fax: (45) 3545 2913
MELD score as a predictor of liver failure and death in patients with acetaminophen-induced liver injury†
Article first published online: 26 FEB 2007
Copyright © 2007 American Association for the Study of Liver Diseases
Volume 45, Issue 3, pages 789–796, March 2007
How to Cite
Schmidt, L. E. and Larsen, F. S. (2007), MELD score as a predictor of liver failure and death in patients with acetaminophen-induced liver injury. Hepatology, 45: 789–796. doi: 10.1002/hep.21503
Potential conflict of interest: Nothing to report.
- Issue published online: 26 FEB 2007
- Article first published online: 26 FEB 2007
- Manuscript Accepted: 25 OCT 2006
- Manuscript Received: 20 MAR 2006
The Model for End-Stage Liver Disease (MELD) scoring system has been established as a reliable measure of short-term mortality risk in patients with end-stage chronic liver disease. The aim of this study was to evaluate the prognostic value of the MELD scoring as a predictor of fulminant hepatic failure (FHF) and death in patients with acetaminophen poisoning. Prospectively, serial measurements of the 3 MELD components—INR, bilirubin, and creatinine—were performed in 460 patients with acetaminophen-induced liver injury. Starting on the first day after the day of overdose, MELD score was significantly higher in patients who eventually developed hepatic encephalopathy (HE) than in those who did not. HE developed in 63 of 142 patients with a MELD score above 18 at 48-72 hours after the overdose (positive predictive value 44%) compared with 2 of 182 patients with a MELD score of 18 or below (negative predictive value 99%). Among 124 patients with FHF, a threshold MELD score of 33 on the day after the onset of HE had sensitivity of 60%, specificity of 69%, positive predictive value of 65%, and negative predictive value of 63%. However, the discriminative power of MELD score was not superior to that of INR alone or of the King's College Hospital criteria. Conclusion: MELD score may be useful as a predictor of FHF in patients admitted with acetaminophen toxicity. However, as a predictor of death from FHF, MELD score did not provide more information than the King's College Hospital criteria or INR alone. (HEPATOLOGY 2007;45:789–796.)
Acetaminophen poisoning is the single most common cause of drug-induced liver injury. The majority of cases of acetaminophen overdose are uncomplicated and the patient will recover without evident organ toxicity.1 A smaller proportion of patients (10%–15%) will show evidence of liver injury, and among these, approximately 20%–25% will progress into fulminant hepatic failure (FHF). The only treatment option that radically improves the outcome of FHF is emergency liver transplantation.2 However, a substantial proportion of patients with acetaminophen-induced FHF have the capacity for spontaneous recovery.
The clinician is required to make vital management decisions at different stages of the intoxication. Patients with uncomplicated acetaminophen poisoning are usually easily managed at the primary hospital. However, among the patients with hepatic injury, those considered at high risk of developing FHF should be transferred to a liver transplant center. At the transplant center, it remains a major clinical challenge to identify those patients who have the capacity to survive without liver transplantation. Among several selection criteria for urgent liver transplantation, the King's College Hospital (KCH) criteria have gained considerable acceptance.3–10 However, shortcomings of the KCH criteria have been a limited sensitivity and a late identification of the patients who require transplantation.11 It has therefore been of interest to discover prognostic indicators with the capability of identifying transplant candidates at an earlier stage of liver failure, perhaps even prior to the onset of hepatic encephalopathy (HE).10, 12
The Model for End-Stage Liver Disease (MELD) scoring system was originally designed for the assessment of short-term prognosis in patients with liver cirrhosis undergoing a transjugular intrahepatic portosystemic shunt (TIPS) to alleviate portal hypertension.13 Subsequently, its use has been expanded to predict survival in end-stage chronic liver disease in general, and the MELD score has been applied as a disease severity index in organ allocation decisions for liver transplantation in patients with chronic liver disease.14 However, lately there has been a growing interest for its use as a prognostic marker in FHF.15, 16
This study was intended to evaluate the capability of the MELD scoring system to predict the development of HE in patients with acetaminophen-induced hepatotoxicity and to predict spontaneous survival in patients with acetaminophen-induced FHF.
