With recent leaps in artificial intelligence, more and more attention is being devoted to algorithms and their effect on society. But beyond chatbots, image generators, social media and search engines, there is major potential to turn computing advances into gains for human health.
Scientists at City of Hope are determined to tap that potential, and the Department of Surgery is one hub where researchers are turning AI to patient benefit. Homegrown City of Hope technologies are emerging, with some starting to exert an impact on care, thanks to the Applied AI and Data Science team’s efforts.
“The possibilities for AI are nearly limitless,” said Yuman Fong, M.D., City of Hope’s Sangiacomo Family Chair in Surgical Oncology and professor and chair of surgery. “We’re already seeing it make the hospital more efficient — cutting down on wait times — and influence patient care.”

Machine learning involves training algorithms using mountains of complex data. This AI can comb through input beyond the human capability to analyze and uncover patterns that people are unable to perceive. The rich data in the electronic health record is a crucial resource for tools that can inform treatment planning.
“We need better answers to the questions that come up along the patient’s journey,” said Mustafa Raoof, M.D., M.S., an assistant professor of surgery and of cancer genetics and epigenetics at City of Hope whose surgical specialty is tumors within the abdomen. “Whatever decisions they need to make, we want to help them make those choices better, and as individualized as possible.”
AI in Action for Better Surgical Outcomes
Starting in late 2021, City of Hope integrated an algorithm in patient records that evaluates risk for serious complications from emergency surgery. The project stretches back about six years, spearheaded by Lily Lai, M.D., professor of surgery and director of the surgical oncology training program. The inspiration was an existing risk calculator created by the American College of Surgeons (ACS) in the early 2000s.

“Our patients have specific clinical characteristics different from the general population,” Lai said. “They’ve gone through blood stem cell transplants, chemo, radiation. So we thought, ‘Let’s develop our own risk score.’”
The team of surgeons and data scientists began with the pivotal variable of whether or not surgery led to subsequent hospitalization, further operations, lasting disability or death. Machine learning connected that binary outcome with the bevy of factors in the electronic health record, including how some numbers changed over time.
“The hope is to give surgeons a more quantitative measure,” said Nasim Eftekhari, M.S., City of Hope’s executive director of Applied AI and Data Science. “When a doctor sees a patient, they do some calculations in their head regarding the risk. With the advantage of millions of data points, we try to offer information that they can use to make better decisions.”

The algorithm ultimately proved a more accurate predictive tool for City of Hope patients than the existing ACS calculator. A feasibility study showed that surgeons using the risk score had a greater reduction in complications than peers who didn’t. Dr. Lai and her colleagues are working diligently to determine the underlying reasons.
In her own practice, she has found that the new risk score — which includes the top 10 factors affecting that risk — helps sharpen focus on whether a given surgery should happen. Sometimes there are factors that can be modified before surgery, such as raising low blood counts via transfusion.
“In some cases, it helps us ask the questions, ‘What is the goal of treatment here?’ or ‘Can we make changes to better prepare them for the operation?’” Dr. Lai said. “In others, it confirms that patients are good risk candidates.”
Extending Risk Reduction to Abdominal Surgeries
As part of Dr. Raoof’s machine-learning investigations, he aims to originate a similar risk score for another population of patients, those with cancers that have spread to the lining of the abdominal cavity, and procedure — surgery to remove tumors, with or without a bath of heated chemotherapy known as hyperthermic intraperitoneal chemotherapy, or HIPEC, to eliminate unseen traces of cancer. This complicated operation can last eight to 12 hours.

Dr. Raoof and his colleagues employ a dataset encompassing more than 2,000 patients from three dozen institutions in the U.S. called the HIPEC Collaborative. Steadily, they have drilled down from evaluating risks for serious complications in the overall patient population to identifying six subgroups at higher risk. Along the way, they have untangled the complex relationships between two previously known risk factors: length of surgery and amount of blood loss.
“How long surgery takes is not actually the most important thing,” Dr. Raoof said. “It’s only relevant with blood loss of half a liter or more. A longer surgery can indicate a meticulous approach that pays off in reducing risk. But if you have a short surgery with high blood loss, the patients suffer more complications.”
He and his team have moved on to deriving individualized risk scores. With success in planned validation studies with data from an external, independent cohort of patients, the algorithm will be in line for clinical application, helping safeguard even more people who seek help from City of Hope.
Recovery as Another Frontier for AI
Meanwhile, Dr. Fong, the surgical chair, has been exploring how AI can inform decisions after surgery. Advances in minimally invasive strategies such as robotic surgery are shortening, or even eliminating, hospital stays after certain procedures. Dr. Fong wants to ensure that choices about care after surgery set patients on the best path to healing.
To that end, he and Virginia Sun, Ph.D., M.S.N., R.N., professor of population sciences at City of Hope, have teamed up to lead a project backed by a $4 million grant from the Patient-Centered Research Institute.

“We’re looking at who to send home early, who to send home a few days later, who needs a call from us or a visit from family, and how soon they should come back to see us after they’re discharged,” Dr. Fong said.
The researchers are stocking their algorithm with a rich array of data collected at home, including sensor readings of vital signs, blood sugar levels and daily steps taken and patients’ own reports of symptoms and quality of life. Work is ongoing, and Dr. Fong is sanguine about the prospects for this strategy.
“Down the line, I’m convinced that we will be monitoring patients — not just surgical patients, but all of them — at home in some fashion,” he said. “There may be subtle patterns that we as humans can’t discern but AI can actually put together.”