AI in oncology is often framed as a breakthrough technology. But inside leading cancer centers, the conversation is far more grounded: what actually works, for which patients, and how do you measure it?

That is the focus of this four-part CDO Magazine interview series featuring Nasim Eftekhari, Chief AI and Analytics Officer at City of Hope, in conversation with Erik Pupo, Director of Commercial Health IT at Guidehouse. In this first installment, Eftekhari unpacks how oncology AI is evaluated in the real world, where it is already delivering impact, and how organizations should think about success versus failure.

Inside City of Hope’s AI-driven oncology ecosystem

City of Hope stands among the most advanced institutions applying AI in healthcare today. As an NCI-designated Cancer Center, it operates as a national academic medical system with more than 40 locations across the U.S., including major hubs in Southern California, Phoenix, Chicago, and Atlanta.

Its capabilities extend well beyond clinical care:

• A dedicated graduate school advancing cancer and biomedical research
• A genomics research institute enabling precision medicine
• The Beckman Research Institute, a leading center for translational science
• An integrated model combining research, clinical care, and data science infrastructure

This breadth creates a unique environment for AI deployment across:

• Patient care and clinical decision-making
• Precision medicine and genomics
• Research and real-world evidence generation
• Operations and financial optimization

Eftekhari, who has spent nine years at City of Hope, leads the organization’s Department of Applied AI and Data Science, where teams focus on applying AI, machine learning, and advanced analytics across the full continuum of oncology.

“Does This Actually Work?”

For Eftekhari, the starting point is not innovation. It is validation: “The patients are at the center of everything we do. So the most important metrics for me are always performance metrics, especially for something that impacts patient care or patient experience.”

She emphasizes that even seemingly basic questions are often the hardest to answer in practice: “Does this thing work? That seems trivial, but a lot of times, it’s hard to measure.”

This challenge becomes even more pronounced when deploying third-party AI solutions. Many tools arrive FDA-approved, and they often demonstrate success in other institutions or populations. Yet, performance is not guaranteed in a unique patient population.

“That doesn’t mean that they would automatically work for us for our very unique patient population.”

As a result, City of Hope prioritizes local validation before scaling any solution.

Looking under the hood: Separating real AI from hype

In a market saturated with AI-labeled solutions, Eftekhari stresses the importance of scrutiny. “A lot of things these days are AI solutions, and there is AI in everything. So we want to make sure that we look under the hood.”

This evaluation centers on two key questions:

• Is this truly an AI-driven capability or just labeled as one?
• Will it deliver the intended clinical or operational impact?

Only after those questions are answered does the organization consider secondary metrics like efficiency or financial return. “Before thinking about return on investment or efficiencies, the most important metrics for us are always performance metrics.”

Where oncology AI is already delivering real impact

While much of AI in healthcare remains experimental, Eftekhari points to two areas where impact is already tangible and measurable.

1. Diagnostics: AI-assisted radiology and pathology

AI is significantly improving how clinicians interpret medical images. “The AI-assisted radiology pathology tools make life easier for clinicians reviewing these images.”

These tools are:

• Embedded directly into diagnostic workflows
• Enhancing accuracy and speed
• Reducing cognitive burden on clinicians

This is one of the clearest examples of AI moving beyond pilot into everyday clinical use.

2. Clinical documentation and knowledge extraction

The second major area of impact lies in handling unstructured clinical data. “Documentation tools like Ambient Scribe and tools like Hope LLM streamline reviewing patient charts, records summarization, and clinical trial matching.”

These tools address one of healthcare’s most persistent challenges: navigating vast volumes of text.

• Patient records
• Clinical notes
• Trial eligibility criteria
• Research data

“Anything that requires human beings to read through thousands of pages of free text, that’s where AI is making a huge impact.”

This capability extends beyond care delivery into research. “Extracting data and real-world evidence for expediting research is another area where I see a lot of impact. This is happening right now,” Eftekhari explains.

A structured approach to AI investment

City of Hope’s approach to AI prioritization reflects years of iteration. Rather than relying on intuition or isolated ROI metrics, the organization uses a multi-dimensional framework to evaluate every AI initiative.

Key dimensions include:

• Patient care outcomes
• Patient experience
• Provider experience
• Direct financial impact
• Indirect financial impact

Eftekhari highlights an important nuance in healthcare AI: “Is this something that is going to impact many patients, but with a small impact? Or a few patients, but the impact could be potentially life and death?”

This distinction shapes prioritization decisions in a way that traditional ROI models cannot capture.

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