Kidney Week

Abstract: FR-PO460

Developing Prognostic Models for Kidney Function Decline Based on Clinical and Metabolite Profiles in an Albuminuria-Stratified Analysis in the CRIC Cohort

Session Information

• Diabetic Kidney Disease: Clinical - I
  October 26, 2018 | Location: Exhibit Hall, San Diego Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Diabetic Kidney Disease

• 602 Diabetic Kidney Disease: Clinical

Authors

• Kwan, Brian, University of California, San Diego, La Jolla, California, United States

Brian Kwan,

• Fuhrer, Tobias, ETH Zurich, Zurich, Switzerland

Tobias Fuhrer,

• Zhang, Jing, University of California, San Diego, La Jolla, California, United States

Jing Zhang,

• Darshi, Manjula, University of Texas Health Science Center at San Antonio, San Antonio, Texas, United States

Manjula Darshi,

• Montemayor, Daniel, University of Texas Health Science Center at San Antonio, San Antonio, Texas, United States

Daniel Montemayor,

• Sharma, Kumar, University of Texas Health Science Center at San Antonio, San Antonio, Texas, United States

Kumar Sharma,

• Natarajan, Loki, University of California, San Diego, La Jolla, California, United States

Loki Natarajan,

Background

There is significant heterogeneity in the rate of kidney function decline among Type 2 diabetics, signaling a critical need to better identify patients at high risk of rapid kidney function decline. We previously identified a metabolomic signature of chronic kidney disease (CKD) using cross-sectional analysis. Metabolomic analysis combined with novel statistical methods could provide clinically useful signatures of CKD using longitudinal cohorts.

Methods

We studied 1003 Type 2 diabetics with up to 10 years of follow-up. The median eGFR decline per year (-1.84 ml/min/1.73m²/yr) cutoff delineated fast vs slow CKD progression. An untargeted flow-injection mass spectrometry method was used to assay an a priori 16-metabolite set, implicated in mitochondrial dysfunction in CKD. Models for fast vs slow decline were built with the 16-metabolite subset and standard clinical variables (e.g. age, race, HbA1c, MAP) as predictors. The accuracy (i.e. % of fast and slow decliners correctly identified by the model) of clinical-only, metabolite-only, and clinical-metabolite statistical models were compared using 5-fold cross-validation. Analyses were stratified by albumin/creatinine ratio (ACR): normal (ACR <30 mg/g), microalbuminuria (ACR 30-300 mg/g), macroalbuminuria (ACR >300 mg/g).

Results

There were significant differences in eGFR decline (p <0.001) between albuminuria groups: mean eGFR slopes (ml/min/1.73m²/yr) were -0.43 in normo-, -1.47 in micro-, and -3.03 in macro-ACR groups. As expected, study entry eGFR, diabetes control (HbA1c) and blood pressure (MAP) were worse (p <0.001) as albuminuria increased. Predictive accuracy of models was 84% normo-, 64% in micro-, 82% in macro-groups. Given the high accuracy, we further investigated the models for the normo- and macro-groups. After adjusting for clinical variables, prognostic metabolites were (i) 2-methyl acetoacetate for normo-ACR and (ii) pyruvic and homovanillic acids for the macro-ACR groups.

Conclusion

The accuracy of the models differed by ACR level. Several a priori metabolites predicted CKD progression, and interestingly, prognostic metabolites varied by ACR status. Our findings suggest that metabolites may offer insights into CKD progression in Type 2 diabetes within albuminuria groups.

Funding

• NIDDK Support