Kidney Week

Abstract: SA-PO395

A Predictive Model of the Time to ESKD Using Machine Learning

Session Information

• Hemodialysis and Frequent Dialysis: CV and Risk Prediction
  November 05, 2022 | Location: Exhibit Hall, Orange County Convention Center‚ West Building
  Abstract Time: 10:00 AM - 12:00 PM

Category: Dialysis

• 701 Dialysis: Hemodialysis and Frequent Dialysis

Authors

• Okita, Jun, Department of Endocrinology, Metabolism, Rheumatology and Nephrology, Faculty of Medicine, Oita University, Yufu, Oita, Japan

Jun Okita,

• Nakata, Takeshi, Department of Endocrinology, Metabolism, Rheumatology and Nephrology, Faculty of Medicine, Oita University, Yufu, Oita, Japan

Takeshi Nakata,

• Noguchi, Emiko, Department of Endocrinology, Metabolism, Rheumatology and Nephrology, Faculty of Medicine, Oita University, Yufu, Oita, Japan

Emiko Noguchi,

• Koumatsu, Nobuchika, Department of Endocrinology, Metabolism, Rheumatology and Nephrology, Faculty of Medicine, Oita University, Yufu, Oita, Japan

Nobuchika Koumatsu,

• Tasaki, Ayako, Department of Endocrinology, Metabolism, Rheumatology and Nephrology, Faculty of Medicine, Oita University, Yufu, Oita, Japan

Ayako Tasaki,

• Uchida, Hiroki, Department of Endocrinology, Metabolism, Rheumatology and Nephrology, Faculty of Medicine, Oita University, Yufu, Oita, Japan

Hiroki Uchida,

• Kudo, Akiko, Department of Endocrinology, Metabolism, Rheumatology and Nephrology, Faculty of Medicine, Oita University, Yufu, Oita, Japan

Akiko Kudo,

• Fukuda, Akihiro, Department of Endocrinology, Metabolism, Rheumatology and Nephrology, Faculty of Medicine, Oita University, Yufu, Oita, Japan

Akihiro Fukuda,

• Shimomura, Tsuyoshi, Hospital informatics center, Oita Medical University Hospital, Yuhu, Oita, Japan

Tsuyoshi Shimomura,

• Tanigawa, Masato, Department of Biophysics, Faculty of Medicine, Oita University, Yuhu, Oita, Japan

Masato Tanigawa,

• Shibata, Hirotaka, Department of Endocrinology, Metabolism, Rheumatology and Nephrology, Faculty of Medicine, Oita University, Yufu, Oita, Japan

Hirotaka Shibata,

Background

Predicting the risk of end-stage kidney disease (ESKD) and the time to renal replacement therapy (RRT) is useful not only for healthcare providers to treat CKD and prepare for renal replacement therapy (RRT) but also for patients to plan their life. Many studies have assessed the risk for ESKD using machine learning, but none have focused on the time to RRT. In this study, we investigated how to predict the time to RRT using machine learning.

Methods

This study was a retrospective cohort study. Patients who started RRT between April 2016 and March 2021 at Oita University Hospital were enrolled. A total of 13,323 data groups were extracted from the electronic medical records, including 34 laboratory data items (e.g., BUN and creatinine [Cr]) and 8 patient background items (e.g., age and gender) from the start of follow-up to the start of RRT. Items with missing values of more than 30% were excluded. The residual 9,838 data groups without missing values were trained by the machine; 80% of the data were randomly divided for training and 20% for testing, and supervised learning was performed with multiple algorithms (mainly nonlinear regression models, such as Random Forest and gradient-boosted decision trees) to create predictive models and evaluate the accuracy of the test data.

Results

In all, 147 patients (99 males) were enrolled with a mean age of 60.8 years. The most common ESRD etiology was diabetic nephropathy (44%). The mean Cr was 7.6 ± 2.0 mg/dL, and the estimated glomerular filtration rate (eGFR) was 6.1 ± 1.7 ml/min/1.73 m² at induction. The optimal algorithm was the Random Forest regression, which created a predictive model that used six identical time variables (age, gender, height, weight, Cr, and eGFR). The model was highly accurate with a coefficient of determination (R²) of 0.96 and a mean absolute error of 159 days. The values predicted from the test data strongly correlated with the true values.

Conclusion

We predicted the time to start RRT with high accuracy using machine learning. Predicting the time to start RRT is useful for CKD treatment, and we plan to improve the accuracy of the model and consider its application to actual clinical practice in the future.