Patients and Methods
The study was performed prospectively in patients admitted with severe acetaminophen-induced hepatotoxicity at the Department of Hepatology, Rigshospitalet, Copenhagen University Hospital, between 1999 and 2004. Patients who fulfilled the Danish criteria for transfer to a tertiary liver center were eligible for inclusion. The Danish criteria for considering transfer to a tertiary liver center are either of the following: international normalized ratio (INR) above 1.7, creatinine above 300 mol/L, platelet count below 50 billion/L, arterial pH below 7.30 after fluid resuscitation, or the presence of HE.17 HE was graded according to the Fogarty criteria, and FHF was defined as the development of grade II-IV HE.18
For each case, daily measurements of bilirubin, INR, and creatinine were recorded. Whenever possible, INR values on a given day were obtained before infusion of plasma factors and creatinine values before dialysis. MELD score was calculated as 3.8 × loge(bilirubin [μM]) + 11.2 × loge(INR) + 9.6 × loge (creatinine [μM]) − 53.77.14 The values were related to the time of overdose (when known) and to the day of onset of HE (in patients with FHF). The term “day 0” refers to the first 24-hour period after overdose, “day 1” to the period 24–48 hours after overdose, and so forth.
The Department of Hepatology at Rigshospitalet, Copenhagen University Hospital, is the tertiary care center of liver disease in Denmark, and the majority of patients are transferred from other hospitals. All patients were treated and monitored in accordance with the department standard. In accordance with the Danish recommendation, a standard regimen of intravenous N-acetylcysteine for a minimum of 36 hours was initiated in all patients without delay upon their primary admission.17 Fresh frozen plasma was not used routinely for correction of clotting factors, but only administered in case of manifest bleeding. However, some patients with FHF were treated with high-volume plasmapheresis as part of an ongoing protocol. Plasmapheresis was only performed after the onset of HE. Patients fulfilling the KCH selection criteria were considered for urgent liver transplantation.3
Values are presented as mean ± SD. The Mann-Whitney test was used for the comparison of a variable between 2 subgroups, and Fischer's test was used for comparison of frequencies. Receiver operating characteristic (ROC) analysis was used to identify the threshold values. Validity of the model was determined by the concordance statistic (c-statistic, equivalent to the area under the ROC curve) with estimation of 95% confidence interval (CI) according to Hanley and McNeill.19, 20 In short, a c-statistic of 0.5 corresponds with a complete lack of discriminative power. A c-statistic greater than 0.7 is generally considered a useful test and a c-statistic greater than 0.8 indicates excellent diagnostic accuracy. Data were analyzed with the Statistica 6.0 statistical software (StatSoft Inc., Tulsa, OK). A 2-tailed P value below 0.05 was considered statistically significant.
A total of 460 consecutive patients with severe acetaminophen-induced hepatotoxicity were included in the study. The majority of 436 patients (95%) were transferred from other hospitals in accordance with the Danish transfer criteria. The remaining 24 patients (5%) were admitted primarily from the local region, but came to meet the same transfer criteria. There were 286 women (62%) and 174 men (38%). The median age was 40 years (interquartile range, 26–52 years; range, 12–86 years). In 403 cases (88%), the acetaminophen poisoning resulted from ingestion of a well-defined single overdose. In the remaining 51 cases (12%), the time of ingestion was unknown or the poisoning was the result of multiple or chronic overdosing. For the single overdose cases, the median acetaminophen overdose was 32 g (interquartile range, 23–50 g) and the median time from overdose to initiation of N-acetylcysteine treatment was 26 hours (interquartile range, 17–48 hours). In the following analyses, a composite endpoint of death or liver transplantation is used. However, performing the analyses with exclusion of the patients who underwent transplantation did not significantly change the findings (data not shown).
