Abstract: TH-OR040

Comparative Performance of Longitudinal Measures of Change in Kidney Function in Predicting ESRD and Death

Session Information

Category: Chronic Kidney Disease (Non-Dialysis)

  • 304 CKD: Epidemiology, Outcomes - Non-Cardiovascular

Authors

  • Bowe, Benjamin Charles, Clinical Epidemiology Center, Research and Development Service, Veterans Affairs St Louis Health Care System, St. Louis, Missouri, United States
  • Yan, Yan, Clinical Epidemiology Center, Research and Development Service, Veterans Affairs St Louis Health Care System, St. Louis, Missouri, United States
  • Xie, Yan, Clinical Epidemiology Center, Research and Development Service, Veterans Affairs St Louis Health Care System, St. Louis, Missouri, United States
  • Xian, Hong, Clinical Epidemiology Center, Research and Development Service, Veterans Affairs St Louis Health Care System, St. Louis, Missouri, United States
  • Li, Tingting, Clinical Epidemiology Center, Research and Development Service, Veterans Affairs St Louis Health Care System, St. Louis, Missouri, United States
  • Al-Aly, Ziyad, Clinical Epidemiology Center, Research and Development Service, Veterans Affairs St Louis Health Care System, St. Louis, Missouri, United States
Background

Four different methods are commonly used in the evaluation of longitudinal changes of eGFR: ordinary least square (OLS), annualized change (AC), group-based trajectory models (GBT), and empirical Bayes slopes (EBS). Their comparative predictive performance for ESRD and death is unknown.

Methods

We built a development (N=274,277) and five validation cohorts (N~54,857 each) of US Veterans with stage 3a CKD and aimed to comparatively evaluate risk discrimination and reclassification of Cox proportional hazard regression models using the various longitudinal eGFR change measurement methods.

Results

For the outcome of death, when eGFR change was captured over a 5-year period, risk discrimination improved gradually from base model, OLS, AC, GBT, and EBS with a corresponding c-statistic of 0.698 (0.696-0.700), 0.702 (0.700-0.704), 0.703 (0.701-0.705), 0.706 (0.704-0.708), and 0.707 (0.705-0.709), respectively. Similarly, for the outcome of ESRD, the base, OLS, AC, GBT, and EBS models had a c-statistic of 0.699 (0.693-0.705), 0.710 (0.704-0.716), 0.711 (0.705-0.717), 0.716 (0.710-0.722), and 0.718 (0.712-0.724), respectively. EBS models offered the largest net reclassification improvement of 4% and 5% for death and ESRD, respectively, and they offered the largest integrated discrimination improvement of 0.56%, and 1.27% for death and ESRD, respectively. Results were reproduced in 5 validation cohorts, and were consistent when change in eGFR was captured over 1-year period.

Conclusion

Longitudinal eGFR change contributed to risk discrimination; measuring change using empirical Bayes slopes offered the most risk discrimination, and the largest improvement in net reclassification, and integrated discrimination for the outcomes of death as well as ESRD.

Funding

  • Veterans Affairs Support