Kidney Week

Abstract: FR-PO1147

Predicting CKD Progression Using Advanced Statistical and Machine-Learning Approaches

Session Information

• CKD: Screening, Diagnosis, Serum and Urine Biomarkers, and Scoring Indices
  November 07, 2025 | Location: Exhibit Hall, Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: CKD (Non-Dialysis)

• 2301 CKD (Non-Dialysis): Epidemiology, Risk Factors, and Prevention

Authors

• Jiang, Huan, Otsuka Pharmaceutical Development & Commercialization, Inc., Princeton, New Jersey, United States

Huan Jiang, PhD

• Yu, Kunbo, Pharmapace, Inc., San Diego, California, United States

Kunbo Yu

• Zhang, Zhen, Otsuka Pharmaceutical Development & Commercialization, Inc., Princeton, New Jersey, United States

Zhen Zhang, PhD

Background

Predicting chronic kidney disease (CKD) stage progression is crucial since it allows healthcare providers to initiate appropriate treatments as early as possible. The purpose of this study was to compare different advanced statistical and machine learning methods for more accurate prediction.

Methods

Using data collected from an observational and longitudinal study (OVERTURE; NCT01430494), multinomial logit models with variable selection and machine learning (ML) approaches including Random Forest, LASSO/Ridge/Elastic Net regression, XGBoost, and Ensemble learning, were used and compared by weighted kappa, for their models’ prediction accuracy and robustness on CKD stage up to 36 months. Baseline demographic characteristics, social determination of health, height adjusted TKV, blood pressures, comorbidities, and pharmacotherapy were included in these models.

Results

3297 patients with CKD from OVERTURE were used for model development. Both advanced statistical and machine learning approaches behaved well. The average weighted kappa values for these models were ranged from 0.902 to 0.904. After model selection in multinomial logit models, the most contributing baseline predictors for CKD progression included age, systolic blood pressure, region, height adjusted TKV, hypertension, proteinuria, hematuria, and calcium channel blockers use. The top influential variables were consistent with variable importance by machine learning methods.

Conclusion

For CKD patients, novel methods for accurate prediction could help to lower the risk of CKD progression.

Funding

• Commercial Support – Otsuka Pharmaceutical Development & Commercialization, Inc., Princeton, NJ

Digital Object Identifier (DOI)

• doi: 10.1681/ASN.2025kteg1ey3