Abstract: PO2322
Three New Race-Free, Community-Based Equations to Estimate GFR: The Machine Learning Estimation of Renal Function (MLERF) Equations
Session Information
- Reassessing Race in Predicting Progression
November 04, 2021 | Location: On-Demand, Virtual Only
Abstract Time: 10:00 AM - 12:00 PM
Category: CKD (Non-Dialysis)
- 2101 CKD (Non-Dialysis): Epidemiology, Risk Factors, and Prevention
Authors
- Zhu, Xiaoqian, The University of Mississippi Medical Center, Jackson, Mississippi, United States
- Shafi, Tariq, The University of Mississippi Medical Center, Jackson, Mississippi, United States
- Norris, Keith C., University of California Los Angeles, Los Angeles, California, United States
- Simino, Jeannette, The University of Mississippi Medical Center, Jackson, Mississippi, United States
- Griswold, Michael, The University of Mississippi Medical Center, Jackson, Mississippi, United States
- Lirette, Seth, The University of Mississippi Medical Center, Jackson, Mississippi, United States
Background
As inclusion of race in glomerular filtration rate (GFR) estimation has become an increasingly controversial issue, it is of vital importance to propose race-free equations and evaluate their performance.
Methods
Using Multivariable Fractional Polynomials (MFP), Generalized Additive Models (GAM), and Random Forests (RF), we developed three new GFR estimating equations from the community-based Genetic Epidemiology Network of Arteriopathy Study (GENOA) study (N=1010). We then compared performance of the new equations to the CKD-EPI creatinine equation using the Epidemiology of Coronary Artery Calcification (ECAC) cohort study and the Assessing Long Term Outcomes in Living Kidney Donors (ALTOLD) (N=792). Due to lack of black participants in external data, we also evaluate performance of equations in Black participants internally using development data. A rigorous bootstrapping method, allowing equation coefficients to change for each bootstrap sample, was used to evaluate performance of our new equations to address the issue of overfitting.
Results
Our final equations were based on creatinine, age and sex. The addition of race yielded only minor nonsignificant improvements in RMSE and thus race was not included in the final equations. In external data (Figure), our new equations showed similar P30, RMSE, bias and precision compared to the CKD-EPI creatinine equation which included race as a predictor. Our equations also showed marked improvements in terms of bias and accuracy for Black participants over the CKD-EPI creatinine equation in the development data.
Conclusion
Performance of our new race-free equations using community-based cohorts were comparable to CKD-EPI creatine equation in external validation and superior in Black participants in internal validation. Our study indicated that race can be removed from equations to estimate GFR in Black and White participants without significantly sacrificing equation performance.
Performance of Equations