Seattle Children’s Hospital may soon pilot AI language translation

Having a sick child is hard enough, but there’s an extra layer of difficulty when you’re not totally fluent in English and are trying to navigate an English-based health system.

That’s the reality for millions in this country. An estimated 8.2% of the US speaks English “less than very well" according to the US Census Bureau's American Community survey. And a 2024 review from researchers at the University of Pennsylvania School of Nursing found that people who don’t speak English very well have lower satisfaction rates and worse health outcomes, including more hospital readmissions and longer stays.

Doctors at Seattle Children’s Hospital hope to soon pilot an artificial intelligence-powered translator that may make things a little less stressful for non-English speakers.

The basics. Hospitals that receive federal funding are required to provide “meaningful” translation and interpretation services when needed, which can include free on-site interpreters and translators, qualified bilingual staff, or remote interpreters via phone or video, according to the Centers for Medicare and Medicaid Services.

Still, three in 10 adults who don’t speak English “very well” said they had trouble understanding their healthcare provider’s instructions or communicating with medical staff, according to a KFF survey released in May.

Searching for solutions. While the “gold standard” for spoken or signed communication is still a human interpreter, AI has the potential to streamline written communication between doctors and patients, according to Casey Lion, research director, Center for Diversity and Health Equity, Seattle Children’s Hospital.

For instance, at Seattle Children’s Hospital, there typically isn’t time to get discharge papers translated into a patient’s preferred language before they depart, Lion told Healthcare Brew, so instead they’re sent through the mail.

“Those are arriving days later, when a lot of the information in there about how to take care of the child at home, what signs to look for, to come back to the hospital for, may already be irrelevant,” she said. “They’ve already gotten through the scariest, most critical period.”

That’s why the hospital is testing an AI-powered tool to translate English-language clinical documents like discharge papers into Spanish, Somali, Vietnamese, and simplified Chinese—the most common languages of its patients, Lion said.

Within the next year, the system hopes to pilot test the tech with clinical documents given to actual patients, Lion said, though human translators will still double-check the results.

Let’s talk ethics. But just because it’s possible doesn’t mean it’s practical.

Seattle Children’s has taken a number of precautions, laid out in a September JAMA article, that Lion hopes other systems will also take if they follow suit.

One of the precautions is to keep everything internal. The hospital built AI language translation tech into its own data system so third parties aren’t given access to sensitive patient information, which could happen if the hospital used an outside system, like generative AI-based models, Lion said.

It’s also important to ensure that this AI tool works with every language in order to avoid exacerbating disparities, Lion said. AI has traditionally been trained on internet data that is mostly in languages like English and Spanish, and it’s not as good at translating less common languages, the New York Times reported in July.

“If we can’t do the high-tech solution well in all the languages, we need to continue investing resources in the low-tech solutions that are more accessible,” Lion said.

What’s more, Seattle Children’s created a dedicated AI ethics review board to monitor the use of the tech in the hospital, she added.

“Seattle Children’s has been really thoughtful and intentional about [its] approach, and my guess is that this is going to be adopted pretty routinely by other systems,” she said.