Critically ill patients admitted to intensive care units (ICUs) exhibit pathophysiological changes that can significantly alter drug kinetics, thereby complicating dose optimisation.1 The accurate assessment of renal function is crucial for drugs eliminated primarily via the kidneys. To date, glomerular filtration rate (GFR) remains the reference standard for assessing renal function.2 It is ideally measured using exogenous markers, such as inulin or radiopharmaceuticals, including technetium-99 m-labelled diethylenetriaminepentaacetic acid, which are freely filtered by the glomerulus and neither reabsorbed nor secreted. However, these techniques are costly and difficult to implement in ICU clinical practice.3

Twenty-four-hour urinary creatinine clearance (CrCl24h) has been proposed to measure GFR; however, its accuracy is also limited by variability in urine collection and in creatinine production during hospitalisation.4 Consequently, serumcreatinine-based equations incorporating anthropometric data are widely used in clinical practice to estimate GFR (eGFR). The most commonly used equations are the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation, the Modification of Diet in Renal Disease (MDRD-4) equation, and the Cockcroft-Gault (CG) equation.5–7

As these equations were not originally developed for critically ill patients, their validity in this clinical setting remains a matter of debate. Incorrect estimation of renal function, whether overestimation or underestimation, can result in inappropriate drug dosing, compromising therapeutic efficacy and increasing the risk of toxicity. This issue is particularly relevant in critically ill patients, since their renal function fluctuates frequently and conventional estimation methods are significantly limited.

The literature indicates that these equations show a modest correlation with GFR, but lack sufficient accuracy in critically ill patients, exhibiting clinically relevant bias and error, particularly in renal failure and augmented renal clearance (ARC).8–10 Augmented renal clearance, defined as an eGFR ≥130 mL/min11 as measured by 24-h urine collection, is of particular relevance for antimicrobial therapy, as suboptimal exposure is associated with an increased risk of treatment failure. Thus, accurate estimation of GFR is essential in this population.12

This study aimed to assess the performance and accuracy of 3 serum creatinine-based equations for estimating GFR in critically ill patients with and without ARC, using CrCl24h measured in ICU patients as the reference method.

The study included 261 patients; 60.2% were men, with a median age of 64 years (IQR 54–73). Table 2 shows the demographic and clinical data of the study population. The median CrCl24h value for the entire cohort was 63 mL/min (IQR 32–107). The medians of eGFR using CKD-EPI, MDRD-4, and CG were 83.8 mL/min (IQR 43.2–104.7), 80.3 mL/min (IQR 42.5–123.2), and 75.9 mL/min (IQR 43.2–121.6), respectively. The prevalence of ARC was 15.7% (41 patients), using a cutoff point of ≥130 mL/min.

Table 3 shows a direct comparison between CrCl24h and each GFR estimating equation. Spearman coefficients for patients with a CrCl24h between 0 and 129 mL/min showed a moderate correlation for all GFR estimating equations (p < 0.01). The strongest correlation with measured CrCl24h was observed for the CG equation (rS = 0.76), followed by the CKD-EPI equation (rS = 0.75), and the MDRD-4 equation (rS = 0.74). For patients with ARC, the correlation between the eGFR equations and CrCl24h was weak (rS ≤0.3). The CG equation showed a slightly higher correlation (rS = 0.30), followed by the MDRD-4 equation (rS = 0.18), and the CKD-EPI equation (rS = 0.14).

Figs. 1 and 2 show Bland–Altman plots for the full and stratified cohorts, respectively. For patients with eGFR between 0 and 129 mL/min, the CKD-EPI equation showed the lowest bias (−13.9 mL/min/1.73 m2; 95% CI: −17.3 to −10.6 mL/min/1.73 m2) and the highest accuracy (±25.6 mL/min/1.73 m2). In contrast, the CG and MDRD-4 equations showed greater bias (−22.0 mL/min/m2; 95% CI: −26.9 to −17.1 mL/min/m2 and −23.5 mL/min/1.73 m2; 95% CI: −29.0 to −18.0 mL/min/1.73 m2, respectively) and lower accuracy, reflected by wider variability (±36.7 mL/min/m2 for CG and ± 41.5 mL/min/1.73 m2 for MDRD-4). No significant differences were found between the equations in terms of bias and accuracy in this subgroup of patients.

