Impaired renal function is a relatively common finding in candidates for transplantation with decompensated cirrhosis, especially those with major impairment in liver function and/or large volume ascites.1–4 A number of mechanisms may be involved, including renal vasoconstriction (leading to hepatorenal syndrome at the most advanced stages) and superimposed intrinsic lesions.5–7 An important finding is that serum creatinine is an independent predictor for early mortality in patients with cirrhosis.8 Pretransplant serum creatinine is also a predictor of posttransplant mortality and posttransplant renal function.9, 10 Serum creatinine has been included into the widely used Model for End-Stage Liver Disease (MELD) score, along with bilirubin and the international normalized ratio (INR).11, 12 Although serum creatinine is not an index of liver function, it weights heavily on the MELD score. As an example, in a patient with a bilirubin level of 50 μmol/L and an INR of 1.1, a 30 μmol/L (0.3 mg/dL) increase in serum creatinine results in a 22% increase in the MELD score.
It has been argued recently that serum creatinine may be too heavily weighted in the existing MELD formula.12 In addition, the bounding of creatinine to 1 mg/dL for values less than 1 mg/dL in order to avoid negative values after logarithmic transformation is questionable.11 Indeed, the hypothesis can be raised that a sizable proportion of patients with cirrhosis with creatinine below 1 mg/dL have a significant impairment in renal function and that mortality is not constant for creatinine below 1 mg/dL. An updated MELD score including loge (1 + creatinine [mg/dL]) instead of loge (creatinine [mg/dL]), without bounding creatinine to 1 mg/dL, has been proposed to overcome this limitation.12 In a large population, this re-weighted score proved more accurate than the existing MELD score.12 However, the improvement was relatively modest. More data is needed on true renal function in patients with cirrhosis with low serum creatinine level.
It can be reasonably assumed that the glomerular filtration rate (GFR) might be a more accurate marker of mortality than creatinine. Creatinine-based formulas are widely used to estimate GFR in the general population.13, 14 However, both the Cockcroft and the Modification of Diet in Renal Disease (MDRD) formulas may be inaccurate in patients with cirrhosis.15–18 Indeed, protein-calorie malnutrition is common in cirrhosis,19–21 which contributes to a reduction in creatinine production and creatinine level. A decrease in creatinine production might result in overly high estimates of GFR using the Cockcroft and MDRD formulas. In practice, disagreement still exists as to whether creatinine-based methods overestimate16, 17, 22, 23 or underestimate15 true GFR in patients with cirrhosis. Furthermore, the factors associated with inaccuracies in the estimation of renal function have not been clearly documented in patients with cirrhosis. The potential impact of these inaccuracies on the MELD score must also be determined.
The aims of this study were to reassess creatinine-based estimates of renal function in candidates for transplantation with cirrhosis, focusing on those with a serum creatinine value of less than 1 mg/dL and to determine the impact of inaccuracies in the estimation of renal function on the MELD score.
BMI, body mass index; CKD-EPI, chronic kidney disease–epidemiology; GFR, glomerular filtration rate; HBV, hepatitis B virus; HCC, hepatocellular carcinoma; HCV, hepatitis C virus; INR, international normalized ratio; IQR, interquartile range; MDRD, Modification of Diet in Renal Disease; MELD, Model for End-Stage Liver Disease; ROC, receiver operating characteristic.
PATIENTS AND METHODS
From January 2004 to December 2007, 157 consecutive patients with cirrhosis who were evaluated for a first liver transplantation in a single institution were included in the study. These patients comprised 116 males (74%) and 41 females (26%). The mean age was 53 ± 9 years (range: 20-69 years). The main cause of cirrhosis was alcohol in 74 (47%), hepatitis B virus (HBV) infection in 16 (10%), hepatitis C virus (HCV) infection in 44 (28%), primary biliary cirrhosis in 2, primary sclerosing cholangitis in 5, autoimmune hepatitis in 5, genetic hemochromatosis in 3, and unknown in 8. In addition to cirrhosis, 57 of these patients (36%) had hepatocellular carcinoma (HCC). All patients with HCC met the Milan criteria, namely, a single nodule less than 5 cm or 2 or 3 nodules each less than 3 cm.24 The characteristics of the patients at evaluation are shown in Table 1.
