Abstract: TH-PO388
Improving Glomerular Filtration Rate Estimation by Semi-Supervised Learning: A Development and External Validation Study
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
Databases of AASK , CRIC and DCCT studies were pooled together for model development, whereas MDRD and CRISP studies for model external validation. The pooled development data set contained 2,719 participants, whereas the pooled external validation data set contained 1,952 participants. 4,829 participants only without GFR records but all other information available were pooled into an unlabeled data set for semi-supervised learning. New Predictors & Established Predictors:Serum creatinine, Age, Sex, Black race, Diabetes status, Hypertension and Body Mass Index. GFR measured as the urinary clearance of 125I-iothalamate. The revised CKD-EPI creatinine equations was selected as benchmark for performance comparisons. The proposed semi-supervised model was essentially an artificial neural network developed by Ladder Network algorithm. Head-to-head performance comparisons were conducted between revised equations and semi-supervised models from 4-variable to 7-variable.
Results
In each independent variables combination, the semi-supervised models consistently achieved superior results in all 3 performance indicators compared with corresponding revised CKD-EPI equations in the external validation data set. When selecting one representative revised equation and semi-supervised model for further comparison, compared with revised 4-variable CKD-EPI equation, the 7-variable semi-supervised model performed less biased (mean of difference: 0.03 [-0.28, 0.34] vs 1.53 [1.28, 1.85], P < 0.001), more precise (interquartile range of difference: 7.94 [7.37, 8.50] vs 8.28 [7.76, 8.83], P = 0.1) and accurate (P30: 88.9% [87.4%, 90.2%] vs 86.0% [84.4%, 87.4%], P < 0.001.
Conclusion
The superior performance of the semi-supervised models during head-to-head comparisons supported the hypothesis that semi-supervised learning technology could improve GFR estimation performance. The semi-supervised model still requires extra and careful validation, and further improvement is expected by integrating more cohort data.
Funding
- Government Support - Non-U.S.