Figure 1.

Bland–Altman plots for the full cohort. The Y-axis represents the difference between measured CrCl24h and the eGFR equation; the X-axis represents the mean of both measurements. A) CKD-EPI equation, B) MDRD-4 equation, and C) CG equation.

Figure 2.

Bland–Altman plots using a cut-off point of ≥130 mL/min for ARC. The Y-axis represents the difference between the measured CrCl24h and the eGFR equation; the X-axis represents the mean of both measurements. Patients with CrCl24h between 0 and 129 mL/min are shown in the upper plots: A) CKD-EPI, B) MDRD-4, and C) CG. Patients with ARC are shown in the lower plots: D) CKD-EPI, E) MDRD-4, E) and F) CG. ARC, augmented renal clearance; GFR, Glomerular filtration rate.

For patients with ARC, the CKD-EPI equation showed the greatest bias (66.4 mL/min/1.73 m2; 95% CI: 54.5 to 78.2 mL/min/1.73 m2) compared with the CG eq. (24.8 mL/min/m2; 95% CI: 5.7 to 43.9 mL/min/m2) and the MDRD-4 eq. (29.3 mL/min/1.73 m2; 95% CI: 9.9 to 48.7 mL/min/1.73 m2), with statistically significant differences between the CKD-EPI equation and the CG and MDRD-4 equations (p < 0.01). Conversely, no significant differences in accuracy were observed among the 3 GFR estimating equations in this cohort, with values of ±38.7 mL/min/1.73 m2 for the CKD-EPI equation, ± 62.4 mL/min/m2 for the CG equation, and ± 63.5 mL/min/1.73 m2 for the MDRD-4 equation. Table 3 shows the limits of agreement, and Fig. 3 shows the estimated bias.

Figure 3.

Estimated bias in eGFR using different creatinine-based equations in patients with CrCl24 < 130 mL/min and in patients with ARC.

Our study shows that the accuracy and clinical utility of GFR estimating equations can vary significantly with patient profile. In the studied cohort, no serum creatinine-based equation proved superior to the others in critically ill patients with a CrCl24h between 0 and 129 mL/min. The 3 evaluated equations (CKD-EPI, MDRD-4 and CG) showed a moderate correlation with CrCl24h (rS = 0.74–0.76; p < 0.01) and all exhibited moderate bias and wide limits of agreement. In line with previous studies,13–15 our findings also suggest that the equations tend to overestimate eGFR; the median eGFR values calculated using CKD-EPI, MDRD-4, and CG were 20.8, 17.3, and 13.0 mL/min higher than the median CrCl24h value, respectively.

The use of equations to estimate eGFR in critically ill patients remains a matter of debate, as these equations are commonly used to adjust drug dosages in clinical practice, despite not being validated for this patient group.16 Patients in intensive care may experience significant physiological changes, including an increased volume of distribution, increased cardiac output, hypoalbuminaemia, a high incidence of acute kidney injury, and frequent use of extracorporeal systems. These factors can significantly impact the pharmacokinetics and pharmacodynamics of many drugs, particularly those primarily eliminated via the kidneys.9,17,18 Furthermore, prolonged hospital stays and catabolic states can alter creatinine production, making serum creatinine an unreliable biomarker for renal function assessment because of its high variability.19

