As AI moves from experimentation to enterprise deployment, leading organizations are anchoring their strategies around domain-specific AI frameworks that operate across clinical, research, and operational workflows. At City of Hope, Hope LLM is an oncology-specific large language model that unifies fragmented cancer data and delivers real-time insights at the point of care, supported by a broader AI governance strategy to ensure safe, scalable deployment.

In this second part of an interview series, Nasim Eftekhari, Chief AI and Analytics Officer at City of Hope, explains how Hope LLM is being applied across the organization’s network to improve oncology outcomes, streamline clinician workflows, and accelerate research.

In conversation with Erik Pupo of Guidehouse, Eftekhari offers a detailed look at where AI is delivering value today and what it takes to scale it responsibly.

Part 1 of the interview covered the real-world evaluation of success and failure in oncology AI and where it is already delivering impact.

Hope LLM at the core of an AI governance strategy

Hope LLM sits at the center of City of Hope’s AI strategy. Rather than relying on a single model, Hope LLM acts as an orchestration layer, bringing together internal and external capabilities to address the complexity of oncology data.

Eftekhari emphasizes that the real value lies not in the model itself, but in its ability to reason across highly fragmented, unstructured clinical information: “Just the ability to read and comprehend and reason over oncology texts in an effective manner can enable many different applications.”

One of the most immediate and widely adopted applications of Hope LLM is patient history summarization. Cancer patients often arrive with years, sometimes decades, of medical records spanning multiple institutions. These records can include:

• Thousands of pages of PDFs
• Scanned handwritten notes
• Radiology and pathology reports
• Fragmented treatment histories

Historically, clinicians or support staff had to manually review and synthesize this information. Hope LLM now takes the information and creates summaries for doctors.

What once took hours can now be done in minutes, enabling care teams to quickly understand a variety of aspects such as prior treatments and responses, adverse events, disease progression, and key decision points in the patient journey.

This shift is not just about efficiency. It directly impacts clinical decision-making quality and speed.

Reducing the hidden burden of care

Beyond clinical insights, Hope LLM is addressing a persistent challenge in healthcare: clinician workload outside of patient hours.

“It has materially decreased what we call ‘pajama time,’ the personal time that doctors spend preparing for their next day, instead of spending with their family,” Eftekhari explains.

By automating documentation review and summarization:

• Clinicians spend less time on administrative tasks
• After-hours workload is reduced
• Engagement with AI increases because the value is immediate and tangible

This is a critical signal. AI adoption in healthcare is not driven by novelty, but by relief from real operational pain points.

Bringing clinical trials into the point of care

Hope LLM is also transforming how patients are matched to clinical trials, a long-standing challenge in oncology. “When the doctor sees a patient, that’s the best place to recruit a patient for clinical trials,” says Eftekhari.

The system continuously evaluates hundreds of active trials, inclusion and exclusion criteria, and patient-specific clinical data.

As a result, at the point of care, clinicians can see the top matching trials along with the justification for each recommendation. This shifts clinical trials from a separate process to an integrated part of care delivery, improving:

• Trial enrollment rates
• Patient access to new therapies
• Clinical decision support

Bridging the gap between patients and trials

In addition to matching patients to trials, Hope LLM is used in reverse to support clinical trial feasibility. Hope LLM helps to find possible patient matches in each location.

This enables City of Hope to:

• Assess potential enrollment before opening a trial
• Select locations based on real patient populations
• Reduce the risk of under-enrollment

Eftekhari highlights the broader impact: “A lot of trials open and close because they cannot recruit enough patients; we are trying to close that gap.”

Accelerating research with real-world evidence

Hope LLM also extends into research, helping teams identify cohorts and extract insights from across databases.

This capability:

• Reduces time to identify study populations
• Enables more precise research design
• Bridges clinical care and research data

While the technology itself is significant, Eftekhari points to a more important shift: how people engage with AI. “At some point, I was always trying to justify my existence, trying to convince people to use these tools. Now it’s the other way around. There is a lot of interest and requests.”

This marks a transition from AI as a push initiative to AI as a pull from clinicians and stakeholders. But this shift introduces new risks.

The need for responsible AI governance implementation

With growing reliance on AI, the focus moves to safe and responsible use. “No matter how good an LLM might be, it still is at risk of hallucinations. We have to make sure that there is no over-reliance,” Eftekhari states.

She describes a fundamental evolution in her role: “I’ve gone from promoting the use of AI to now trying to put governance in place. Governance and AI education go hand-in-hand.”

AI governance roles

City of Hope has implemented a multi-layered model with specific AI governance roles that balances oversight with speed:

• Executive governance committee for strategy and high-risk decisions
• Multidisciplinary AI work group reviewing every use case

The committee includes:

• Clinical and research experts
• Legal, compliance, and ethics
• IT, InfoSec, and AI teams
• HR for workforce impact

Importantly, Eftekhari stresses that governance is designed to keep pace with innovation: “We don’t want something to be stuck in governance; we can review and approve or reject use cases in a matter of days.”

CDO Magazine appreciates Nasim Eftekhari for sharing her insights with our global community.