Teaching EHR to Mimic Liver Specialist Expertise

Whenever a story on fatty liver disease appears in the media, Columbia hepatologist Julia Wattacheril does what most people try to avoid: she reads all the comments.

"I always learn a lot from what the public asks in the comment sections in public health pieces, it really reveals the gaps in communication," says Wattacheril, who directs a program for people with the disease (officially called MASLD, for metabolic dysfunction-associated steatotic liver disease) at Columbia University Vagelos College of Physicians and Surgeons. "It's the most common liver disease in the world, but it's still surprising how underrecognized this disease is among people on the street, patients, and even doctors."

Until recently, MASLD was called nonalcoholic fatty liver disease to differentiate the condition from fatty liver disease caused by heavy intake of alcohol and a host of other causes of fat in the liver. In MASLD, fat progressively builds up in the liver, sometimes triggering inflammation and scarring that over the years impairs the liver's functions. It is believed to affect about 30-40% of U.S. adults and up to 45% of the global population.

"In media stories about MASLD, the most common comment I read is: 'If 30% or 40% of the population has this disease, how do I know? What do I do?'" Wattacheril says.

"It's still surprising how underrecognized this disease is among people on the street, patients, and even doctors."

Part of the issue is that patients in the early and even some more advanced stages of the condition often have no symptoms. The finding can also fall down the list of competing issues to tackle in a provider visit.

"Expertise is limited, and I'm fortunate if a patient is referred to me after someone recognizes the risk factors and does a test," says Wattacheril, who specializes in treating patients with MASLD and the more severe form of the disease, MASH (metabolic dysfunction-associated steatohepatitis), through to cirrhosis and those requiring transplant. "Tons of patients make their way to us too late and already have cirrhosis of the liver-which is about to crescendo as the leading indication for transplantation. We could do a lot better at identifying and stratifying them earlier. But it does require a proactive approach towards health. Over the past years, it's often that patients recognize the finding in their report and ask their doctor to refer. And sometimes we find they are even beyond cirrhosis-so we shift the conversation and evaluation towards transplantation."

Mining electronic health records

A new technology Wattacheril is developing may do a better job of identifying patients by mining millions of data points within electronic health records. Her machine-based approach-recently run on the EHR system at Columbia University-identified approximately 16,000 potential MASLD patients, including two-thirds whose electronic records had no indication of a MASLD diagnosis. The findings-including validating results from two other medical centers-were published in January in the journal Clinical and Translational Science.

"We have basically taught a computer to think like a hepatologist...[we're] flagging patients who have scar tissue and don't know it, patients who may have the disease and don't know it."

"Our machine-based algorithm is flagging patients who have scar tissue and don't know it, patients who may have the disease and don't know it, and patients with atypical (or no) risk factors that put them into a high-risk category," Wattacheril says.

The technique does not currently rely on machine learning or AI but instead works like an automated physician by using natural language processing to read lab data, clinical notes, and radiology reports in a fraction of the time.

"We have basically taught a computer to think like a hepatologist," Wattacheril says. "We can now leverage expertise at a systems level. It's raising the antenna on patients who have risk factors but whose diagnosis remains hidden to provider recognition, often buried in an enormous chart that's hard to find in a time- and resource-limited practice."

Future iterations of the technology may incorporate machine learning and AI, but only after the researchers ensure that the data used to train an AI system won't cause any errors or lead to unanticipated problems. Wattacheril's team is about to start their first in-human trial to validate the software's diagnoses and stratifications in conjunction with regulatory guidance. In the next few months, Wattacheril's team will start reaching out to doctors and nurses whose patients were identified as potentially having MASLD with evidence of moderate fibrosis by the software to determine if they'll refer them for research-related testing and confirmation of computer findings.

A different era of therapeutics

MASLD patients have more options today than just a few years ago, which makes identifying them and accurately staging their disease key. Patients may benefit from several interventions, including lifestyle changes, new GLP-1 weight loss drugs, and drugs approved just last year that can reduce scarring in some patients with MASH and fibrosis, the advanced form of the disease. "We are in a different era of therapeutics, noninvasive tests, clinical trials, and public frustration with the status quo," Wattacheril says. "Therefore, we need to meet that new era with new technologies such as software as a medical device."

"The essence of our algorithm is how it makes precision medicine possible, by filtering patients into different categories of MASLD," Wattacheril says. "A lot of people will benefit from new GLP-1 drugs to reduce weight; but other patients have less-common forms of MASLD (like those with normal weight) and will need other approaches, including genomic testing to identify rare diseases like familial lipid disorders."

"If they're willing, we'll test patients for MASLD with non-invasive tests, determine if they really have the disease identified by the software, and if so, connect them to clinical care and other research studies if they are interested. Then we'll know if our technology is ready to be deployed for multisite validation. The timing could not be better."

Julia Wattacheril, MD, MPH, is an associate professor in the Department of Medicine (Division of Digestive and Liver Diseases) at Columbia University Vagelos College of Physicians and Surgeons.