Several studies have evaluated the performance of serum creatinine-based equations in intensive care settings. However, the results have been inconsistent. Ruiz et al. reported that the CKD-EPI and CG equations were slightly more accurate than the MDRD equation, with biases of −11.3 mL/min/1.73 m2, −18.8 mL/min/1.73 m2, and −22.5 mL/min/1.73 m2, respectively. These results are consistent with those obtained in our study.20 A study in Australia by Tsai et al. reported that, although the GFR equation adjusted for total body weight had a low bias (−8.2 mL/min/1.73 m2), the CKD-EPI equation showed the narrowest limits of agreement (115.9 mL/min/1.73 m2) in the non-Indigenous subpopulation.21 In our study, the CKD-EPI equation showed the lowest bias and narrowest limits of agreement (100.2 mL/min/1.73 m2) in patients with a CrCl24h between 0 and 129 mL/min.

In contrast, Wongpraphairot et al. found limited validity of eGFR equations in 49 critically ill patients; the CG equation showed the best performance, with a bias of −8.6 mL/min/1.73 m2 compared with the CKD-EPI and MDRD-4 equations, which showed biases of −19.0 mL/min/1.73 m2 and −32.0 mL/min/1.73 m2, respectively.8 These findings contrast with those of our study: among patients with CrCl24h between 0 and 129 mL/min, the CKD-EPI equation showed the lowest bias (−13.9 mL/min/1.73 m2), followed by the CG equation (−22.0 mL/min/m2), and the MDRD-4 equation (−23.5 mL/min/1.73 m2). These differences could arise from the lower incidence of ARC in the study by Wongpraphairot et al. than in our study and the previously mentioned studies. This lower incidence may reflect a population with characteristics associated with a lower ARC risk: older age, a lower proportion of male patients, and fewer cases of traumatic and surgical aetiology, in contrast to our cohort and those of Ruiz et al. and Tsai et al.

Several studies have shown that young age, trauma as the cause of ICU admission, and male sex are independent factors that increase the risk of ARC.18,22 Identifying patients with ARC is of vital importance to avoid underdosing of drugs eliminated via the kidneys. Although data on the performance of eGFR equations in critically ill patients with ARC remain limited, several studies have highlighted the importance of dose adjustment in this population, particularly for hydrophilic antibiotics, where appropriate dosing is essential to prevent treatment failure and the emergence of resistance due to suboptimal serum drug concentrations.9,17

The incidence of ARC described in the literature ranges from 30% to 65%.23 We observed a lower incidence of ARC, which could be partly attributed to the heterogeneity of the ICUs included (2 general ICUs, 1 respiratory ICU, and 1 surgical ICU). Among the subgroup of patients with ARC, all eGFR equations evaluated showed a modest correlation (rS ≤0.3; p > 0.05) and high bias values relative to CrCl24h, consistent with international studies indicating the low accuracy of these equations in the context of renal hyperfiltration.8,20,21 The CG equation showed the lowest bias (24.8 mL/min/m2), followed by the MDRD-4 eq. (29.3 mL/min/1.73 m2) and the CKD-EPI eq. (66.4 mL/min/1.73 m2), although all significantly underestimated renal function in patients with ARC (148.7 mL/min/m2; 136 mL/min/1.73 m2 and 114 mL/min/1.73 m2, respectively). It is worth noting that significant differences in bias were observed between the CKD-EPI equation and the other 2 equations (p < 0.01).

