The glomerular filtration rate (GFR), estimated from serum creatinine (SCr), is widely used in clinical practice for kidney function assessment, but SCr-based equations are limited by non-GFR determinants and may introduce inaccuracies across racial groups. Few studies have evaluated whether advanced modeling techniques enhance their performance.

Using multivariable fractional polynomials (MFP), generalized additive models (GAM), random forests (RF), and gradient boosted machines (GBM), we developed four SCr-based GFR-estimating equations in a pooled data set from four cohorts (n = 4665). Their performance was compared to that of the refitted linear regression-based 2021 CKD-EPI SCr equation using bias (median difference between measured GFR [mGFR] and estimated GFR [eGFR]), precision, and accuracy metrics (e.g., P10 and P30, percentage of eGFR within 10% and 30% of mGFR, respectively) in a pooled validation data set from three additional cohorts (n = 2215).

In the validation data set, the greatest bias and lowest accuracy, were observed in Black individuals for all equations across subgroups defined by race, sex, age, and eGFR. The MFP and GAM equations performed similarly to the refitted CKD-EPI SCr equation, with slight improvements in P10 and P30 in subgroups including Black individuals and females. The GBM and RF equations demonstrated smaller biases, but lower accuracy compared to other equations. Generally, differences among equations were modest overall and across subgroups.

Our findings suggest that advanced methods provide limited improvement in SCr-based GFR estimation. Future research should focus on integrating novel biomarkers for GFR estimation and improving the feasibility of GFR measurement.