Table 1. Baseline Characteristics of the Patients in the Study Population
At the time of evaluation for transplantation, in addition to standard workup, all patients underwent direct measurement of GFR using plasma clearance of iohexol, an exogenous marker, as previously described.25 All patients received a 5 mL intravenous dose of iohexol. Each patient then simultaneously ingested 150 mL of tap water within 30 minutes. Blood samples were taken at 0, 60, 120, 180, 240, and 300 minutes after injection. Examples of decrease in plasma iohexol concentrations are shown in Fig. 1. Clearance of iohexol was calculated by the following formula: Clearance = Dose/AUC, where AUC is the area under the plasma concentration curve.
At evaluation, demographic, biochemical, and anthropometric variables required for the calculation of GFR according to Cockcroft and MDRD equations were systematically collected. The Cockcroft estimation was calculated according to the following equation: GFR (mL/minute) = (140 − age [years] × body weight [kg])/(serum creatinine [μmol/L] × k), with k = 1.04 if female and k = 1.23 if male.13 GFR was normalized in mL/minute/1.73 m2 by using the Dubois formula for the calculation of body surface area. The simplified MDRD equation using 4 variables was calculated as follows: GFR (mL/minute/1.73 m2) = 186 × (serum creatinine [mg/dL]) −1.154 × (age [year]) −0.203 × (0.762 if patient is female) × (1.21 if patient is black).26 We also computed the CKD-EPI (chronic kidney disease–epidemiology) creatinine-based equation, which was recently proposed as an alternative to the Cockcroft and MDRD equations (Table 2).27
GFR= 166 × (creatinine [mg/dL]/0.7) – 0.329 × (0.993) Age
> 62 (> 0.7)
GFR= 166 × (creatinine [mg/dL]/0.7) – 1.209 × (0.993) Age
≤ 80 (≤ 0.9)
GFR= 163 × (creatinine [mg/dL]/0.9) – 0.411 × (0.993) Age
> 80 (> 0.9)
GFR= 163 × (creatinine [mg/dL]/0.9) – 1.209 × (0.993) Age
White or other
≤ 62 (≤ 0.7)
GFR= 144 × (creatinine [mg/dL]/0.7) – 0.329 × (0.993) Age
> 62 (> 0.7)
GFR= 144 × (creatinine [mg/dL]/0.7) – 1.209 × (0.993) Age
≤ 80 (≤ 0.9)
GFR= 141 × (creatinine [mg/dL]/0.7) – 0.411 × (0.993) Age
> 80 (> 0.9)
GFR= 141 × (creatinine [mg/dL]/0.7) – 1.209 × (0.993) Age
The MELD score was calculated according to the following equation: 9.6 × loge (serum creatinine [mg/dL]) + 3.8 × loge (serum bilirubin [mg/dL]) + 11.2 × loge (INR) + 6.4.11 In patients with serum creatinine below 1 mg/dL, values were rounded off to 1 mg/dL. The MELD-Na score was calculated according to the following equation: MELD-Na − (0.025 × MELD × [140 − Na]) + 140, where serum sodium concentration was bound between 125 and 140 mmol/L.28 The updated MELD score was calculated according to the following equation: 1.266 × loge (1 + serum creatinine [mg/dL]) + 0.939 × loge (1 + serum bilirubin [mg/dL]) + 1.658 × loge (1 + INR).12
During the study, 110 of 157 patients (70%) underwent transplantation 7 ± 5 months on average after listing. Fifteen patients (10%) died on the waiting list, 6 ± 6 months on average after listing. Twenty-seven patients (17%) were removed from the waiting list due to improvement (n = 12), progression of HCC (n = 9), or for other reasons (n = 6). For the analysis of outcome, the patients who were removed from the waiting list due to tumor progression were censored at the time of dropout from the waiting list. Therefore, patients who died due to tumor progression were not included in the group of patients who died on the waiting list due to end-stage liver insufficiency. Patients who were removed for other reasons were also censored. Finally, 5 patients (3%) were still awaiting a liver allograft at the end of this study. Up to March 2007, allocation of allograft to recipients was essentially based on waiting time. From March 2007, the MELD score–based allocation policy was implemented.
