Serum sodium based modification of the MELD does not improve prediction of outcome in acute liver failure

The study was carried out according to the Declaration of Helsinki and the guidelines of the International Conference for Harmonization for Good Clinical Practice, it was approved by the local Ethics Committee of the University Hospital Essen (Institutional Review Board). In a prospective monocenter study (11/2006–12/2010), we recruited 108 consecutive ALF patients (64% females / 36% males), who met the criteria defined by the “Acute Liver Failure Study Group Germany” [20]. In brief, ALF was diagnosed by significant liver dysfunction with pathologically increased laboratory parameters (bilirubin, AST, ALT, AP, γ-GT) and an international normalized ratio (INR) of >1.5 with the concomitant presence of any degree of encephalopathy. Reference values for normal ranges are presented in Table 1. A pathological increase was defined as any value above these ranges. Other causes of liver dysfunction were excluded, such as acute-on-chronic liver failure or pre-existing cirrhosis. All patients had presented within four weeks of disease onset without pre-existing liver disease. Upon admission, clinical data were collected. Outcome (spontaneous recovery, SR; non-spontaneous recovery, NSR: comprising transplantation or death) was defined by the status after 4 weeks post admission.

Serum sodium, KCC, MELD, UKELD and MELD-Na were assessed upon admission. Parameters were correlated with the outcome at 4 weeks after admission. In order to calculate the individual MELD score we used the formula as published by Kamath et al.[21]. If the bilirubin and creatinine values were below 1.0 mg/dl they were set to 1.0 mg/dl. If the creatinine was above 4.0 mg/dl or the patients underwent dialysis during two weeks before assessment, the creatinine value was adjusted to 4.0 mg/dl. In order to calculate the individual MELD-Na score we used the formula as published by Ruf et al.[17]. If the sodium values were below 125 mmol/l they were set to 125 mmol/l, if the values were above 140 mmol/l they were adjusted to 140 mmol/l. In order to calculate the individual UKELD score we used the formula as published by Barber et al.[19]. The KCC were evaluated following published criteria. Differentiation between acetaminophen induced ALF and non-acetaminophen induced ALF have been made, as previously published [22]. Detailed formulas of the utilized scores are given in Table 2.

Differences between parameters were evaluated by one-way Analysis of Variance, repeated-measure Analysis of Variance, or paired Student’s t-test and t-test for independent samples t-test. For MELD and modified MELD statistics the Mann–Whitney test was used. For categorical variables, frequencies and percentages were estimated. χ² or Fisher’s exact tests were used for categorical factors. ROC calculations were undertaken where applicable. Screening, optimal, and diagnostic cutoff values were calculated, and the optimal cutoff, including specificity and sensitivity as well as the AUC are shown in boxes included in the ROC plots. A p < 0.05 was considered statistically significant. All values are given as means ± standard error of means. Analyses were performed with SPSS 19.0.1, version 2008 (SPSS, Chicago, IL, USA).

Mean age of the 108 patients [69 (64%) females and 39 (36%) males] was 43.4 ± 16.2 Underlying etiologies are presented in Figure 1A. Detailed laboratory parameters and general data are listed in Table 1. Sixty-two (60%) patients recovered spontaneously (SR), fifty-four (40%) deceased or underwent LTx (NSR) (Figure 1B). Differences concerning outcome in patients with non-acetaminophen induced ALF are presented in Table 3.

To assess the predictive value on the clinical outcome of sodium in comparison to individual MELD parameters, we analyzed the individual parameters at the date of the maximum MELD score within 4 weeks after admission. Upon the point of maximum MELD, significant differences between SR and NSR groups were found for all of the individual MELD parameters, including INR (SR 1.9 ± 0.08, NSR 3.99 ± 0.32, p < 0.05; Figure 2A), total bilirubin (SR: Median: 11.4 Range: 40.5 [mg/dl], NSR: Median: 20.2; Range: 43.1 Figure 2B) and creatinine (SR 1.32 ± 0.14 mg/dl, NSR 2.15 ± 0.19 mg/dl, p < 0.05; Figure 2C). In contrast to these findings, serum sodium did not show any difference between these two groups (SR 138.63 ± 0.57 mmol/l, NSR 138.63 ± 0.99 mmol/l; Figure 2D). Analyses of the performance for the individual MELD parameters (Figure 3A) showed that increased INR at date of maximum MELD had a modest sensitivity and specificity (both 88%) (INR: cutoff: 2.47, AUC: 0.922; 95% CI: 0.867-0.977; p < 0.05). Increased creatinine alone had a sensitivity of 75% and a specificity of 70% at a cutoff value of 1.16 mg/dl (creatinine: AUC 0.723; 95% CI: 0.629-0.846; p < 0.05). In contrast, serum bilirubin lacked specificity at date of maximum MELD. However, a cutoff for serum bilirubin was calculated with an anticipated low specificity (bilirubin: cutoff: 19.8 mg/dl; AUC 0.661; 95% CI 0.54-0.774; p < 0.05). ROC curve analysis for serum sodium revealed no significant prognostic value.

The classical MELD score showed a strong correlation with the clinical outcome (Figure 3B: MELD: AUC 0.967; CI 0.935 – 0.998; p < 0.01). Since hyponatremia is associated with an unfavorable prognosis, inclusion of serum sodium into the MELD score was found to improve its predictive value in cirrhotic patients [13, 17]. However, in our ALF cohort, the classical MELD was superior to sodium based modifications like the MELD-Na (AUC: 0.960; CI: 0.924-0.966; p < 0.01) or the UKELD (AUC: 0.828; CI: 0.738-0.918; p < 0.01).

A comparison of the MELD score, MELD-Na and the KCC in non-acetaminophen induced ALF demonstrated a better performance of the MELD based scores (Table 4). Concerning sensitivity, specificity, negative predictive value (NPV), positive predictive value (PPV) and overall diagnostic accuracy, the classic MELD was superior to KCC (MELD: Sensitivity: 0.90, Specificity: 0.92, PPV: 0.90, NPV: 0.92, diagnostic accuracy: 0.912; KCC: Sensitivity: 0.72, Specificity: 0.90, PPV: 0.86, NPV: 0.79, diagnositc accuracy: 0.818). While the accuracy of MELD-Na to predict outcome was lower, compared to classic MELD score, MELD-Na performed still better than KCC (MELD-Na: Sensitivity: 0.90, Specificity: 0.90, PPV: 0.88, NPV: 0.92 and diagnostic accuracy: 0.898).

A comparison of the MELD score at time of admission and the timepoint with a maximum MELD score within four weeks after admission in non-acetaminophen induced ALF revealed a better performance of the “maximum MELD” based scores (Table 4). Concerning sensitivity, specificity, negative predictive value (NPV), positive predictive value (PPV) and overall diagnostic accuracy the maximum MELD values were superior to the initial ones (initial MELD: AUC: 0.839; CI: 0.757-0.922; p < 0.01 Sensitivity: 0.78, Specificity: 0.76, PPV: 0.73, NPV: 0.81, diagnostic accuracy: 0.772, initial MELD-Na: AUC: 0.825; CI: 0.738-0.913, p<0.01 Sensitivity: 0.80, Specificity: 0.76, PPV: 0.72, NPV: 0.83, diagnostic accuracy 0.780).