ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

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

email@asn-online.org

202-640-4660

The Latest on X

Kidney Week

Please note that you are viewing an archived section from 2020 and some content may be unavailable. To unlock all content for 2020, please visit the archives.

Abstract: PO0527

Machine Learning Prediction of ESKD and Death in CKD Patients: Electronic Medical Record-Based Cohort Study

Session Information

Category: CKD (Non-Dialysis)

  • 2101 CKD (Non-Dialysis): Epidemiology, Risk Factors, and Prevention

Authors

  • Kanda, Eiichiro, Kawasaki Medical School, Kurashiki, Okayama, Japan
  • Tokuyama, Atsuyuki, Kawasaki Medical School, Kurashiki, Okayama, Japan
  • Itano, Seiji, Kawasaki Medical School, Kurashiki, Okayama, Japan
  • Nagasu, Hajime, Kawasaki Medical School, Kurashiki, Okayama, Japan
  • Kashihara, Naoki, Kawasaki Medical School, Kurashiki, Okayama, Japan
Background

Chronic kidney disease (CKD) is a risk factor for end-stage kidney disease (ESKD) and death. An accurate prediction of these risks is required to improve their prognosis. We developed the new machine learning models for the prediction of CKD progression and death using the electronic-medical-record-based CKD-patient big database in Japan (n=3,714, 66,981 claim data).

Methods

We developed 20 risk prediction models with 45 variables for the classification of the patients on the basis of their outcomes (ESKD and death) over 1 to 3 years using random forest (RF), Gradient Boosting Decision Tree (GB), eXtreme Gradient Boosting (XG), support vector machine, and multivariate logistic regression models using validation datasets including baseline or time-series datasets. The performance characteristics of the models were compared with those of the laboratory indices, and the kidney failure risk equation (KFRE) using the area under the prediction curves (AUCs) by bootstrapping 1000 times.

Results

53.1% were male: age, 60.1±17.6 years; eGFR, 54.2±30.7 ml/min/1.73 m2; diabetes mellitus, 23.1%. In the validation dataset, 14 models showed statistically significantly higher AUCs for the prediction of outcomes than KFRE 0.782 (0.682, 0.881), and the RF, GB, and XG models based on time-series data showed the highest AUCs: 0.924 (95% CI 0.895, 0.953) (Fig. A). These three models also demonstrated the highest performance in the subgroup analysis that considers eGFR, DM, gender, and age. Moreover, the models’ sensitivities were 0.971 (95% CI 0.914, 1.0). Cox proportional hazards models revealed that the probabilities predicted by these models represented the risk of the outcome (p<0.0001) (Fig. B).

Conclusion

The machine learning models exhibit better performance than pre-existing models in identifying patients at an increased risk of CKD progression and death. Thy will enable us to implement effective measures to improve patient’s prognosis.

Funding

  • Government Support - Non-U.S.