The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki and was approved by the local ethics committee.
Results for continuous variables are expressed as means ± standard deviation. Student t test, chi-squared test, Fisher's exact test, Mann-Whitney test, logistic regression analysis, and Cox regression analysis were used where appropriate. GFR estimates according to the Cockcroft and MDRD equations were compared to plasma iohexol clearance by using a paired-samples t test. Pearson's correlation analysis was used for correlation between measured GFR with plasma iohexol clearance and calculated GFR according to the Cockcroft and MDRD formulas. Cox regression analysis was used to compare the prognostic value of a score including loge serum creatinine (mg/dL) along with loge serum bilirubin (mg/dL) and loge INR (the same variables as those of the MELD score) to that of a score including loge true GFR (mL/minute/1.73 m2) along with loge serum bilirubin (mg/dL) and loge INR. The resulting scores were compared using receiver operating characteristic (ROC) curve analysis and the derived c-statistic. For all tests, P < 0.05 was considered statistically significant. Analysis was performed using SPSS (Chicago, IL) and SAS (Cary, NC) software.
Comparison Between Direct Measurement of GFR Using Plasma Iohexol Clearance and Creatinine-Based Estimates of Renal Function
In the study population, the mean value of GFR measured by plasma iohexol clearance was 85 ± 30 mL/minute/1.73 m2 (median: 87, range: 16-163, interquartile range [IQR]: 38). Mean serum creatinine was 81 ± 37 μmol/L (median: 73, range: 35-305, IQR: 30). Mean values of GFR estimated by Cockcroft and MDRD equations were 102 ± 37 mL/minute/1.73 m2 (median: 97, range: 19-253, IQR: 43) and 101 ± 37 mL/minute/1.73 m2 (median: 99, range: 15-227, IQR: 42), respectively. Mean values of GFR estimated by CKD-EPI was 93 ± 25 mL/minute/1.73 m2 (median: 98, range: 14-167, IQR: 30).
There was a statistically significant although relatively weak correlation between serum creatinine and measured GFR with an R2 value of 0.6 (P < 0.001). The correlations between estimated GFR according to Cockcroft and MDRD equations on the one hand and true GFR on the other hand were also relatively weak (R2 values of 0.5 and 0.6, respectively) but statistically significant (P < 0.001 for both). The correlation between estimated GFR according to the CKD-EPI equation and true GFR (R2 of 0.48) was weaker than that of Cockcroft and MDRD with true GFR, although it was statistically significant (P < 0.0001).
Overall, both Cockcroft and MDRD formulas tended to overestimate measured GFR (Fig. 2A,B). The mean difference between estimated GFR according to Cockcroft formula and measured GFR using iohexol clearance was 17 ± 32 mL/minute/1.73 m2 (P < 0.001). The mean difference between estimated GFR according to the MDRD formula and measured GFR was 16 ± 29 mL/minute/1.73 m2 (P < 0.001). CKD-EPI also tended to overestimate true GFR with a mean difference of 8 ± 22 mL/minute/1.73 m2 (P < 0.001) (Fig. 2C). The average difference between true GFR and CKD-EPI was numerically lower compared to Cockcroft and MDRD formulas. However, in patients with true GFR lower or equal to 70 mL/minute/1.73 m2, the difference between CKD-EPI and true GFR was 19 ± 20 mL/minute/1.73 m2 (P < 0.0001).
