Clinical factors associated with biased estimation of glomerular filtration rate: a cross-sectional study

Equations estimating glomerular filtration rate (GFR) based on creatinine and/or cystatin C incorporate demographic variables such as age and sex. However, clinical determinants may lead to substantial bias in GFR estimation. We aimed to identify clinical factors associated with biased GFR estimation with the Modification of Diet in Renal Disease, Chronic Kidney Disease-Epidemiology Collaboration, and European Kidney Function Consortium equations.

In this retrospective cross-sectional study, we included patients referred for GFR measurement from March 2008 to February 2024 in the Physiology unit of Bichat Hospital, Paris, France. GFR was measured as the urinary clearance of a radio-isotopic tracer and the error of estimated GFR (eGFR), was expressed as log(eGFR/measured GFR) and analyzed with linear regression models.

Among 3838 patients (mean age 51 ± 15 years, mean measured GFR 59 ± 26 ml/min/1.73 m²), several clinical variables were associated with significant estimation error in the multivariable analysis. For all creatinine-based equations, underestimation occurred with HIV infection, high BMI, loop diuretics, and cotrimoxazole use, while overestimation occurred with younger age, female sex, lower BMI, history of kidney transplantation, and cirrhosis. For all cystatin C-based equations, underestimation was associated with older age, HIV infection, corticosteroid use, and history of kidney transplantation; overestimation was associated with younger age, female sex, and sub-Saharan African origin. Equations using both biomarkers performed better in conditions affecting each biomarker in opposite directions. By cumulating several conditions, bias can vary from -50% to more than +50% leading to a completely erroneous GFR estimation.

Common clinical features are independently associated with biased GFR estimation that can be of high clinical relevance, especially when cumulated. Our results guide GFR evaluation by giving a qualitative and most importantly quantitative estimation of the expected bias depending on individual patient profile and comorbidities.