NLP medical review may help identify adults with undiagnosed PFIC

The study findings were reported in a poster presented at AASLD The Liver Meeting in Washington, DC, USA by Seema Meloni (Ipsen, Cambridge, Massachusetts, USA) and co-workers who explain that the diagnosis and treatment of late-onset PFIC is a “major challenge” due to its “heterogenous presentation,” with 13 known types and varied clinical phenotypes.

The team searched for patients whose symptoms fit with a potential diagnosis of late-onset PFIC in de-identified EHRs from the Mayo Clinic and Duke Health Systems between 2005 and 2024.

They identified 119 adults with one or more relevant ICD codes for specified disorders of the biliary tract and other liver diseases in their EHRs (structured data, n=2), but also used NLP to screen clinical notes for diagnosis (unstructured data, n=106), and/or genetic test results indicating a relevant mutation (structured or unstructured data, n=15).

The medical records of those with a tentative diagnosis of late-onset PFIC were then manually reviewed by experts and the diagnosis was confirmed in 14 patients, the team reported.

This included 13 patients at the Mayo Clinic, aged an average of 23.0 years, of whom 69.2% were women and 53.8% were White, and one man of unknown race at Duke Health who was aged 24 years. These patients were followed up for a median of 244 months and 93 months, respectively.

Analysis showed that the most common clinical features associated with late-onset PFIC were cholestasis (78.6%), cirrhosis (35.7%), pruritus (28.6%), and jaundice (14.3%). In addition, 14.3% of the patients had end-stage liver disease and/or liver decompensation, 7.1% had splenomegaly, and 7.1% portal hypertension.

Cholestasis was the first clinical characteristic to present in the patients, followed by pruritus, the researchers say.

“To our knowledge, this is the first study to leverage EHR data from 2 large medical institutions and to use NLP technology to identify late-onset PFIC,” Meloni et al comment.

Noting that a prior study of 356 patients also found that cholestasis, pruritus, jaundice, and end-stage liver disease were common diagnoses in patients with adult-onset cholestasis, the team concludes that assessing EHRs “may aid earlier diagnoses” and allow “earlier intervention” which could, in turn, “improve disease management, and ultimately, patient outcomes.”