On univariate analysis using the Cox regression model, true GFR was an independent factor of waiting list mortality (P = 0.001; hazard ratio = 0.97; 95% confidence interval = 0.95-0.99). On multivariate analysis, when true GFR, loge bilirubin (mg/dL), and loge INR were entered in the model, true GFR and loge INR (P = 0.001 and 0.003, respectively) but not loge bilirubin were independent predictors of waiting list mortality. Similarly, patients with true GFR lower or equal to 60 mL/minute/1.73 m2 (n = 33) were at higher risk of death on the waiting list compared to patients with true GFR greater than 60 mL/minute/1.73 m2 (n = 124; P = 0.02).
Factors Associated with the Overestimation of Calculated GFR with Cockcroft Formula Compared to True GFR
Cockcroft formula overestimated true GFR by 30% or more in 59 patients (37.6%) and by 20% or more in 68 patients (43.3%). Table 3 shows that on univariate analysis, age, body mass index (BMI), body surface area, HCC, Child-Pugh grade, MELD score, MELD-Na, INR, bilirubin, albumin, platelet count, ascites (past or present), and encephalopathy (past or present) were significantly correlated with the overestimation of GFR by 20% or more using the Cockcroft formula. Serum sodium was not significantly related with the risk of misclassification. On multivariate analysis, only age (P = 0.0003), BMI (P = 0.0008), and ascites (P < 0.0001) were significantly and independently related to an overestimation of true GFR by 20% or more. The rate of overestimation was 63% for patients below 50 years old, 39% for patients between 50 and 60 years old and 25% for patients older than 60 years old. Similarly, the rate of overestimation was 3% for patients with a BMI lower than 20 (n = 13) compared to 34% for those with a BMI between 20 and 25 (n = 59), 31% for those with a BMI between 25 and 29 (n = 45) and 32% for those with a BMI over 29 (n = 50).
Table 3. Factors Associated with an Overestimation of GFR by 20% or More with Cockcroft Formula on Univariate Analysis
Overestimation of Measured GFR by 20% or More with Cockcroft Formula
yes n = 68 (43%)
No n = 89 (57%)
50 ± 9
55 ± 7
Body mass index (kg/m2)
26 ± 5
24 ± 4
Body surface area (m2)
1.88 ± 0.21
1.77 ± 0.22
Child-Pugh grade (A/B/C)
14 ± 4
12 ± 4
15 ± 5
13 ± 6
1.5 ± 0.4
1.3 ± 0.4
46 ± 38
30 ± 41
29 ± 6
32 ± 7
67 ± 33
76 ± 39
Platelets count (109/L)
81 ± 50
103 ± 59
136 ± 4
135 ± 4
Episodes of ascites
Episodes of encephalopathy
Factors Associated with the Overestimation of Calculated GFR with MDRD Formula Compared to True GFR
The MDRD formula overestimated true GFR by 30% or more in 58 patients (36.9%) and by 20% or more in 72 patients (45.8%). Table 4 shows that on univariate analysis, age, HCC, Child-Pugh grade, MELD, INR, bilirubin, albumin, creatinine, ascites (past or present), refractory ascites, and encephalopathy (past or present) were significantly correlated with the overestimation of GFR by 20% or more using the MDRD formula. Neither serum sodium nor MELD-Na was significantly related to overestimation. In contrast to the Cockcroft formula, neither BMI nor body surface area were significantly correlated to overestimation by the MDRD formula. On multivariate analysis, only age (P = 0.003) and ascites (P < 0.0001) were significantly and independently correlated with the overestimation of true GFR by 20% or more using the MDRD formula. The rate of misclassification was 63% below 50 years old, 40% between 50 and 60 years old, and 34% for those more than 60 years old.