Our results in patients with ARC are consistent with the current literature. Baptista et al. demonstrated, in a cohort of 86 patients with ARC, that eGFR equations are inadequate for estimating eGFR in these patients. In their study, Spearman correlation coefficients were 0.26 for the CG equation and 0.22 for the MDRD-4 equation, with a greater bias for the latter (48 mL/min/1.73 m2) compared with the CG eq. (39 mL/min/1.73 m2).24 These findings are consistent with a previous study that evaluated the performance of the CG and MDRD-4 equations in patients with ARC, reporting Spearman coefficients of 0.34 and 0.29, and biases of 11.2 mL/min/1.73 m2 and 19.9 mL/min/1.73 m2, respectively.25 Subsequent studies included the CKD-EPI equation in the analysis. For example, Ruiz et al. obtained results consistent with those of our study in terms of bias, observing that the CG equation exhibited the lowest bias (35.7 mL/min/1.73 m2), followed by the MDRD-4 eq. (40.9 mL/min/1.73 m2), and the CKD-EPI eq. (57.9 mL/min/1.73 m2).21 Similarly, Al-Dorzi et al., in a cohort of patients with ARC, reported the best Spearman correlation coefficient for the adjusted CG equation (rS = 0.38), followed by the MDRD-4 and CKD-EPI equations (rS = 0.27 and rS = 0.22, respectively), with biases of 0.7 mL/min, 29.2 mL/min/1.73 m2 and 44.0 mL/min/1.73 m2, respectively.26 Taken together, our results and the literature suggest that the CKD-EPI equation may not be suitable for patients with ARC, given its poor performance, as evidenced by a low Spearman correlation coefficient (rS = 0.14) and a high bias (66.4 mL/min). Therefore, in patients with ARC, eGFR should be estimated using the CG or MDRD-4 equations.

The results of this study should be interpreted with caution, taking into account both its strengths and limitations. Its strengths include the large sample size of critically ill patients and the evaluation of 3 different eGFR equations commonly used in routine clinical practice. Its limitations include the heterogeneity of the study cohort, as patients were recruited from 4 different hospital ICUs. The retrospective nature of the study also limits the ability to ensure the accuracy of the 24-h urine measurements, as undetected errors may have occurred during collection. Another limitation is the use of CrCl24h as the reference standard rather than substrates that are exclusively filtered, such as inulin. However, our objective was to evaluate methods commonly used in daily clinical practice. Furthermore, cystatin C-based equations were not included, as this parameter is not routinely measured in patients admitted to the ICUs at our institution. We also assumed a steady-state condition to estimate GFR, although it is recognised that critically ill patients may experience fluctuations in creatinine production during their ICU stay. Finally, this study is limited by the lack of analysis of the impact of using different GFR estimating equations on drug dosing in patients with ARC. Future studies could also focus on assessing systemic exposure for drugs where therapeutic concentration monitoring is available as a tool for dose optimisation.

This study highlights that the equations commonly used to estimate glomerular filtration rate in critically ill patients have significant limitations compared with CrCl24h. Clinicians should be aware of this limitation when using serum creatinine-based equations to assess renal function in patients admitted to ICUs. No significant differences in bias and accuracy were observed between the CG, MDRD-4, and CKD-EPI equations in patients with a CrCl24h between 0 and 129 mL/min. However, in patients with ARC, the CG and MDRD-4 equations performed better than the CKD-EPI equation, suggesting that the latter should not be used in this clinical context.

Equations for estimating glomerular filtration rate are routinely used in critically ill patients, despite not being validated in this population. This study expands the available knowledge on the performance of 3 creatinine-based equations for estimating glomerular filtration rate commonly used in this clinical setting. The results are particularly relevant for the subpopulation of patients with augmented renal clearance (≥130 mL/min), as they have an increased risk of underexposure to predominantly renally eliminated antimicrobials, which could lead to treatment failure and impact mortality.

In this setting, the data may raise awareness among clinicians and pharmacists of the limitations of these equations, thereby supporting appropriate dosing of renally eliminated drugs.

This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the hospital's ethics committee (registration reference HCB/2022/1012). Exemption from informed consent was requested as the study design was retrospective and non-interventional.

Joan Ramon Roma: Writing – review & editing; Writing – original draft; Methodology; Investigation; Formal analysis; Data curation; Conceptualisation. Natàlia Arranz-Pasqual: Writing – review & editing; Investigation; Data curation. Carla Bastida: Writing – review & editing; Writing – original draft; Supervision; Project administration; Methodology; Conceptualisation; Validation. Dolors Soy: Writing – review & editing; Supervision; Methodology; Conceptualisation; Validation.

None declared.