AI Trial Matching Comes Of Age At City Of Hope
By Dan Schell, Chief Editor, Clinical Leader
I first connected with Simon Nazarian after my colleague Abby Proch included him in an article last year about emerging AI use in clinical research. His comments hinted at something bigger underway at the City of Hope hospital system. When we spoke again, I wanted to go beyond the sound bite and understand what the organization had actually built — and whether it was solving real problems or just riding the AI wave.
What emerged is not a pilot project or a proof of concept gathering dust. It’s a systemwide tool already in use across City of Hope’s national network, helping clinicians digest massive patient records and identify clinical trial opportunities in real time.
At the center of this effort is HopeLLM, an internally trained AI platform designed to support oncology care and research. I was impressed during our conversation to learn how deeply it has been embedded into the AMC’s clinical workflows.
Turning Data Overload Into Usable Insight
Nazarian is quick to clarify that City of Hope did not build a large language model from scratch. Instead, the team trained open-source models on oncology data and layered natural language processing and orchestration capabilities on top.
Essentially, HopeLLM addresses a familiar challenge in cancer care: overwhelming documentation. “Some of our patients could have thousands of pages of notes associated with their care over many years,” Nazarian said. “HopeLLM is able to intake that information and quickly summarize it for the physician.” The time savings are substantial. Physicians often spend hours reviewing records after hours; what Nazarian calls “pajama time.” By summarizing complex histories, the system can save two to three hours per patient. Just as important, the summaries transform unstructured data into information clinicians can use at the point of care.
Trial Matching As Access, Not Just Efficiency
Of course, the capability that most captured my attention was related to clinical trial matching. Historically, oncology trial matching has been slow, manual, and fragmented. Chart reviews can take weeks, and eligible patients are often missed. Meanwhile, oncology records span years of treatment lines, diagnostics, and genomic data.
“It was not just an efficiency issue, but an access issue,” Nazarian said. HopeLLM analyzes multimodal data — including clinical records, claims, genomics, radiology, and pathology — to identify trials suited to a patient. It can also reverse the process, scanning patient populations to identify candidates when new trials open. The goal is immediate visibility. “When a patient comes in for care, we wanted to be able to let them know right away what trials they are eligible for,” he explained. That immediacy matters in oncology, where timing can determine whether a patient gains access to a potentially lifesaving therapy.
City of Hope evaluated commercial tools before building its own. Many struggled with longitudinal oncology data and failed to integrate into real clinical workflows. Designing internally allowed the team to reflect how their oncologists actually think and practice.
Despite its sophistication, though, HopeLLM is not autonomous. As with any discussion involving AI these days, Nazarian stressed that clinician trust and oversight are non-negotiable. “We validate it through side-by-side comparisons with human reviews. Clinicians are embedded throughout development and pilots, AI recommendations are explainable and traceable, and bias monitoring is continuous. Licensed clinicians retain final decision authority. “No algorithm really gets moral authority,” Nazarian said. “AI can assist… but it can’t own clinical accountability.” Further, City of Hope treats governance, privacy, and safety as foundational requirements rather than afterthoughts. The technology is integrated into clinical infrastructure, not layered on top.
Faster Feasibility And A National Trial Model
Beyond matching individual patients, HopeLLM is transforming feasibility assessments across the network. City of Hope has evaluated more than 200 trials using the platform. Tasks that once took weeks can now be completed in minutes, enabling teams to identify eligible patients across multiple locations quickly.
Nazarian describes this as part of a national clinical trials model: centralized coordination paired with geographically distributed research sites. The approach improves access to trials while allowing studies to launch across multiple regions almost simultaneously. Feasibility becomes a strategic advantage instead of a bottleneck.
Quick Adoption … And Hugs
Clinicians are not known for rushing to adopt new technology. Nazarian expected skepticism. Instead, adoption came quickly. “We’ve gotten hugs in the hallway,” he said, initially referring to chart summarization but noting broader enthusiasm.
He attributes uptake to practical value. “It reduces the cognitive burden… and shifts their time away from administrative work and more toward patient care.”
The tool also supports research coordinators and trial staff by surfacing eligible patients and reducing manual screening tasks.
City of Hope gathers continuous feedback and iterates rapidly. Champions helped drive adoption through a train-the-trainer model, and workflow integration made the system intuitive. The most common request so far: expand coverage to legacy patient records.
HopeLLM went live systemwide in early Q4 last year and continues expanding into clinical trials operations workflows. Future development will focus on physician needs, patient benefit, and caregiver support — the three groups Nazarian sees at the center of care. (I really liked that he mentioned the caregiver here, as that echoed some of the points I heard at a recent SCOPE panel discussion I wrote about in my article “SCOPE Takeaways On Inclusion And Real Patient-Centricity.”) In the meantime, the organization is quietly addressing one of clinical research’s most persistent problems: connecting the right patient to the right trial at the right time. And it doesn’t hurt that they’re helping clinicians reclaim a few hours of their evenings, too.