Table 4. Factors Associated with an Overestimation of GFR by 20% or More with MDRD Formula on Univariate Analysis
Overestimation of Measured GFR by 20% or More with MDRD Formula
Yes n = 72 (46%)
No n = 85 (54%)
50 ± 10
55 ± 7
Body mass index (kg/m2)
25 ± 5
25 ± 4
Body surface area (m2)
1.8 ± 0.2
1.8 ± 0.2
Child Pugh grade (A/B/C)
14 ± 4
12 ± 4
16 ± 4
14 ± 5
1.5 ± 0.4
1.3 ± 0.4
46 ± 40
30 ± 40
29 ± 6
32 ± 7
66 ± 25
80 ± 43
Platelets count (109/L)
88 ± 56
98 ± 57
136 ± 3
135 ± 4
Episode of ascites
Episodes of encephalopathy
Impact on the Accuracy of the MELD and MELD-Na Scores
In this population, the c-statistic derived from the ROC curve analysis for predicting waiting list mortality was 0.75, 0.72 and 0.73 for MELD, MELD-Na and “updated MELD” scores,12 respectively. For the calculation of the MELD score (as well as the MELD-Na), a serum creatinine value below 1 mg/dL (88 μmol/L) is rounded off to 1 mg/dL.11 In the study population, 116 patients (74%) had a serum creatinine value lower than or equal to 1 mg/dL. In this group of patients, mean indexed GFR was 94 ± 24 mL/minute/1.73 m2 with a range 34-163 mL/minute/1.73 m2. Sixteen of these 116 patients (14%) had true GFR below 70 mL/minute/1.73 m2. The distribution of GFR in patients with serum creatinine lower than, or equal to 1 mg/dL is shown in Fig. 3. Table 5 shows some examples from the study cohort illustrating the discrepancies between serum creatinine and true GFR and showing that patients with low serum creatinine and low GFR may be disadvantaged by the MELD score.
Table 5. Examples from the Study Population Illustrating the Disadvantage of Patients with Low Creatinine and Low GFR Compared to Patients with High Creatinine and Similarly Low GFR
Creatinine Bounded for the MELD score (mg/dL)
True GFR (mL/minute/1.73 m2)
Patients 1 and 2 have a comparable MELD score, comparable values of bilirubin and INR and are given the same creatinine value while in patient 2, true GFR is 167% higher compared to Patient 1. Patient 3 has a higher MELD score than Patient 1 with comparable values of bilirubin and INR while, in Patient 3, true GFR is 158% higher than in Patient 1. Eventually, the MELD score is markedly higher in Patient 4 than in Patient 1 while these two patients have comparable true GFR because serum creatinine better reflects true GFR in Patient 4 than in Patient 1.
Among the 116 patients with serum creatinine below or equal to 1 mg/dL (≦ 88 μmol/L), refractory ascites was significantly more frequent in those with true GFR below 70 mL/minute/1.73 m2 (44%) compared to those with true GFR over 70 mL/minute/1.73 m2 (10%, P = 0.0004). In contrast, serum sodium was not significantly different between groups (134 ± 3 mmol/L compared to 136 ± 3 mmol/L, respectively).
In order to compare the prognostic value of serum creatinine (according to the MELD score equation) to that of true GFR, we created two distinct prognostic models based on Cox regression analysis in the study population. In the first model (Score 1) we entered the same variables as those of the MELD score; namely, loge bilirubin (mg/dL), loge INR and loge creatinine (mg/dL), where creatinine was bounded to 1 when below 1 mg/dL and capped to 4 mg/dL. In the second model (Score 2), we entered loge bilirubin (mg/dL), loge INR and loge GFR (mL/minute/1.73 m2) (instead of loge creatinine [mg/dL]). Each variable entered in the score was weighted by the regression coefficients obtained by the Cox regression analysis. The equation for Score 1 was as follows: (0.43 × loge bilirubin [mg/dL]) + (2.01 × loge INR) + (1.51 × loge creatinine [mg/dL]). The equation for Score 2 was as follows: (0.5 × loge bilirubin [mg/dL]) + (2.23 × loge INR) − (1.67 × loge GFR [mL/minute/1.73 m2]). A comparison between Score 1 and Score 2 was used instead of a comparison between the MELD score and Score 2 because this latter score was directly derived from our study population. Indeed, the superiority of Score 2 over the MELD score could have resulted from overfitting rather than from a difference in the prognostic value of creatinine compared to that of true GFR. However, as shown in Table 6, the accuracy of the MELD score was similar to that of Score 1 (0.75 for both).
