Admission Urinary and Serum Metabolites Predict Renal Outcomes in Hospitalized Patients With Cirrhosis

Abstract

Background and aims: Acute kidney injury (AKI) has a poor prognosis in cirrhosis. Given the variability of creatinine, the prediction of AKI and dialysis by other markers is needed. The aim of this study is to determine the role of serum and urine metabolomics in the prediction of AKI and dialysis in an inpatient cirrhosis cohort.

Approach and results: Inpatients with cirrhosis from 11 North American Consortium of End-stage Liver Disease centers who provided admission serum/urine when they were AKI and dialysis-free were included. Analysis of covariance adjusted for demographics, infection, and cirrhosis severity was performed to identify metabolites that differed among patients (1) who developed AKI or not; (2) required dialysis or not; and/pr (3) within AKI subgroups who needed dialysis or not. We performed random forest and AUC analyses to identify specific metabolite(s) associated with outcomes. Logistic regression with clinical variables with/without metabolites was performed. A total of 602 patients gave serum (218 developed AKI, 80 needed dialysis) and 435 gave urine (164 developed AKI, 61 needed dialysis). For AKI prediction, clinical factor-adjusted AUC was 0.91 for serum and 0.88 for urine. Major metabolites such as uremic toxins (2,3-dihydroxy-5-methylthio-4-pentenoic acid [DMTPA], N2N2dimethylguanosine, uridine/pseudouridine) and tryptophan/tyrosine metabolites (kynunerate, 8-methoxykyunerate, quinolinate) were higher in patients who developed AKI. For dialysis prediction, clinical factor-adjusted AUC was 0.93 for serum and 0.91 for urine. Similar metabolites as AKI were altered here. For dialysis prediction in those with AKI, the AUC was 0.81 and 0.79 for serum/urine. Lower branched-chain amino-acid (BCAA) metabolites but higher cysteine, tryptophan, glutamate, and DMTPA were seen in patients with AKI needing dialysis. Serum/urine metabolites were additive to clinical variables for all outcomes.

Conclusions: Specific admission urinary and serum metabolites were significantly additive to clinical variables to predict AKI development and dialysis initiation in inpatients with cirrhosis. These observations can potentially facilitate earlier initiation of renoprotective measures.

FIG. 1.

Patient flow after entry into the study.

FIG. 2.

Random forest mean decrease accuracy for AKI development using serum and urine metabolites for the entire group. (A) Mean decrease accuracy on random forest for serum metabolites in patients who developed AKI. (B) Representative LS mean differences: AKI serum yes or no. (C) Mean decrease accuracy on random forest for urine metabolites in patients who developed AKI. (D) Representative LS mean differences: AKI urine yes or no.

FIG. 3.

Random forest mean decrease accuracy for dialysis (RRT) requirement using serum and urine metabolites for the entire group. (A) Mean decrease accuracy on random forest for serum metabolites in patients who required RRT. (B) Representative LS mean differences: serum yes or no. (C) Mean decrease accuracy on random forest for urine metabolites in patients who required RRT. (D) Representative LS mean differences: urine yes or no.

FIG. 4.

Random forest mean decrease accuracy for dialysis requirement using serum and urine metabolites for the subgroup with AKI. (A) Mean decrease accuracy on random forest for serum metabolites in patients who required dialysis (AKI-RRT) within the AKI group. (B) Representative LS mean differences: serum, yes or no. (C) Mean decrease accuracy on random forest for urine metabolites in patients who required dialysis (AKI-RRT) within the AKI group. (D) Representative LS mean differences: urine yes or no.