Abstract: FR-PO0011
Machine Learning-Based Prediction of Rapid Kidney Function Decline Using Clinical and Socioenvironmental Data: Analysis from a Large Health System
Session Information
- Artificial Intelligence and Digital Health at the Bedside
November 07, 2025 | Location: Exhibit Hall, 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
- Al Awadhi, Solaf, Houston Methodist Academic Institute, Houston, Texas, United States
- Casarin, Stefano, Houston Methodist Academic Institute, Houston, Texas, United States
- Al-Kindi, Sadeer, Houston Methodist, Houston, Texas, United States
- Waterman, Amy D., Houston Methodist Academic Institute, Houston, Texas, United States
Group or Team Name
- TeXas Cardiometabolic and Kidney Disparities (Tx-CKD) Consortium.
Background
Chronic kidney disease (CKD) progression is highly variable and influenced by both clinical and environmental factors. Early identification of individuals at risk for rapid kidney function decline is critical to guide proactive interventions.
Methods
We identified 10,186 adults with CKD stages 2–4 from the Houston Methodist health system and calculated annualized estimated glomerular filtration rate (eGFR) slopes using serial laboratory data. Rapid progression was defined as an eGFR decline of ≥5 mL/min/1.73m2 per year. Predictor variables included demographic, clinical, and geocoded census tract–level socioenvironmental data (191 Climate Vulnerability Index). An extreme gradient boosting (XGBoost) model was trained on 60% of the cohort and validated on 40%.
Results
The cohort had a mean age of 70.6 years; 61% were female, 22.8% Black, and 7.1% Hispanic. Comorbidities included hypertension (94.2%), diabetes (50.9%), and history of smoking (41.4%). A total of 1,210 patients (11.9%) were identified as fast progressors. Top clinical and demographic predictive features included heart failure, baseline systolic blood pressure (SBP), baseline eGFR and age. Top socioenvironmental factors included presence of hazardous waste management facilities and expected annual building loss (Figure 1). The model achieved an area under the receiver operating characteristic curve of 0.682 and showed good calibration (Figure 2).
Conclusion
A machine learning model integrating clinical and socioenvironmental data predicted rapid kidney function decline with good performance, highlighting the role of neighborhood context and the utility of predictive tools for targeted interventions.
Funding
- Other NIH Support