Kidney Week

Abstract: FR-PO0015

A Multimodel Causal Survival Analysis Framework for Individualized Dialysis Decision Support in Patients with Advanced CKD: A Retrospective Cohort Study

Session Information

• Artificial Intelligence: Clinical Prediction, Decision Support, and Digital Kidney Care
  October 23, 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

• Leung, Ka Chun, Tuen Mun Hospital, Hong Kong, Hong Kong

Ka Chun Leung, MBBS, MPH, FASN

• Yap, Yat Hin Desmond, The University of Hong Kong, Hong Kong, Hong Kong

Yat Hin Desmond Yap, MD, MBBS, PhD, FASN

Background

Mortality benefit from dialysis at the advanced CKD decision point, defined as eGFR ≤10 mL/min/1.73m², is heterogeneous. We developed PRISM, a 4-zone causal survival framework to estimate individualized benefit and support shared dialysis decisions(1).

Methods

A retrospective cohort in Hong Kong (n = 20184) was included in model development to predict treatment benefit from dialysis initiation within 90 days when eGFR <= 10ml/min/1.73m². PRISM combined prediction of four models: Random Survival Forest(RSF), Causal Forest(CF), R-Learner(RL), and an all-cause mortality model(ACMM), all were tested in held-out spatial, temporal, and external cohorts. Individual treatment effects(ITEs) in mortality reduction were estimated over 5 years. Mortality risk and model-predicted treatment benefit were mapped to four decision zones.

Results

All models showed significant treatment-effect heterogeneity. Zone validation showed no significant survival advantage from early dialysis in Zone A (low risk/low benefit), or Zone D (high risk/low benefit), while survival benefit was strongest in Zone C, (HR 0.28–0.38, p<0.005) across cohorts.

Conclusion

PRISM provides patient-specific predictions of dialysis benefit. It identified a high-risk subgroup in whom early dialysis showed no significant survival benefit. This subgroup warrants prospective evaluation for conservative kidney management.