ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005


The Latest on Twitter

Kidney Week

Abstract: FR-OR023

Tissue Transcriptomic Profiles Perform Similarly to Clinical and Pathology Features for Nephrotic Syndrome Outcome Prediction

Session Information

Category: Glomerular

  • 1005 Clinical Glomerular Disorders


  • Mariani, Laura H., University of Michigan, Ann Arbor, Michigan, United States
  • Huang, Huateng, University of Michigan, Ann Arbor, Michigan, United States
  • Godfrey, Brad A., University of Michigan, Ann Arbor, Michigan, United States
  • Kretzler, Matthias, University of Michigan, Ann Arbor, Michigan, United States
  • Guan, Yuanfang, University of Michigan, Ann Arbor, Michigan, United States

Clinical parameters do not accurately predict outcomes in nephrotic syndrome (NS). Traditional statistics are not well equipped to identify predictors from many potential parameters across the genotype-phenotype continuum. Machine learning techniques have been developed to address this, but have not been widely applied to NS.


NEPTUNE is a cohort study of NS patients enrolled at the time of biopsy. Clinical data, pathologic features and kidney tissue genome wide mRNA expression levels are collected. Elastic net regularization was used to build Cox proportional hazards models for time to (1)composite of ESRD/40% eGFR decline and (2)complete remission (UPCR <0.3mg/mg) using different sets of predictors: clinical+pathology data, gene expression modules, all variables. In 200 bootstrap replicates, models were built in training sets and time-dependent area under the curve (tAUC) was computed in test sets. Paired t-test of mean tAUCs across replicates was used to compare prediction accuracy between models.


432 patients were in clinical/pathology models [mean age 33(21), eGFR 83(36), UPCR 3.7(5.5), 41% male, 27% black,18% MN, 31% MCD, 13% IgA, 38% FSGS]. 198 patients in gene expression models had similar characteristics. Elastic net models had higher tAUC than simple cox models(Table). Signifcant predictors are shown(Fig).


Machine learning elastic net models had highest accuracy and identified novel predictors. Tissue mRNA expression modules were more accurate predictors of composite outcome than routine clinical parameters and may better capture the underlying biologic heterogeneity of NS.

Model Comparison of Prediction Accuracy
CohortModel (p-value comparison is default model)tAUC Composite tAUC Complete Remission
Clinical CohortSimple Cox Model (Default; demographics, eGFR, UPCR)0.680.71
Elastic Net Clinical and Pathology0.76 (p=0.02)0.73 (p<0.01)
Gene Expression CohortElastic Net Clinical and Pathology (Default)0.730.71
Elastic Net Gene Expression Modules0.75 (p<0.01)0.70 (p=0.03)
Elastic Net Clinical, Pathology, Gene Expression Modules*0.77 (p<0.01)0.72 (p<0.01)

*Composite:14 clinical, 7 pathology, 14 modules; Remission: 27 clinical, 2 pathology, 24 modules


  • NIDDK Support