Summary and Historical Background
Summary
Background: Prognosis in cirrhotic patients has had a resurgence of interest because of liver transplantation and new therapies for complications of end-stage cirrhosis. The model for end-stage liver disease score is now used for allocation in liver transplantation waiting lists, replacing Child-Turcotte-Pugh score. However, there is debate as whether it is better in other settings of cirrhosis.
Aim: To review studies comparing the accuracy of model for end-stage liver disease score vs. Child-Turcotte-Pugh score in non-transplant settings.
Results: Transjugular intrahepatic portosystemic shunt studies (with 1360 cirrhotics) only one of five, showed model for end-stage liver disease to be superior to Child-Turcotte-Pugh to predict 3-month mortality, but not for 12-month mortality. Prognosis of cirrhosis studies (with 2569 patients) none of four showed significant differences between the two scores for either short- or long-term prognosis whereas no differences for variceal bleeding studies (with 411 cirrhotics). Modified Child-Turcotte-Pugh score, by adding creatinine, performed similarly to model for end-stage liver disease score. Hepatic encephalopathy and hyponatraemia (as an index of ascites), both components of Child-Turcotte-Pugh score, add to the prognostic performance of model for end-stage liver disease score.
Conclusions: Based on current literature, model for end-stage liver disease score does not perform better than Child-Turcotte-Pugh score in non-transplant settings. Modified Child-Turcotte-Pugh and model for end-stage liver disease scores need further evaluation.
Historical Background
In 1964, Child and Turcotte published a classification to assess the operative risk in cirrhotic patients who recovered from variceal bleeding, undergoing portosystemic shunt surgery. They considered five variables selected by clinical experience: ascites, encephalopathy, nutritional status and serum levels of bilirubin and albumin, classifying patients in class A, B or C in relation to best (A), moderate (B), or worse (C) prognosis.
In 1973, Pugh et al. used a modified version of this classification for patients undergoing surgical transection for oesophageal varices. They replaced nutritional status with prothrombin time (PT) and assigned a score ranging from 1 to 3 to each variable. Subsequently, this classification was used to predict the outcome of surgery in cirrhotic patients in general, and more recently, to stratify patients on the waiting list for liver transplantation (LT). Similar to the original Child's grading, neither the division of the grades nor the scoring system was validated statistically.
Over many years, the routine use of these classifications have revealed some problems. First, both the degree of ascites and encepahlopathy are subjective assessments evaluated by physical examination alone. The widespread use of ultrasonography has led to more sensitive detection of ascites, but it is unclear how ascites diagnosed by ultrasonography alone is being categorized by various authors. Hepatic encephalopathy (HE) is often assessed by psychometric testing or slowing of frequency on an electroencephalography (EEG). It is not clear how these abnormalities are fitted into Child-Turcotte-Pugh (CTP) classification. Secondly, both ascites and HE can be influenced by therapy such as diuretics, albumin infusion and lactulose and it is not clear if ascites and HE are scored at their best, or worst, or independent of specific therapy.
As regards the continuous variables in the CTP classification, they are categorized with arbitrary cut-off points. Thus, patients with bilirubin of 55 µM who have a better prognosis than those with a bilirubin of 250 µM; in the CTP classification both these patients have the same score of severity for bilirubin concentration ('the ceiling effect'). Changes in bilirubin concentration with specific therapy such as ursodeoxycholic acid (for which improvements is in doubt) are difficult to interpret using CTP classification. A similar problem exists for serum albumin so that the CTP classification does not differentiate between patients with an albumin of 17 g/L vs. 25 g/L ('the floor effect'). The increased use of intravenous albumin if given close to an assessment may further complicate the interpretation.
Pugh's modification of Child-Turcotte criteria substituted PT for nutrition. However, the PT (expressed in s) as well as the prothrombin index (expressed as a percentage of the control value), varies depending on the sensitivity of the thromboplastin reagent used so that it can vary greatly from laboratory to laboratory. CTP scoring for PT does not take this into account. The International Normalized Ratio (INR) now overcomes this using a standardized thromboplastin reagent. The conversion from INR to PT and vice versa using a normogram is not used routinely and INR has not been incorporated into CTP score. Although it would seem sensible to use INR, it should be remembered that INR was designed to standardize the anticoagulation effect of warfarin and not to evaluate the severity of liver disease. As a result, INR may not be valid to assess liver impairment, despite its use in model for end-stage liver disease (MELD) scoring (see below).
Partly due to the grading system, the CTP classification does not distinguish the 'mild' grade C from the severe grade C patient with sufficient clinical discrimination. Moreover, it does not include a measure of renal function, which is a well-established prognostic marker in cirrhosis, as well as in acute liver disease.
