Comparing methods for glomerular filtration rate estimation.

Area of Science:

• Nephrology
• Biostatistics
• Biomarker Discovery

Background:

• Serum creatinine (SCr)-based equations are standard for estimating glomerular filtration rate (GFR) but have limitations.
• Inaccuracies in SCr-based GFR estimation can arise from non-GFR determinants and racial disparities.
• Few studies have explored advanced modeling to improve SCr-based GFR estimation performance.

Purpose of the Study:

• To develop and evaluate advanced modeling techniques for SCr-based GFR estimation.
• To compare the performance of multivariable fractional polynomials (MFP), generalized additive models (GAM), random forests (RF), and gradient boosted machines (GBM) against the 2021 CKD-EPI SCr equation.
• To assess performance across diverse demographic subgroups, including race, sex, age, and GFR levels.

Main Methods:

• Four advanced modeling techniques (MFP, GAM, RF, GBM) were used to create SCr-based GFR-estimating equations.
• Performance was evaluated using bias, precision, and accuracy metrics (P10, P30) in a pooled validation dataset (n=2215).
• Comparison was made against a refitted linear regression-based 2021 CKD-EPI SCr equation.

Main Results:

• All equations exhibited the greatest bias and lowest accuracy in Black individuals.
• MFP and GAM equations showed performance similar to the refitted CKD-EPI equation, with minor improvements in P10/P30 for Black individuals and females.
• GBM and RF equations had smaller biases but lower accuracy compared to other methods; overall differences were modest.

Conclusions:

• Advanced modeling techniques offer minimal improvement for SCr-based GFR estimation.
• Current SCr-based methods show persistent disparities, particularly in Black individuals.
• Future research should prioritize novel biomarkers and enhance GFR measurement feasibility.