ASN's Mission

ASN leads the fight to prevent, treat, and cure kidney diseases throughout the world by educating health professionals and scientists, advancing research and innovation, communicating new knowledge, and advocating for the highest quality care for patients.

learn more

Contact ASN

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


The Latest on Twitter

Kidney Week

Abstract: TH-PO758

Predicting Pediatric CKD Progression with Urinary Metabolomics

Session Information

  • Pediatric CKD
    November 07, 2019 | Location: Exhibit Hall, Walter E. Washington Convention Center
    Abstract Time: 10:00 AM - 12:00 PM

Category: Pediatric Nephrology

  • 1700 Pediatric Nephrology


  • Blydt-Hansen, Tom D., University of British Columbia, Vancouver, British Columbia, Canada
  • Sharma, Atul Kumar, University of Manitoba, Winnipeg, Manitoba, Canada
  • Mak, Robert H., UCSD, La Jolla, California, United States
  • Schwartz, George J., University of Rochester, Rochester, New York, United States
  • Warady, Bradley A., Children's Mercy Kansas City, Kansas City, Missouri, United States
  • Wishart, David, University of Alberta, Edmonton, Alberta, Canada
  • Furth, Susan L., The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, United States

Group or Team Name

  • Chronic Kidney Disease in Children (CKiD) Study

Predicting progression in children with chronic kidney disease (CKD) may improve care and planning for renal replacement therapy. Adding urinary metabolome changes to existing clinical models may improve prognostication.


Urine samples from patients in the CKiD Study had targeted urinary profiling (138 metabolites/creatinine ratio). Time to event (TTE) was measured, with event defined as 50% decline in eGFR, eGFR <15 or start of renal replacement (composite). A TTE predictor was trained using partial least squares discriminant analysis (PLS-DA), reported as discriminant score (dscore). Log-logistic accelerated failure time (AFT) models were fitted to predict survival based on clinical features without/with dscore, and compared the mean bias for accuracy (t test) and precision (F test).


703 patients (61% male, aged 12.1±4.4 years, iGFR 51.7±23.9) were divided into 2/3 training, 1/3 test sets, and 222 had an event (cases). The dscore was trained on cases (2 PLS components) and validated in the test set. The dscore was significantly correlated with TTE (r=0.61, p<0.001). The model performed similarly using only the top 10 metabolites (r=0.60, p<0.001). No improvement was noted when training separate glomerular vs. non-glomerular strata. An AFT survival model fit to training data (N =469 with 148 cases) included GFR, proteinuria, and glomerular disease as predictors. The model including dscore & clinical features (AIC=756) had improved fit compared to clinical features alone (AIC=793, likelihood ratio test p<0.001). Using test cases (N=74), the mean bias (years) was 1.8±3.1 vs. 2.4±4.3, demonstrating superior accuracy (mean difference 0.6, p=0.04) and precision (F ratio=0.53, p=0.007). Empiric and predicted AFT survival curves were compared using test data (N=234, 74 cases) stratified by baseline GFR <60. The predicted survival curves ± dscore are shown for both strata (Figure).


The addition of a urinary metabolite classifier improves both predictive accuracy and precision of a CKD progression model, compared with clinical features alone.


  • NIDDK Support