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: SA-OR066

Development of an Automatic Risk-Prediction System for Hemodialysis Patients Using Artificial Intelligence: A Nationwide Dialysis Cohort Study in Japan

Session Information

  • Hemodialysis Potpourri
    November 09, 2019 | Location: 144, Walter E. Washington Convention Center
    Abstract Time: 06:06 PM - 06:18 PM

Category: Dialysis

  • 701 Dialysis: Hemodialysis and Frequent Dialysis

Authors

  • Kanda, Eiichiro, Kawasaki Medical School, Kurashiki, Okayama, Japan
  • Tsuruta, Yuki, Tsuruta Itabashi Clinic, Tokyo, Japan
  • Kikuchi, Kan, Shimoochiai Clinic, Tokyo, Japan
  • Kashihara, Naoki, Kawasaki Medical School, Kurashiki, Okayama, Japan
  • Abe, Masanori, Nihon University School of Medicine, Tokyo, Japan
  • Masakane, Ikuto, Honcho-Yabuki Clinic, Yamagata, Japan
  • Nitta, Kosaku, Tokyo Women's Medical University, Shinjuku-ku, TOKYO, Japan
Background

Dialysis patients are at high risks of death and cardiovascular disease. An accurate prediction of these risks at an individual level is required to improve the prognosis of dialysis patients. In this study, we developed a new system for predicting five-year death using machine learning and big data from a nationwide prospective cohort study of the Japanese Society for Dialysis Therapy Renal Data Registry.

Methods

We categorized hemodialysis patients in Japan into new clusters generated by k-means clustering method. The associations between clusters and an outcome (death) in five years were evaluated using multivariate Cox proportional hazards models. Then, the accuracy of the prediction of five-year mortality was compared among the machine learning models.

Results

Among the hemodialysis patients (n=78,854); average age, 65.7±12.2 years; male, 61.5%; and diabetes mellitus, 32.8%. The k-means clustering method using baseline characteristics in the training dataset automatically generated three new groups. Hazard ratios of the high- and middle-risk groups were 45.2 (95% confidence intervals 40.8, 50.0) and 4.8 (4.4, 5.4), respectively. In the test dataset, the accuracy of the cluster was 0.79, which was higher than those of multivariate logistic regression models including baseline characteristics (0.76). The accuracies of deep learning and support vector machine (SVM) models including baseline characteristics were 0.88 and 0.91, respectively. Moreover, that of the SVM model with the cluster was improved to 0.93.

Conclusion

It was found that artificial intelligence can categorize hemodialysis patients on the basis of their characteristics, which reflects their prognosis. This system is useful for identifying high-risk patients at an early stage.