Of the 460 patients, 124 patients (27%) progressed into FHF.18 Forty-four of these 124 patients (35%) showed clear signs of spontaneous recovery, and all survived without being listed for orthotopic liver transplantation (OLT). Of the remaining 80 patients, 21 patients had a somatic contraindication to OLT (severe clinical instability, 14; malignancy, 3; significant heart disease, 2; chronic obstructive pulmonary disease, 1; dementia, 1), and 2 of these patients survived (9.5%). Further 41 patients had one or more psychiatric contraindications to OLT (ongoing alcohol or drug abuse, 37; chronic psychiatric disorder, 23; repetitive suicidal activity, 16), and 7 of these patients (17%) survived. The remaining 18 patients were listed for OLT; 8 patients eventually underwent transplantation, 5 patients died while waiting for a transplant, and 5 patients (28%) recovered spontaneously. In total, 58 (47%) of the 124 patients with FHF survived spontaneously, 58 patients (47%) died without liver transplantation, and 8 patients (6.5%) received emergency liver transplantation. Thirty-five (28%) of the 124 patients with FHF received treatment with high-volume plasmapheresis.
MELD Score as a Predictor of FHF.
Of the 124 patients who developed FHF, 28 patients (23%) presented with HE at primary admission. Consequently, the majority of 96 (77%) of the patients who developed FHF initially just presented with signs of hepatic injury without HE. Of these 96 patients, 27 patients developed HE prior to transfer to the liver center, whereas 69 patients only developed HE after transfer to the liver center. Sequential MELD score values relative to time of overdose (when known) are shown in patients with or without FHF (Fig. 1). From the first day (i.e., 24–48 hours) after the day of the acetaminophen overdose, MELD score was significantly higher in patients who eventually developed FHF than in those who did not. Similarly, significant differences were demonstrated in each of the 3 MELD components from day 1 (Fig. 2).
In order to be considered a predictor of FHF, MELD score is required to discriminate between groups prior to the expected onset of HE. Because the majority of patients developed HE on day 3 or later, we decided to evaluate MELD score on day 2 as a predictor of FHF. At this time point, data were available in 324 patients including 65 patients (20%) who eventually developed FHF. The distribution of MELD score and of each MELD component in patients with FHF versus those without FHF is shown in Fig. 3, whereas the corresponding ROC curves are shown in Fig. 4. The overlap between groups was visibly smaller for the MELD score than for the single components. This observation corresponded with a statistically significant difference between the c-statistic for MELD (0.92, CI 0.87–0.96) and each of the 3 components (INR 0.85, CI 0.79–0.91; bilirubin 0.81, CI 0.75–0.87; creatinine 0.77, CI 0.69–0.85). If cutoffs were selected to achieve a high sensitivity (e.g., 95%), the ROC curve illustrates that the loss of specificity was smaller for MELD score than for each of the 3 components (Fig. 4). For MELD score, a cutoff of 18 was required to identify 95% of the FHF patients. HE developed in 63 of 142 patients with a MELD score above 18 [positive predictive value (PPV) 44%] compared with 2 of 182 patients with a MELD score of 18 or below (negative predictive value 99%). Similarly, HE developed in 63 of 210 patients with an INR above 2.6 (PPV 30%), in 62 of 207 patients with a bilirubin above 32 μmol/L (PPV 30%), and in 63 out of 309 patients with a creatinine above 55 μmol/l (PPV 20%). Thus, using MELD rather than INR as a screening indicator for FHF reduced the number of false positive patients from 147 to 79. Bilirubin was similar in discriminative power to INR, whereas creatinine was obviously inferior.
MELD Score as a Predictor of Death.
Among the 124 patients with FHF, MELD score was significantly lower in survivors than in nonsurvivors starting on the day after the onset of HE (Fig. 5). Accordingly, there was no significant difference in MELD score between nonsurvivors and survivors on the day of onset of HE (32.7 ± 9.2 versus 30.6 ± 7.7). However, starting on the day of onset of HE, MELD score consistently increased in nonsurvivors and decreased in survivors (ANOVA: P < 0.00001). In order to further quantify this trend, a ΔMELD was defined as the difference in MELD score from day 0 to day 1 after the onset of HE. ROC analysis revealed that ΔMELD was a better discriminator than just the MELD score at day 1 (c-statistic: 0.78, CI 0.70–0.87 versus 0.68, CI 0.57–0.78; P = 0.03), whereas MELD score on the day of onset of HE had no discriminative power (c-statistic: 0.58, CI 0.47–0.69).