Table 6. Accuracy of Different Prognostic Scores, Based on ROC Curve Analysis and c Statistic, for Predicting Early Mortality in the Study Population
According to Sharma P. et al. Gastroenterology 2008; 135: 1575-81.
Score 1 included the same variables as MELD score (loge creatinine [mg/dL], loge bilirubin [mg/dL] and loge INR), where each variable was weighted according to Cox regression analysis in the study population.
Score 2 included loge true GFR [mL/min1.73m2], loge bilirubin [mg/dL] and loge INR, where each variable was weighted according to Cox regression analysis in the study population.
Score 1 based upon loge creatinine (mg/dL) in the study population**
Score 2 based upon true GFR (mL/min/1.73m2) in the study population†
Both Score 1 and Score 2 were significantly correlated to waiting list mortality. The P values corresponding to the Hosner-Lemshow goodness of concordance fit test were of 0.22 and 0.96 for Score 1 and Score 2, respectively, both markedly over 0.05. However, as shown in Fig. 4, Score 2 (loge GFR-based score) was superior to Score 1 (loge creatinine-based score) for predicting outcome, with a c statistic of 0.8 compared to 0.75. The updated MELD score based on loge (1 + creatinine [mg/dL]) was not superior to the MELD score (Table 6). The size of the study population was not powered for identifying statistically significant differences.
The results of this study strongly suggest that true GFR has a better prognostic value than serum creatinine and creatinine-based formulas in patients with cirrhosis. These findings are in line with preliminary results showing that the MELD score was improved when loge creatinine (mg/dL) was changed for loge GFR (mL/minute/1.73 m2).29 Unfortunately, direct measurement of GFR using exogenous agents such as iohexol or inulin is costly and time consuming. The MELD score which includes simple and readily available variables can be easily updated according to the patient's status. Although it can be reasonably proposed that direct measurement of GFR is performed systematically during pretransplant workup in order to assess more precisely baseline renal function, true GFR is impractical for routine use if a score must be updated. Therefore, more accurate indirect assessment is still needed.
It has been pointed out that bounding serum creatinine to 1 mg/dL in patients with creatinine below 1 mg/dL may represent a significant limitation of the existing MELD score.11, 12 In keeping with other studies,16, 17, 22, 23 we found that a substantial proportion of patients with cirrhosis with serum creatinine within the normal range had impaired renal function. An “updated” MELD score including loge (1 + creatinine [mg/dL]), without bounding values below 1 mg/dL to 1 mg/dL, instead of loge creatinine [mg/dL] has been proposed to overcome this limitation.12 However, the improvement in the accuracy of the updated score was relatively limited, with an index of concordance for predicting 90-day waiting list mortality of 0.73 compared to 0.75 for the existing MELD score. The results of our study (Fig. 2) show that the range of true GFR in patients with serum creatinine below 1 mg/dL is quite broad. Therefore, even when variations in serum creatinine below 1 mg/dL are taken into account, a number of patients with impaired renal function and who are possibly at higher risk of early mortality are underscored. Again, creatinine alone is an inaccurate marker of renal function in cirrhosis, which could explain that the updated MELD score still has limitations.
