How can hospitals boost cancer screening rates? Researchers report that an AI helped

CHICAGO — People can be mightily reluctant to get a colonoscopy for reasons like cost, language barriers, or fear of what the exam might discover. One of the best counters to such barriers are patient navigators who can help resolve most such reasons. The trouble is that there are only so many navigators at a given cancer center. Scientists at Montefiore Einstein wondered if AI could help.

Alyson Moadel-Robblee, a psychologist working on psycho-oncology at Montefiore Einstein Comprehensive Cancer Center, said the navigator team there — as at many hospitals — simply can’t reach all of the nearly 3,000 patients a year who don’t make their colorectal cancer screening.

“Since the pandemic, the demand and load on our healthcare team is huge. They’re stretched thin. They can only do so much,” Moadel-Robblee said. “They can’t call in the evening or whenever.”

So, Moadel-Robblee and other scientists at Montefiore Einstein partnered with the tech company MyndYou to create a conversational artificial intelligence to do the bulk of the outreach for patients who no-showed or canceled colonoscopy screening appointments. In an abstract presented here at the meeting of the American Society of Clinical Oncology on Saturday, the team reported the AI helped double the colonoscopy completion rate among this group.

The AI, dubbed MyEleanor, isn’t designed to replace human navigators, Moadel-Robblee explained. “She” calls patients who didn’t show up or canceled their colonoscopy appointments. If they pick up, she has two primary directives: transfer them over to a human navigator and, if the patient consents, guide them through a brief survey on why they missed their appointment.

“Our virtual navigator, she doesn’t sleep. So she can call earlier, later, or on different days. The navigators that are human are invaluable. They have the human touch. We can’t replace them, but we can supplement,” Moadel-Robblee said.

In an age where computers seem to be galloping ever closer to passing the Turing test, the scientists made some very deliberate decisions on MyEleanor’s sophistication. For one, she doesn’t sound particularly human, Moadel-Robblee said. “Our peer navigators, cancer survivors, said ‘I don’t want this bot to sound like a person and not know if they’re not real,’” she said. So, MyEleanor retains the speech of a robot. But, Moadel-Robblee insisted, without sounding impersonal.

“She’s really warm,” Moadel-Robblee said. “She says, ‘I’m Eleanor. I speak English and Spanish, which would you prefer? I’m part of the care team.”

It seems to have worked with a good chunk of patients. Of 2,400 patients that MyEleanor called, 57% stayed on the line with the bot. Among those, over half agreed to transfer to a human navigator to try to reschedule their appointment. Ultimately, human navigators were able to increase the proportion of no-show patients who completed their colonoscopy from 10% to 19% with MyEleanor’s assistance. Not only that, but MyEleanor helped free up an average of 52 work hours per month for each human navigator.

“This quality improvement initiative is truly an innovative means of increasing cancer screening,” said Fumiko Chino, a radiation oncologist at Memorial Sloan Kettering Cancer Center during an online ASCO press briefing. She did not work on the study. “It really offloads the burden from an overworked healthcare workforce by leveraging AI technology to optimize the outreach capacity to vulnerable populations.”

These are particularly good numbers considering a significant proportion of the Bronx is underserved and may face multiple barriers to accessing care. In the study, Moadel-Robblee said, about a quarter of patients are primarily Spanish speakers, about 40% are Black, 32% are unemployed, and about 40% have education beyond high school. Black and Latino patients are both more likely to be diagnosed with colorectal cancer at younger ages and more likely to be diagnosed at more advanced stages, Chino said, when cancers are harder to cure.

“It’s a group that we’re always trying to reach,” said Cynthia Mojica, a cancer researcher at Oregon State University who did not work on the study. In that sense, the study demonstrated how AI might be a tool that could improve engagement with vulnerable communities for cancer prevention — and possibly help close some of the health disparities between low-income communities and communities of color.

“It has a lot of potential. There are a lot of advantages to it,” said Mojica, who is also working on a similar project to use AI for patient outreach and navigation. “The idea is to decrease disparities using AI, but I think we also need to be careful about how we’re designing it.”

There are a few things that experts like Mojica and Chino hope to see next. For one, Mojica said, there’s a chance that some populations may be more comfortable engaging with AI than others, so it’ll be important to find ways to make sure AI doesn’t create tools that are biased towards helping only more privileged groups.

“It’s important to note this is a quality improvement project, not a randomized controlled trial,” Chino also pointed out. “The benefits really need to be continually reevaluated over time and rigorously tested in other populations.”

But assuming the benefits are real, the technology has a lot of potential, Chino said. And not just for colorectal cancer screening programs, said Montefiore Einstein’s Moadel-Robblee. Soon, they hope to roll out MyEleanor to other cancer prevention programs, including lung and breast cancer.