When the components of the MELD score were examined individually (Fig. 6), none of the 3 components was able to discriminate between survivors and nonsurvivors until after the onset of HE. Most noticeably, there was no significant difference in INR between nonsurvivors and survivors on the day of onset of HE (4.9 ± 1.9 versus 4.9 ± 2.0). INR could be identified as the qualitatively most important contributor to the change in MELD score after onset of HE. When Δbilirubin, ΔINR, and Δcreatinine were defined as the difference in the respective values from day 0 to day 1 after the onset of HE, the c-statistic of ΔINR (0.76, CI 0.66–0.87) was not significantly different from that of ΔMELD, whereas the c-statistic of Δcreatinine was significantly lower (0.65, CI 0.54–0.75). ΔBilirubin completely lacked discriminative power (c-statistic: 0.52, CI 0.41–0.63). The ROC analysis identified a ΔMELD of −0.4 and a ΔINR of −0.3 as the optimum discriminative thresholds. Applying these thresholds, both ΔMELD and ΔINR appeared to be equal in prognostic value to the KCH criteria at this time point (Table 1).
|MELD > 32 at day of HE (day 0)||63/121||37||58||54||59||53|
|MELD > 33 at day +1 of HE||52/115||34||60||69||65||63|
|ΔMELD > −0.4 from day 0 to day +1||63/113||46||82||70||73||80|
|ΔINR > −0.3 from day 0 to day +1||56/113||41||73||74||73||74|
|KCH criteria at day of HE||46/124||36||55||83||78||62|
|KCH criteria at day +1 of HE||50/124||38||58||79||76||62|
|KCH criteria at any time||56/124||44||67||79||79||68|
In this study, MELD score was strongly associated with development of HE in patients with severe acetaminophen-induced hepatotoxicity. In patients with FHF, MELD score at the onset of HE did not discriminate between survivors and nonsurvivors, whereas an increase in MELD score (ΔMELD) after the onset of HE was associated with a poor prognosis. MELD score showed no advantages over INR or existing selection criteria for identifying nonsurvivors.
The MELD score has been established as a reliable measure of short-term mortality risk in adult patients with end-stage chronic liver disease.14 In a recent study in OPTN/UNOS Status 1 patients, MELD score was a significant predictor of overall survival in patients with non–acetaminophen induced FHF.15 The study was unable to identify any significant association between survival and MELD score in patients with acetaminophen-induced FHF, which may have been due to a low number of events. In addition, a preliminary report suggested the MELD score may also be useful in the setting of mainly non–acetaminophen induced FHF to complement other prognostic criteria.16 This study was therefore performed in order to evaluate the MELD scoring system, particularly in patients with acetaminophen-induced hepatotoxicity.
Our results illustrate that acetaminophen-induced hepatotoxicity is a dynamic condition. Because this dynamic nature involves all 3 MELD components, timing of the observation is important for the meaningful interpretation of MELD score at a given time. In particular when assessing a single observation, the time of observation should be standardized in relation to a well-defined time point of the disease (e.g., time of overdose or onset of HE). A continuous scale scoring system such as MELD is potentially more versatile than dichotomous tests such as the KCH criteria. In theory, different cutoffs for the MELD score may be applied at different stages of disease progress, in different populations, and for different indications. Thus, MELD score may be applied in patients with liver injury before the onset of HE in order to identify a high-risk subgroup that should be transferred to a specialized liver unit, whereas MELD score in patients with FHF may assist in identifying suitable candidates for liver assist treatment or even liver transplantation. Furthermore, a continuous scale is a requirement for a meaningful calculation of changes in score such as the suggested ΔMELD.
In FHF, the onset of HE is always preceded by biochemical signs of progressive liver injury and dysfunction. Thus, even though it is the defining feature of FHF, HE may be regarded as a late event in the disease process.21, 22 In the management of patients with FHF, it is essential patients be referred to specialist liver units as early as possible in order to identify and treat deleterious aggravating cofactors.22, 23 Awaiting the development of HE would delay referral unacceptably, and the challenge at the referring hospitals must be to identify and transfer patients at a high risk of FHF before onset of HE.22, 24 Surprisingly, very few studies focused on prediction of HE in acetaminophen poisoning, and suggested criteria for when to contact a specialist center tend to be based on expert opinion rather than published data.12, 17, 25, 26 In this study, a difference in MELD score was already detectable 24 to 48 hours after ingestion of the acetaminophen overdose between patients who eventually developed FHF and those who did not. Because all 3 MELD components contributed to the difference between groups, the MELD score was significantly better than each single component at predicting FHF. The data suggest that MELD score may be useful in the pre-encephalopathic stage of acetaminophen-induced liver injury as a screening indicator for identification of a subgroup of patients with a high risk of developing FHF.
