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: PO1017

Pre-Access Vein Transcriptomics as a Predictor of Arteriovenous Fistula Failure: A Machine Learning Approach

Session Information

Category: Dialysis

  • 703 Dialysis: Vascular Access

Authors

  • Rojas, Miguel G., University of Miami School of Medicine, Miami, Florida, United States
  • Martinez, Laisel, University of Miami School of Medicine, Miami, Florida, United States
  • Challa, Akshara Sree, University of Miami School of Medicine, Miami, Florida, United States
  • Tabbara, Marwan, University of Miami School of Medicine, Miami, Florida, United States
  • Duque, Juan Camilo, University of Miami School of Medicine, Miami, Florida, United States
  • Salman, Loay H., Albany Medical College, Albany, New York, United States
  • Vazquez-Padron, Roberto I., University of Miami School of Medicine, Miami, Florida, United States
Background

As the number of patients with end-stage renal disease continues to rise, the creation of a robust and efficient hemodialysis access is more important than ever. A mature arteriovenous fistula (AVF) is the preferred method for long-term hemodialysis. However, the nationwide maturation rate continues to be as low as 50-60%, and we currently lack an effective risk stratifying method to identify patients at higher risk of AVF failure.

Methods

To address this clinical need we developed a predictive model based on supervised machine learning from transcriptomics of the pre-access vein. Forty-eight pre-access veins obtained at the time of AVF creation (24 matured and 24 failed postoperatively) were randomly selected from the University of Miami Vascular Biorepository and submitted for bulk RNA sequencing. Both outcome groups were matched by age, sex, demographics, and baseline clinical characteristics. The highest expressing genes (normalized gene expression counts >200) were used as input in KNN, SVM, XGBoost, and other machine learning algorithms. Area under the curve (AUC) and receiver-operating characteristic (ROC) plots were used to compare the performance of the models relative to each other. The best performing algorithm, XGBoost, was optimized with the following hyperparameters {gamma=0.25, learning_rate=0.001, max_depth=4, reg_lambda=10, scale_pos_weight=3}. The SHapley Additive exPlanations (SHAP) analysis was then used to evaluate the highest contributing features to the XGBoost model.

Results

Ten highly predictive and abundantly expressed genes were identified using this methodology (RIC1, CLIC5, DNAL1, FOXO4, TIMMDC1, GALNT11, CDH13, KLHDC10, ZNF8, and DBT). Using these transcripts, the AUC in the logistic regression model is 97.6%.

Conclusion

In conclusion, this study has identified 10 potential pre-access gene predictors of postoperative AVF failure, which could be used clinically as a stratifying or risk management tool in vascular access patients.

Funding

  • NIDDK Support