Kidney Week

Abstract: FR-PO937

Predicting Rapid eGFR Decline in the CURE-CKD Registry

Session Information

• CKD: Epidemiology, Risk Factors, Prevention - II
  November 04, 2022 | Location: Exhibit Hall, Orange County Convention Center‚ West Building
  Abstract Time: 10:00 AM - 12:00 PM

Category: CKD (Non-Dialysis)

• 2201 CKD (Non-Dialysis): Epidemiology‚ Risk Factors‚ and Prevention

Authors

• Davis, Tyler Austin, University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Tyler Austin Davis,

• Petousis, Panayiotis, University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Panayiotis Petousis,

• Zamanzadeh, Davina J., University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Davina J. Zamanzadeh,

• Norris, Keith C., University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Keith C. Norris,

• Duru, Obidiugwu, University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Obidiugwu Duru,

• Tuttle, Katherine R., Providence St Joseph Health, Spokane, Washington, United States

Katherine R. Tuttle,

• Bui, Alex, University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Alex Bui,

• Nicholas, Susanne B., University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Susanne B. Nicholas,

• Sarrafzadeh, Majid, University of California Los Angeles Department of Computer Science, Los Angeles, California, United States

Majid Sarrafzadeh,

Group or Team Name

• CURE-CKD

Background

Patients with rapid eGFR decline tend to progress to kidney failure. Automated tools can identify individuals at risk of severe kidney function decline and facilitate disease mitigation. We describe a machine learning model for predicting the risk of rapid eGFR decline (>40% over 2 years) and identify specific populations with elevated risk using the CURE-CKD Registry.

Methods

Variables include age, sex, race and ethnicity, ACE inhibitor/ARB, SGLT2 inhibitor, GLP-1 receptor agonist, NSAID, and PPI use, eGFR, systolic blood pressure (SBP), HbA1C, hypertension, type 2 diabetes, and chronic kidney disease (CKD) based on ICD-9/10 coding from patients with CKD (N=234,219) and at-risk for CKD (N=935,329) with CKD-EPI eGFR ≥15 ml/min/1.73 m² and 2 years of follow-up. We trained, tuned, and validated a gradient boosted tree ensemble model (GBTe) using a 60/20/20 train/validation/test split. We computed the risk distribution of all 8,503,055 subpopulations, based on all possible expert defined combinations of the above variables, and compared each risk distribution to the whole population’s risk distribution using the Kolmogorov-Smirnov (KS) test. Subgroups with the highest risk of eGFR decline were identified using the KS test (Holm-Bonferroni method with α=0.05).

Results

The GBTe model achieved an area under the precision-recall curve (PR-AUC) of 0.099 and an area under the receiver operating characteristic curve of 0.75 on the test set. 480,344 subpopulations were significantly above average predicted risk in the test set. We identified the most frequent predictors of rapid eGFR decline across the highest risk subpopulations. Of the top 100 significantly higher risk subpopulations the following variables are the most frequent: CKD (100%), PPI use (100%), SBP >140 mmHg (98%), HbA1C >8% (87%), and age 45-66 years (79%). Patients in these 100 subpopulations were 13.7 times more likely to experience rapid decline than the overall study population.

Conclusion

We developed a methodology that uses a risk model for rapid eGFR decline to identify subpopulations with significantly high risk for rapid eGFR decline. These subpopulations are strong candidates for closer study and early intervention.

Funding

• Other NIH Support