In the patients with FHF, neither MELD score nor any of its 3 components were able to discriminate between survivors and nonsurvivors at the time of onset of HE. Because the decision whether to list a patient for emergency liver transplantation typically has to be made at this time point, MELD score appears to be of little value in identifying candidates for liver transplantation. MELD score did tend to increase after the onset of HE in nonsurvivors and decrease in survivors. This trend could be quantified as a ΔMELD, which contained more information than MELD score alone at any of the 2 time points. The c-statistic of ΔMELD of 0.78 would suggest that ΔMELD was a valuable prognostic test. However, ΔMELD was not superior to ΔINR, which is much easier to calculate. The association between a decrease in INR and a favorable prognosis is already well established. Coagulation factors are among the most important prognostic indicators in acetaminophen-induced FHF.3, 27, 28 In particular, a continued improvement in serial determinations of prothrombin (corresponding with a negative ΔINR) is strongly indicative of recovery.29, 30 In reality, the clinical implications of our findings on ΔMELD are that patients who continually deteriorate after the onset of HE are likely to die, whereas patients who improve are likely to survive. In most cases, the experienced clinician will have reached this conclusion without the need for calculating a MELD score.
Of the 3 MELD components, INR was the quantitatively most important, whereas bilirubin was the least contributory. The development of renal failure is an established prognostic indicator in acetaminophen-induced FHF, and creatinine and well as INR are incorporated into the KCH criteria.3, 31 In contrast, bilirubin, which is an important risk factor in non–acetaminophen induced FHF, has never been equivocally established as a risk factor in acetaminophen-induced FHF.3, 32 Indeed, bilirubin has been suggested to be higher in survivors than in nonsurvivors.3 This was not the case in our study, because a more pronounced increase in bilirubin was observed in nonsurvivors than in survivors. However, the difference was rather small and appeared too late in the clinical course to be of significant clinical value.
A practical limitation of the use of MELD score in FHF is that the 3 components of the MELD score may be artificially improved by treatment modalities such as renal replacement therapy, plasma exchange treatment, and artificial liver support. This artificially achieved biochemical improvement will result in a relative decrease in MELD score that does not reflect recovery of organ function and should be interpreted with caution. However, a high or increasing MELD score will still be associated with a poor prognosis, whereas a spontaneous decrease in MELD score will be associated with a good prognosis. These considerations are, of course, not specific to the MELD scoring system, but will apply to all scoring systems, including the KCH criteria.
No significant difference could be established between the existing KCH criteria and ΔMELD or ΔINR in this study. The study also illustrated the 2 main problems with the KCH criteria: their sensitivity is limited and several patients are only identified at a late stage, when their clinical condition may be too advanced to allow liver transplantation.11 Attempts have been made to amend these drawbacks by proposing the addition of sensitive and early prognostic markers such as lactate and phosphate.10, 12, 33–35 However, none of these markers have been generally accepted, and there is still a need for better prognostic indicators in acetaminophen poisoning.
In conclusion, MELD score may be useful in the pre-encephalopathic stage as a predictor of acetaminophen-induced FHF. At the time of onset of HE, MELD score failed to discriminate between survivors and nonsurvivors from FHF, which limits its use for identification of liver transplantation candidates. However, a decrease in MELD score after the onset of HE was strongly indicative of survival from acetaminophen-induced FHF.
- 16Application of MELD scoring system to predict prognosis in patients with acute liver failure [Abstract]. HEPATOLOGY 2003; 38(Suppl 1): 554A., , , .
- 18Diseases of the Liver and the Biliary Tract. New York: Raven Press, 1994., , , .