The results of this study confirm that, in patients with cirrhosis, Cockcroft and MDRD equations overestimate true GFR (by 25 ± 26 and 16 ± 22 mL/minute/1.73 m2, respectively). The recently proposed CKD-EPI equation27 also overestimates true GFR, especially in those with impaired renal function. Several reasons may explain that creatinine-based equations tend to overestimate true GFR in patients with cirrhosis. First, creatinine is synthesized by the liver before being stored by skeletal muscle and eliminated after conversion to creatinine.16 Patients with impaired liver function have a lower production of creatinine. Second, muscle waste which is common during cirrhosis may also contribute to a decreased production of creatinine.30, 31 Third, patients with cirrhosis have an increased rate of tubular excretion of creatinine.17, 32 Considered together, these factors contribute to misleading values of serum creatinine level and/or an increase in the ratio between creatinine secreted by the tubule and creatinine filtered by the glomerulus. In turn, these changes in creatinine levels result in inaccuracies in creatinine-based equations. It has been suggested that creatinine-based equations may underestimate true GFR in patients with cirrhosis with relatively preserved liver function whereas, in patients with more advanced liver disease, GFR may be overestimated.15 A similar trend was observed in our population; for patients with a high MELD score, true GFR was more often overestimated by both the Cockcroft and MDRD formulas compared to patients with low MELD score (Tables 3 and 4). However, it must be noted that in this population which included candidates for transplantation with HCC, the mean “physiological” MELD score was relatively low (14 ± 4). Therefore, overestimation of GFR by creatinine-based equations was possible at an early stage in the course of cirrhosis.
In this population, we found that not all patients with cirrhosis had the same level of discordance between estimated GFR and true GFR. Patients with a high MELD score were more likely to have an overestimation of true GFR when the Cockcroft and MDRD equations were used. However, with multivariate analysis, the MELD score had no significant impact on the level of discordance. Ascites and younger age were independently and significantly associated with a higher rate of overestimation of true GFR by creatinine-based equations. Interestingly, overestimation was more pronounced for patients less than 50 years old. Both the Cockcroft and MDRD equations take into account the age of the patient.13, 26 However, the results of this study suggest that, for a given value of serum creatinine, the impact of age on true renal function is markedly different in patients with cirrhosis compared to the general population and/or patients with chronic kidney diseases. The weight given to age by Cockcroft and MDRD formulas is clearly inappropriate for patients with cirrhosis, and it should be reassessed for this population.
Independent of ascites, BMI had a significant impact on the rate of overestimation by the Cockcroft equation. The rate of overestimation was markedly different in patients with a BMI below or over 20, patients with a BMI over 20 being at higher risk of overestimation compared to true GFR. In contrast to MDRD, the Cockcroft equation takes into account body weight. This difference could explain why BMI had a significant influence on the risk of overestimation of GFR with the Cockcroft equation but not with the MDRD equation. These findings suggest that the influence of weight and/or BMI on creatinine and creatinine-based equations is markedly different in patients with cirrhosis compared to the general population, probably due, at least in part, to ascites and edemas. Overall, creatinine-based equations should be interpreted with caution in patients with cirrhosis below 50 years old and/or with ascites. Independent of ascites, the MDRD formula should be used in preference to the Cockcroft formula for patients with a BMI of over 20. In this population, there was no significant interaction between serum sodium and the accuracy of the Cockcroft and MDRD equations.
In conclusion, the results of this study strongly suggest that a score including GFR instead of serum creatinine would be more accurate at predicting early mortality in patients with cirrhosis, even if serum creatinine is not rounded off to 1 mg/dL when below 1 mg/dL. Indeed, patients with cirrhosis with normal serum creatinine may have a markedly impaired renal function. Direct measurement of GFR using exogenous agents is inappropriate for routine use as it is costly and complex. Unfortunately, creatinine-based equations are inaccurate in patients with cirrhosis and tend to result in overestimation compared to true GFR. The use of the existing equations instead of serum creatinine may not improve prognostic scores. Efforts should be made to create and validate specific equations for assessing renal function in patients with cirrhosis. Based on the results of this study, it can be anticipated that more specific equations, either based on creatinine or other markers such as cystatin C, should take into account factors which significantly correlated with overestimation of true GFR, including age below 50 years old, ascites, and BMI. The results of the Cockcroft and MDRD equations should be interpreted with caution in patients with cirrhosis below 50 years old, with ascites and/or or with a BMI over 20.
The authors are indebted to Dr. Hélène Voitot and Dr. Marie Christine Guimont, Department of Biochemistry and Molecular Genetics, Hôpital Beaujon, Clichy, France, for their advice. The authors are also indebted to Claire Worledge for her helpful assistance in the preparation of the manuscript.