Presentation (CTI)

OBJECTIVES: Chronic Kidney Disease (CKD) often progresses silently until advanced stages, leading to higher morbidity, costly interventions, and long‑term health system burden. Early identification of high‑risk individuals with obesity is essential for prevention and targeted care. This study evaluated machine‑learning (ML) models to predict CKD onset and stage transitions using real‑world US data.
METHODS: A retrospective cohort was created using Optum® Market Clarity data (January 2016-June 2025). Adults aged ≥18 years with obesity‑related ICD‑10 codes and at least one eGFR or serum creatinine measurement were included. Predictors encompassed demographics (age, BMI), comorbidities (hypertension, diabetes, cardiovascular disease), laboratory parameters (eGFR, creatinine, A1c, urine albumin), medication classes (ACEIs/ARBs, diuretics, GLP‑1 agonists), and healthcare utilization metrics. Logistic Regression, Random Forest, and XGBoost models were developed to estimate CKD onset and stage‑progression risks. Model performance was evaluated using F1‑score, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). The best model was selected based on predictive accuracy and clinical interpretability.
RESULTS: Ensemble‑based approaches outperformed single‑algorithm models. Random Forest demonstrated the strongest predictive performance (F1: 77%; AUC: 80%), followed by XGBoost (F1: 75%; AUC: 78%). Key predictors contributing significantly to CKD onset and progression risk included: Age, baseline BMI, and history of hypertension.
CONCLUSIONS: ML‑based predictive models using real‑world data effectively identify obesity‑related CKD risk and anticipate stage progression, supporting proactive and personalized care strategies. By integrating clinical, laboratory, and utilization predictors, these models can inform payer‑provider initiatives focused on early intervention, slowing CKD progression, and reducing long‑term healthcare costs.