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

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: PO0815

Predicting Decline in Residual Renal Urea Clearance via Random Forest Regression

Session Information

Category: Dialysis

  • 701 Dialysis: Hemodialysis and Frequent Dialysis

Authors

  • Berkowitz, Jacob S., University of Pittsburgh, Pittsburgh, Pennsylvania, United States
  • Akbilgic, Oguz, Loyola University Chicago, Chicago, Illinois, United States
  • Kalantar-Zadeh, Kamyar, University of California Irvine, Irvine, California, United States
  • Streja, Elani, University of California Irvine, Irvine, California, United States
Background

For incident hemodialysis patients, the decline of residual kidney function (KRU) over the first year of dialysis therapy is associated with adverse outcomes such as higher death risk. While several studies have identified biomarkers associated with higher risk of steeper KRU decline, we sought to employ more novel prediction utilizing a random forest regressor to identify important predictors of KRU after one year on hemodialysis.

Methods

We retrospectively reviewed a cohort of 5,141 patients who initiated in-center hemodialysis from 2007 to 2011 and had available KRU data at both baseline and during the 90 days after the one-year mark. 80% of the cohort was selected for the training dataset, with the remaining 20% used to test the model. Cross validation was utilized to optimize the number of trees and the mtry parameter. For feature selection, we used the 20 most important features from a random forest using all available predictors.

Results

In our cohort, mean age was 61 ± 14 years, with 66% men, 25% Black, 70% diabetes, and mean baseline albumin was 3.62 ± 0.42 g/dL. Median baseline KRU was 4.24 (6.39 – 2.69). Median KRU after one year was 1.74 (3.14-0.76). The random forest model yielded an overall mean squared error of 2.13 with noticeably stronger performance on the lower end of final KRU values. Using the median response as a classification threshold, the model achieved an AUC of 0.74. A variable importance analysis revealed that the model’s five most important predictors consist of baseline KRU, albumin, weight post treatment session, blood urea nitrogen level, and body mass index.

Conclusion

We showed that a random forest regressor can predict KRU values for hemodialysis patients after one year of treatment with moderately high accuracy. Utilizing our predictive models could aid patients and clinicians in determining the best course of treatment, which should be validated in future studies.