Chart notes, natural language processing can help spot a serious brain issue tied to liver disease
Researchers at VCU and the Richmond VA Medical Center have found that a few words in a medical chart can help flag a serious mental condition linked to liver disease.
The condition is called hepatic encephalopathy, or HE. It affects how the brain works when the liver is too sick to clean toxins from the blood. People with cirrhosis and HE may get confused, lose focus or act unlike themselves, as though they have brain fog. It can often lead to long hospital stays and many return visits.
Why this is important
Even though HE can be life-changing, it does not play a direct role when doctors decide who gets a liver transplant. One reason is that HE can be hard to pin down. Doctors may describe symptoms in different ways.
To improve the objectivity of an HE diagnosis, researchers at VCU Health and the VA turned to natural language processing, or NLP, a kind of AI computer tool that can read thousands of notes far faster than any human. Their results were published by the American Journal of Gastroenterology.
“HE puts a heavy load on patients and the people who care for them,” said the paper’s corresponding author, Jasmohan Bajaj, M.D. “We need a fair and simple way to spot it. Using plain words already written by clinicians in the chart is a promising path.”
What the team did
The team reviewed more than 4,000 medical notes about 432 veterans in hospital with cirrhosis and used a computer to look for words in the notes related to HE. Doctors also reviewed each chart to confirm who truly had the condition.
After sorting through everything, they found five terms that frequently appeared in the notes:
- asterixis, a hand tremor
- altered mental status
- confusion
- starting lactulose or rifaximin, two drugs that treat HE
- continuing lactulose or rifaximin
If at least one of these five terms showed up in a patient’s chart, that patient was very likely experiencing HE. If none showed up, the patient probably was not. This quick rule of thumb worked well, catching most cases and rarely missing them.
Testing it again and again
To make sure the findings were solid, the team tested the five terms with two more groups of patients. One group came from the VA through the NACSELD research network. The other came from VCU Health through their NACSELD patients.
In the VA group, the five-term list caught every HE case. In the VCU group, it was still very accurate, demonstrating that this approach can work for different hospitals and different patient populations.
The team also used machine learning to check how much each term added to the diagnosis. The results matched what they saw before. The same terms showed up as the biggest clues, and using them together worked better than using any one term alone.
Why this matters
These findings could help hospitals track HE in a more consistent way. That could improve patient care, support national research studies, and guide future updates to liver transplant policy. This is because liver transplant priority points, which are sorely needed to ensure patients with HE have a fair representation, are difficult to operationalize with the current subjective environment.
The five key terms also help separate HE from other symptoms that might look similar. Feelings like anxiety, tiredness, or occasional tremors are common in cirrhosis, but they do not reliably point to HE. The five-term list offers better clarity.
In addition to transplant, drugs or trials that use HE occurrence as a clinical endpoint suffer from this subjectivity also. This is because the majority of HE events occur in hospitals far away from the trial sites and a lot of resources are spent judging whether the research volunteer had developed HE or not. This could also give national clinical trials a shared, dependable way to measure HE across many sites.
Looking ahead
The team hopes physicians could someday use this method in real time to help determine when a patient suffers from HE.
“This study shows how combining clinical insight with technology can make care better,” the team said. “Better tools mean better support for people living with advanced liver disease.”
Authors of the article also included Scott Silvey and Brian J. Bush of VCU’s Department of Population Health; and Nilang Patel, M.D., Brian C. Davis, M.D., Richard K. Sterling, M.D., Anas Aljabi, M.D., Abhishek Shenoy, M.D., Puneet Puri, M.D., and Michael Fuchs, M.D.