Abstract: TH-PO387
Study on Glomerular Filtration Rate Equation Using Ensemble Learning and Linear Regression
Session Information
- CKD: Risk Scores and Translational Epidemiology
November 07, 2019 | Location: Exhibit Hall, Walter E. Washington Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: CKD (Non-Dialysis)
- 2101 CKD (Non-Dialysis): Epidemiology, Risk Factors, and Prevention
Authors
- Liu, Xun, The Third Affiliated Hospital of Sun-Yat-Sen University, Guangzhou, GuangDong, China
- Lv, Linsheng, The Third Affiliated Hospital of Sun-Yat-Sen University, Guangzhou, GuangDong, China
- Ye, Yuqiu, Department of Nephrology, The Third Affiliated Hospital of Sun Yat-sen University, Guangdong, China, Guangzhou, China
- Li, Shaomin, The Third Affiliated Hospital of Sun-Yat-Sen University, Guangzhou, GuangDong, China
- Hu, Wentao, Department of Nephrology, The Third Affiliated Hospital of Sun Yat-sen University, Guangdong, China, Guangzhou, China
Background
Accurate estimating glomerular filtration rate (GFR) is crucial both in clinical practice and epidemiological survey. We incorporated semi-supervised learning technology to improve GFR estimation performance.
Methods
Data from patients with CKD and healthy people who were examined by radionuclide renal dynamic imaging or dual plasma DTPA in the Tianhe and Lingnan districts of The Third Affiliated Hospital of Sun Yat-sen University from January 2012 to April 2018 including baseline indicators, laboratory indicators, Kidney color Doppler ultrasound results, medication status. A total of 1,732 CKD patients and healthy people who had ECT examinations were included, of which 932 patients in Tianhe District used data as modeling data, and other 400 patients in Tianhe District as an internal validation data set, 400 patient data from Lingnan districts and outpatient department was used as the external validation data set. In the modeling group, the conventional data is modeled by linear regression and ensemble learning method, and then the verification group data is imported into the model to judge the prediction performance. We use internationally accepted bias, precision and accuracy comprehensively evaluate the predictive the model effectiveness and use the bootstrap method to calculate the 95% confidence interval and select the optimal model.
Results
A total of 1,732 CKD patients and healthy people were included in the study. The mean age was 57.15±13.56 years, and the mean GFR (mGFR) was 87.01±37.63 ml/min/1.73 m2. The newly revised CKD-EPI equation using smooth linearity (N-Spline technique) is superior to the equations modeled by traditional linear regression methods in terms of precision (p < 0.001). In terms of accuracy and precision, the XGboost model equation is superior to the linear regression equation of the same variable (p<0.05). Compared with the 4-variable revision CKD-EPI model, the 15-variable, 3-variable XGboost model improved in bias, but it was not statistically significant; it was better in accuracy and precision (P<0.05). Compared with the widely used eGFR model (4 and 6 variable MDRD models), the bias, accuracy and precision of the 4-variable revision CKD-EPI model were better (P<0.05).
Conclusion
The ensemble learning model can optimize the predictive performance of the GFR model.
Funding
- Government Support - Non-U.S.