Trial Matching, Source Data, And More: How Clinical Researchers Used AI In 2025

By Abby Proch, executive editor, Clinical Leader

Across industry and academia, from discovery to source data and more, AI is becoming more commonplace in helping clinicians focus more on patients while managing vast streams of complex data. Discover how a handful of companies tapped into AI over the past year.

City of Hope Adds In-House, Trial-Matching AI Tool

Simon Nazarian, chief digital and technology officer, City of Hope

Getting the right patient to the right clinical trial at the right time has long been a challenge in cancer research. Manual chart reviews take weeks, opportunities get missed and eligible patients slip through the cracks. At City of Hope, we created and deployed HopeLLM, a proprietary AI tool, to fundamentally change that equation.

 The tool scans longitudinal patient records against hundreds of active trials, surfacing matches that might otherwise be overlooked. HopeLLM also scans hundreds of thousands of patients against one trial to find eligible clinical trial patients across City of Hope's 38 locations nationally. It’s not just about speed: it’s about processing volumes of data more thoroughly than any human could working alone.

 We’re building the digital infrastructure capabilities that enable truly data-driven precision medicine at scale.

Since go-live, HopeLLM Clinical Trials Matching has supported feasibility for more than 200 trials, rapidly identifying potentially eligible patients across all City of Hope locations. In parallel, a limited group of physicians is already using it in real clinical decision making to surface life-saving trial options for patients who may otherwise have no viable path.

We built this capability in-house because existing solutions did not meet the complexity and the realities of our oncology population, our workflows, or the level of precision and flexibility needed at an academic cancer center.

Importantly, HopeLLM is always human-in-the-loop: physicians and study coordinators remain the decision makers, while the system reduces cognitive burden by organizing evidence, accelerating screening and making trial matching dramatically faster and easier.

We plan to expand access broadly to all clinicians in 2026.

AI Stack Helps Cancer Doc Find Treatments And Trials

Roupen Odabashian, M.D., FRCPC, FASCO, Hematology/Oncology Fellow, Barbara Ann Karmanos Cancer Institute

In my daily oncology practice, AI and clinical decision support have become indispensable tools for delivering optimal patient care. I routinely leverage AI-powered platforms, such as Doximity GPT, OpenEvidence, and UpToDate AI to navigate the rapidly evolving landscape of cancer treatment. These tools help me identify the most current evidence-based treatment options, locate relevant clinical trials — including early-Phase 1 and 2 studies — and access the latest research papers that directly inform my patient management decisions.

What I've found particularly valuable is how these AI tools can synthesize vast amounts of clinical data and literature in real time, allowing me to make more informed decisions at the point of care. Whether I'm evaluating a complex case or searching for novel therapeutic options for a patient who has exhausted standard treatments, AI-powered clinical decision support significantly enhances my ability to provide comprehensive, cutting-edge care. Additionally, AI scribe and documentation tools now help me generate clear, accessible summaries for patients, improving communication and shared decision-making.

Beyond my clinical practice, I am developing an AI-powered educational platform designed specifically for oncology learners and fellows. This tool creates personalized learning environments where residents, fellows and early-career oncologists can work through clinical questions and case scenarios with an AI assistant guiding them at every step. Rather than simply providing answers, the AI assistant offers real-time, personalized feedback tailored to each learner's knowledge gaps and learning pace. It's about creating a supportive learning companion that helps clinicians develop the critical thinking skills they need for board certification and real-world practice. My goal is to bridge the gap between traditional medical education and the practical demands of modern oncology, ensuring the next generation of oncologists is prepared to deliver the highest standard of care.

Technetium Uses Proprietary AI For Discovery

Cheng Hu, CEO and cofounder, Technetium Therapeutics

At Technetium Therapeutics, we've engineered a paradigm shift in drug discovery with our proprietary AI Agentic Workflow, TechnetiumX. This platform moves us from a linear, high-attrition pipeline to a dynamic, in-silico evolution engine.

Our focus is de novo drug design, creating novel small molecules from first principles. The core challenge is the intractable vastness of chemical space (~10⁶⁰ compounds), a problem too complex for any single model.

Our solution is a recursive, multi-agent system that decomposes discovery into specialized tasks. The workflow integrates a Generative Agent for molecular design and a Distillation Agent for initial filtering.

Critically, we replace the limiting "human-in-the-loop" approach with an "agent-in-the-loop" core. Our Interaction Review Agent performs a deep, residue-level analysis of binding interactions—evaluating hydrogen bond stability and π-stacking—to validate true binding mode quality beyond simplistic docking scores. This agent augments human expertise by handling the iterative, scaling workload, freeing our scientists to guide high-level strategy while the AI executes the vast exploration.

This architecture is powered by precision training. By leveraging SandboxAQ's high-quality SAIR dataset (5 million data points) and our team's deep domain expertise to extract specific, meaningful features, we moved beyond brute-force training. The result is a predictive capability that sets a new industry benchmark.

The tangible outcome is a remarkable 23% hit rate validated by wet experiments—a massive improvement over the industry average of less than 1%.

This performance compresses the initial discovery timeline from years to weeks and has enabled us to generate novel lead series for five distinct therapeutic targets across oncology, immunology, and neurodegeneration. We are now navigating the infinite molecular search space with unprecedented speed, precision, and success.

PI Relies On AI For Notetaking And Source Data

Mikel Daniels, DPM, MBA, president and chief medical officer, WeTreatFeet Podiatry

When I am recruiting for a study or conducting a follow-up visit, my attention is split. I am trying to build rapport with a nervous patient while simultaneously ensuring I capture every needed detail required by the protocol, not to mention other regulatory bodies. Questions are always on my mind, as I don’t want to miss anything. Did they take the study medication? Any new concomitant meds? A slight headache last Tuesday? If I miss those details because I’m staring at a screen, the data quality suffers.

That’s where AI ambient listening has changed everything for our clinical research at WeTreatFeet Podiatry. It allows our interactions to flow naturally, without awkward pauses for typing. It lets me keep my eyes (and mind) on the patient, not the keyboard, ensuring that the "source data" for a trial is actually coming from the patient, not my memory of the conversation 10 minutes later.

With ambient AI, I am actually present. I can read a patient’s body language, spot hesitations when I ask about adverse events, and take in the things they’re not saying. My patients notice. They feel like I am understanding every word, which builds the trust necessary for retention. Because the AI logs everything in real time, the record is less likely to miss important details. It can also be screened, placed into the correct documentation platform, and satisfy the requirements for the study and data integrity.