Predicting kidney failure from longitudinal kidney function trajectory: A comparison of models.

Area of Science:

• Nephrology
• Biostatistics
• Epidemiology

Background:

• Early prediction of chronic kidney disease (CKD) progression to end-stage kidney disease (ESKD) is crucial for patient management.
• Current prediction models often rely solely on baseline estimated glomerular filtration rate (eGFR).
• More complex models, including eGFR slope, time-varying eGFR, and joint modeling, have been proposed to enhance prediction accuracy.

Purpose of the Study:

• To evaluate whether advanced modeling approaches improve the prediction of ESKD compared to traditional methods.
• To assess the performance of Cox models with eGFR slope, time-varying eGFR, and joint modeling for ESKD prediction.
• To benchmark these models against the Kidney Failure Risk Equation (KFRE).

Main Methods:

• Prospective cohort study design utilizing data from two CKD cohorts (MASTERPLAN and NephroTest).
• Development and validation of various Cox models incorporating baseline eGFR, eGFR slope, and time-varying eGFR.
• Comparison of model discrimination (C-statistics) and calibration, benchmarking against the KFRE.

Main Results:

• The Kidney Failure Risk Equation (KFRE) achieved a C-statistic of 0.94.
• Models with time-varying eGFR (0.92) and baseline eGFR (0.95) showed similar predictive performance to KFRE.
• The Cox model incorporating eGFR slope demonstrated the best calibration, while joint models did not offer significant improvements.

Conclusions:

• In this study, advanced modeling techniques like joint models did not outperform traditional approaches for ESKD prediction.
• Models utilizing eGFR slope or time-varying eGFR provided comparable predictive performance to established equations.
• The findings suggest that simpler, well-calibrated models may be sufficient for predicting ESKD in specific clinical scenarios with complete follow-up data.