Dynamically predicting renal failure after development of diabetes across biobanks

Area of Science:

• Nephrology
• Endocrinology
• Data Science

Background:

• End-stage renal disease (ESRD) is a significant diabetes complication.
• Static risk scores struggle with evolving patient profiles and competing mortality risks.
• Accurate, dynamic risk assessment is crucial for timely intervention.

Purpose of the Study:

• Develop and validate a landmark-based dynamic risk score for ESRD in diabetes patients.
• Improve individualized risk prediction by accounting for evolving patient data and mortality.
• Compare the novel score against existing static risk equations.

Main Methods:

• Retrospective cohort study of 708,435 U.S. Veterans with newly diagnosed diabetes.
• Developed a landmark-based ESRD Dynamic Risk Score (ESRD-DRS) using EHR data.
• Validated externally in the All of Us (AoU) cohort (n=13,223).
• Utilized penalized Fine-Gray subdistribution hazard models with over 400 variables.
• Evaluated discrimination (AUROC) and calibration (Brier score) over 1-, 5-, and 10-year horizons.

Main Results:

• ESRD-DRS demonstrated high discrimination (AUROCs 0.85-0.94) and good calibration in both VHA and AoU cohorts.
• The dynamic score outperformed established static equations (RECODe, KFRE).
• Key predictors included eGFR, albuminuria, systolic blood pressure, and age, consistent across landmarks.
• ESRD-DRS effectively updates risk estimates at 1, 5, and 10 years post-diabetes diagnosis.

Conclusions:

• The landmark-based ESRD-DRS is a scalable and accurate tool for dynamic ESRD risk prediction in diabetes.
• It surpasses static scores by incorporating evolving patient data and competing risks.
• Integration into EHR systems can enhance personalized ESRD risk assessment and management.