Abstract: SA-PO0087
Artificial Intelligence Approach to Understanding CKD Progression
Session Information
- Artificial Intelligence: Kidney Imaging, Pathology, and Precision Phenotyping
October 24, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Gaweda, Adam E., University of Louisville, Louisville, Kentucky, United States
- Nouman, Muhammad Khuram, University of Louisville, Louisville, Kentucky, United States
- Nayak, Vibha S., University of Louisville, Louisville, Kentucky, United States
- Ouseph, Rosemary, University of Louisville, Louisville, Kentucky, United States
Background
Diabetes and hypertension are the two main risk factors behind chronic kidney disease (CKD). We present a state-of-the-art Artificial Intelligence (AI) approach that reveals how diabetes, hypertension, and the related treatments affect the progression of CKD.
Methods
We abstracted the following data of 960 subjects with 11 or more eGFR measurements from Chronic Renal Insufficiency Cohort (CRIC): eGFR (CKD-EPI), age, BMI, smoking status, hypertension, proteinuria, diabetes, and the use of antidiabetic and anti-hypertensive agents. For each subject we estimated their eGFR trajectory over time (GFRslope). We trained an Autoencoder (AE) neural network using these data and performed clustering in the AE internal representation space. From each cluster, we estimated mean values (for continuous parameters) and percentages (for nominal parameters), as well as partial correlations to create networks representing interactions between parameters.
Results
Table 1 shows estimated cluster parameters in a 4 cluster setup. Figure 1 shows partial correlation networks representing all revelant interactions between eGFR trajectory and the parameters of interest. Stable eGFR trajectory is associated with highest eGFR level, lowest prevalence of hypertension and diabetes, and the lowest level of proteinuria. Partial correlation networks show that eGFR trajectory is most closely related to eGFR level, age, diabetes, and diabetes treatment.
Conclusion
Slowing down the progression of CKD is more likely at higher GFR levels. Diabetes has a stronger effect on CKD progression than hypertension. Interventions in early CKD stages, primarily targeting diabetes, may be especially important for preventing kidney function decline.
Acknowledgment
CRIC data were provided by NIDDK Central Repository, a program of the National Institute of Diabetes and Digestive and Kidney Diseases
Cluster parameters. Legend: HTN – hypertension, DIAB – diabetes, PROT – proteinuria, SMK – smoking status, ACE – ACE inhibitor, BBK – beta blocker, ADB – antidiabetic agent.
Partial correlation network showing interactions between parameters within each cluster.