestable using the data. Nevertheless, emulation becomes particularly valuable for designing better trials via data-driven simulation and determining when augmentation with real-world evidence is appropriate.
The AI Trial Coordinator
AI has the potential to transform not just the logistics of trial operations, but the participant experience itself. Informed consent conversations are often ineffective and may bias towards enrollment. Already large language models have been successful in reducing the complexity and reading time of surgical consent forms and their use is being considered for clinical trials.
"The chatbot isn't time constrained, doesn't get impatient, and can speak any language you need and explain things at every grade level required," notes Jain. To boot, it may reduce barriers to enrollment by reducing patient visits.
AI is enhancing the collection and refinement of continuous monitoring data from wearable devices. This "digital biomarker data" may include metrics derived from vital sign measurements, skin temperature, arrhythmia monitoring, accelerometry, GPS data to identify potential health care encounters, and even speech analysis to detect pulmonary congestion.1 This decentralized approach reduces the burden on participants but also has the potential to expand trial participation to patients living far from academic medical centers.
Automated Outcome Adjudication
Perhaps one of the most currently impactful applications of AI in cardiovascular trials is automated clinical event adjudication. Currently, outcomes like heart failure (HF) hospitalizations are adjudicated by clinical events committees – a process that is labor-intensive, expensive and time-consuming.
Natural language processing (NLP) models can now perform these tasks automatically. In the INVESTED trial, an NLP model for adjudicating HF hospitalizations demonstrated 87% agreement with human clinical events committees, with performance improving to match human-level reproducibility after fine tuning.3 Such tools could dramatically reduce costs while providing more consistent and rapid outcome assessment.
Business Model Disruption
While the technical capabilities of AI are impressive, Jain and colleagues emphasize in a recent article in JACC that realizing this vision requires more than just better algorithms.4 "I don't think we're going to get there unless we advance efforts across digital infrastructure, business model innovation and regulatory frameworks."
Current clinical trial business models are built around labor-intensive processes managed by contract research organizations. Noting it can be hard to innovate within that infrastructure, she argues that organizations must create "direct lines to leadership to incubate and innovate and advance efforts like this."
Regulatory modernization presents another critical challenge. The current Food and Drug Administration (FDA) clearance process for AI tools differs fundamentally from drug approval, requiring only that a tool performs as claimed. "There's a chasm between FDA clearance and actual proven benefit and deployment in health care systems and clinical trials," Jain explains. She emphasizes the need for academic and industry stakeholders to develop robust evaluation frameworks.
The Path Forward
The vision of AI-transformed clinical trials is already beginning to materialize. Tools like ChatEHR at Stanford, RECTIFIER for patient screening, and NLP-based outcome adjudication are demonstrating proof of concept today.
By 2030, Jain predicts, many of the innovations discussed will be standard practice: AI agents explaining protocols in personalized language, wearable devices providing continuous monitoring, automated outcome adjudication, and seamless integration of trial data with EHRs. The challenge now is not whether these technologies can work, but whether health care institutions and regulatory bodies can adapt quickly enough to realize their full potential.
AI Unlocked: Trends, Tools and Tomorrow
Ready to bring AI into practice? Mark your calendar now for the three-part AI Intensive at ACC.26, including not-to-miss Keynotes on Scaling AI in Cardiology: Moving From Paper and Podium to Product and The Future of AI in Cardiology and Bottlenecks Translating Research to Clinical Care. Visit ACCScientificSession.org to learn more and to search the entire program for more on AI.
This article was authored by Debra L. Beck, MSc.
References
- Cunningham JW, Abraham WT, Bhatt AS, et al. Artificial intelligence in cardiovascular clinical trials. JACC. 2024;84:2051-62.
- Unlu O, Shin J, Mailly CJ, et al. Retrieval augmented generation enabled generative pre-trained transformer 4 (gpt-4) performance for clinical trial screening. medRxiv. Preprint posted online Feb. 8, 2024:2024.02.08.24302376.
- Cunningham JW, Singh P, Reeder C, et al. Natural language processing for adjudication of heart failure in a multicenter clinical trial: a secondary analysis of a randomized clinical trial. JAMA Cardiol. 2024;9:174-81.
- Jain SS, Sarraju A, Shah NH, et al. The Coming AI Revolution in Clinical Trials. JACC. 2025;85:378-80.
- Dahabreh IJ, Yeh RW, De Bartolomeis P. Trial emulation, simulation, and augmentation using electronic health records and generative AI. NEJM AI. 2025;2:AIe2500894.
Keywords: Cardiology Magazine, ACC Publications, CM-Jan-Feb-2026, Electronic Health Records, Artificial Intelligence, Data Science, Clinical Trials as Topic
