Comparative performance of CKD-EPI equations in people with diabetes: An international pooled analysis of individual participant data

Abstract

Aim

This study assessed the concordance and misclassification of chronic kidney disease (CKD) stages between directly measured glomerular filtration rate (mGFR) and estimates of GFR (eGFR) using the creatinine-based CKD-EPI-2009 and the CKD-EPI-2021 equations in individuals with diabetes.

Methods

Data from 5,177 individuals across six international diabetes cohorts included mGFR measurements using exogenous filtration markers. We calculated an intra-class correlation coefficient (ICC), bias, precision and accuracy between mGFR and CKD-EPI estimates using a four-level mixed-effect linear variance component model.

Results

The pooled cohort included people with type 1 (n = 1,748, median age: 33 years [IQR: 27, 40], mGFR = 104.2 ml/min per 1.73 m²) and type 2 diabetes (n = 3,429, median age: 66 years [IQR: 58, 73], mGFR = 58.4 ml/min per 1.73 m²). Both CKD-EPI equations showed good agreement (2009 ICC: 0.90; 2021 ICC: 0.87) but substantial bias (2009: 3.7 ml/min/1.73 m²; 2021: 8.6 ml/min/1.73 m²), low precision (2009: 12.4 ml/min/1.73 m²; 2021: 13.91 ml/min/1.73 m²), and limited accuracy (2009 p30: 77 %; 2021 p30: 70 %) compared to mGFR.

Conclusion

The use of CKD-EPI equations has the potential for misdiagnosis and suboptimal CKD management in people with diabetes. Alternative methods of estimating kidney function for people with diabetes are needed to optimally manage diabetes-related kidney disease.