SAN DIEGO—An artificial intelligence machine learning algorithm applied to imaging data successfully identified patients with metabolic dysfunction–associated steatotic liver disease who likely would have otherwise been missed, according to new data presented at TLM 2024.
Related Content
Traditional methods to identify MASLD rely on ICD-10 codes or other proxies, according to lead researcher Ariana Stuart, MD. “It means that someone has recognized [MASLD] and designated the diagnosis, or it relies on a lot of factors that are more discrete. There’s a lot of narrative information that is put in notes related to diagnoses, and a lot of that gets missed in traditional chart pulls or identification,” said Dr. Stuart, an internal medicine resident at the University of Washington, in Seattle.
The researchers created an iterative natural language processing AI algorithm that used the American Association for the Study of Liver Disease MASLD criteria to analyze hepatic steatosis in radiology reports drawn from multiple centers in the Seattle area between Dec. 1, 2023, and May 31, 2024 (abstract 2360). The researchers excluded infectious, autoimmune and genetic causes of liver disease, as well as alcohol-associated liver disease.
The algorithm identified 957 patients with imaging findings that were consistent with MASLD. Two physicians manually reviewed medical records to validate the cohort against AASLD MASLD criteria. After multiple rounds of iterations, the algorithm achieved an accuracy of more than 88%, according to Dr. Stuart.
Improving MASLD Identification
The mean age of the identified cohort was 51 years, and 44% were female. Of these patients, only 17% had a MASLD ICD diagnosis and 26% had had contact with a gastroenterologist or hepatologist.
The low percentage of the cohort with an ICD-10 diagnosis of MASLD is a significant finding, “especially for a disease where time to initiation of management really affects disease progression and outcome,” Dr. Stuart said.
The group is now testing the algorithm against a much larger cohort spanning several years. “Then we’re also planning on implementing a quality improvement initiative aimed at increasing awareness for primary care providers, specifically on what to do when your patient has an incidental reading of hepatic steatosis,” Dr. Stuart said.
Long term, she anticipates that AI could automatically flag patients with MASLD to undergo further testing and management. It could also boost research and quality improvement efforts by reducing the need for chart review, she said, which is a tedious process that can be a drag on research.
“It’s an opportunity,” she said, “for machine learning artificial intelligence to help relieve the burden and do the recognition.”
Dr. Stuart reported no relevant financial disclosures.