Author information

• ¹Virginia Commonwealth University, Richmond VA Medical Center, Richmond, Virginia. Electronic address: jasmohan.bajaj@vcuhealth.org.
• ²University of Alberta, Edmonton, Canada.
• ³Dallas VA Medical Center, Dallas, Texas.
• ⁴University of Pennsylvania, Philadelphia, Pennsylvania.
• ⁵Yale University Medical Center, New Haven, Connecticut.
• ⁶Mercy Medical Center, Baltimore, Maryland.
• ⁷University of California San Francisco, San Francisco, California.
• ⁸Emory University Medical Center, Atlanta, Georgia.
• ⁹Mayo Clinic Arizona, Phoenix, Arizona.
• ¹⁰University of Toronto, Toronto, Canada.
• ¹¹Virginia Commonwealth University, Richmond VA Medical Center, Richmond, Virginia.
• ¹²Mayo Clinic Rochester, Rochester, Minnesota.

Abstract

Background & aims: Grades 3 to 4 hepatic encephalopathy (advanced HE), also termed brain failure, is an organ failure that defines acute-on-chronic liver failure. It is associated with poor outcomes in cirrhosis but cannot be predicted accurately. We aimed to determine the admission metabolomic biomarkers able to predict the development of advanced HE with subsequent validation.

Methods: Prospective inpatient cirrhosis cohorts (multicenter and 2-center validation) without brain failure underwent admission serum collection and inpatient follow-up evaluation. Serum metabolomics were analyzed to predict brain failure on random forest analysis and logistic regression. A separate validation cohort also was recruited.

Results: The multicenter cohort included 602 patients, of whom 144 developed brain failure (105 only brain failure) 3 days after admission. Unadjusted random forest analysis showed that higher admission microbially derived metabolites and lower isoleucine, thyroxine, and lysophospholipids were associated with brain failure development (area under the curve, 0.87 all; 0.90 brain failure only). Logistic regression area under the curve with only clinical variables significantly improved with metabolites (0.65-0.75; P = .005). Four metabolites that significantly added to brain failure prediction were low thyroxine and maltose and high methyl-4-hydroxybenzoate sulfate and 3-4 dihydroxy butyrate. Thyroxine alone also significantly added to the model (P = .05). The validation cohort included 81 prospective patients, of whom 11 developed brain failure. Admission hospital laboratory thyroxine levels predicted brain failure development despite controlling for clinical variables with high specificity.

Conclusions: In a multicenter inpatient cohort, admission serum metabolites, including thyroxine, predicted advanced HE development independent of clinical factors. Admission low local laboratory thyroxine levels were validated as a predictor of advanced HE development in a separate cohort.