Selecting important predictors for arteriovenous fistula maturation in older hemodialysis patients by using random survival forests

Joyce Qian, Mara McAdams-DeMarco, Derek Ng, Bryan Lau

Research output: Contribution to journalArticlepeer-review

Abstract

Background: Placing an arteriovenous fistula (AVF) in older hemodialysis patients at great risk of primary failure leads to prolonged dependency on central venous catheter (CVC). A model which accurately predicts AVF maturation can facilitate better clinical determination for AVF placement. Methods: We assembled a retrospective cohort of 14 892 patients aged 67 years and older who started hemodialysis with a CVC between 7/1/2010 and 6/30/2012 and had a subsequent, incident AVF placement from the United States Renal Data System (USRDS). We used random survival forests (RSF) with competing risks to identify important predictors for AVF maturation. Results: Approximately 49.7% patients achieved AVF maturation and 13.6% had a competing event. The median time to maturation was 4 (IQR: 3-5) months. Patient's gender had the highest variable importance (VIMP, 0.0027), followed by race, being institutionalized, days on hemodialysis without an AVF, estimated glomerular filtration rate, and body mass index with borderline importance (VIMP ≥0.0005). The out-of-bag (OOB) error rate of the RSF was 45.3% and 45.8% for AVF maturation in the training and validation data sets, respectively. Conclusions: Predictors in USRDS data have limited ability to predict AVF maturation. Patient's gender might be considered as the most important predictor for AVF maturation.

Original languageEnglish (US)
Pages (from-to)148-155
Number of pages8
JournalSeminars in dialysis
Volume33
Issue number2
DOIs
StatePublished - Mar 1 2020

Keywords

  • arteriovenous fistula
  • elderly
  • hemodialysis
  • maturation
  • random forests

ASJC Scopus subject areas

  • Nephrology

Fingerprint

Dive into the research topics of 'Selecting important predictors for arteriovenous fistula maturation in older hemodialysis patients by using random survival forests'. Together they form a unique fingerprint.

Cite this