During recent years, several new prognostic models have been developed and validated in order to overcome these limitations of the CTP score. The performance of the majority of these models was evaluated statistically by measurement of their discriminative ability, estimated by the concordance (c)-statistic [i.e. the area under the Receiver Operator Characteristic (ROC) curve]. The latter describes the ability of the model to separate patients who die, from those with the same score who live. The c-statistic ranges from 0 to 1, with 0.5 corresponding to what is expected by chance alone and 1.0 to perfect discrimination. In general, a c-statistic >0.7 indicates a useful test, whereas a value >0.8 indicates very good prognostication, but can never be 1.0. For this reason, prognostic models cannot predict the outcome of individual patients but give estimates: e.g. for a model with an area under the ROC curve 0.8-0.9, by definition a significant proportion of the patients (10-20%) has an outcome which is not predicted accurately.
In 2000, Malinchoc et al. published an article on 'a model to predict poor survival in 231 patients who had undergone transjugular intrahepatic portosystemic shunt (TIPS)'. They developed a statistical model to (i) predict survival and (ii) identify those patients whose liver-related mortality post-TIPS would be 3 months or less. Cox proportional hazard regression identified serum concentration of bilirubin and creatinine, the INR (for PT), and the aetiology of cirrhosis as predictors of survival. In particular, these variables were used to calculate a risk score for undergoing elective TIPS by combining their prognostic values with their regression coefficient: R = 0.957 × loge (creatinine mg/dL) + 0.378 × loge (bilirubin mg/dL) + 1.120 × loge (INR) + 0.643 × (cause of cirrhosis: 0 for alcohol related and cholestatic liver disease; 1 for viral hepatitis and other liver disease). This model was found to be superior to both CTP classification and score in predicting survival, especially in patients with CTP class B with impaired renal function.
In 2001 the same group, used this MELD, to score severity of liver disease and consequent risk of mortality for patients awaiting LT. The MELD score is slightly modified from the original TIPS risk score, multiplying the score by 10 and then rounding up to the nearest integer. The MELD score formula is: R = 9.6 × loge (creatinine mg/dL) + 3.8 × loge (bilirubin mg/dL) + 11.20 × loge (INR) + 0.64 × (cause of cirrhosis: 0 for alcohol related or cholestatic liver disease; 1 for viral hepatitis and other liver disease).
The model's validity was tested with data obtained from different patient populations including (i) patients hospitalized for hepatic decompensation, (ii) ambulatory patients with non-cholestastic cirrhosis, (iii) patients with primary biliary cirrhosis (PBC) and unselected patients from the 1980s with cirrhosis ('historical' patients). In these groups the model's ability to accurately predict death within 3 months was evaluated using the c-statistic.
The MELD score compared with the CTP score was able to predict death within 3 months with a c-stastistic of (i) 0.87 for hospitalized patients (compared with 0.84 for CTP score), (ii) 0.80 for non-cholestatic ambulatory patients, (iii) 0.87 for PBC patients and (iv) 0.78 for 'historical' cirrhotic patients. The authors evaluated the addition of a history of important complications of portal hypertension such as ascites, encephalopathy, variceal bleeding and spontaneous bacterial peritonitis (SBP) to the MELD score for each patient. However, only a minimal improvement in the prediction of the 3-month mortality was obtained. Importantly, no calibration of the model was evaluated, i.e. an assessment of whether it performed uniformly well across the spectrum of severity of liver disease.
In the original prognostic model, designed for patients selected for TIPS, cholestatic and alcoholic liver disease were given a lower risk score. In cholestatic disease, the bilirubin concentration has a different association with liver dysfunction [and reduction with ursodeoxycholic acid (UDCA) therapy] compared with other liver diseases. This was also true in the Child-Pugh (CP) modification in which the A, B, C categories have different bilirubin boundaries. The modified model that excluded the aetiology of liver disease found that in the four patient validation sets, which there was minimal change of the c-statistic, and so aetiology was dropped. However, there remains an issue of whether individuals with diseases in which bilirubin falls with therapy, e.g. ursodeoxycholic acid in primary sclerosing cholangitis (PSC) and PBC, do have an improved prognosis or not, whereas in alcoholic cirrhotics with abstinence, even if initially very jaundiced, prognosis improves.
The authors concluded that MELD score showed advantages compared with the CPT score and proposed the MELD score as criterion for organ allocation for LT. An editorial agreed with them '... MELD score, with its applicability to the pretransplantation prognosis, easy to use and verifiability, is a useful addition to the array of prognostic instruments and appears likely to dislodge the CP system from its perch'.
The study by Wiesner et al., published in 2003, confirmed the predictive accuracy of MELD score for short-term (3 months) outcome in patients on the liver transplant waiting list and could be applied to the liver allocation system. Although one must bear in mind that in this context, allocation by MELD score represents a justice and not a utility system, the first results after MELD implementation are satisfactory: for the first time after 20 years, there was a reduction in the number of patients on the waiting list and a significant decrease in the presence of severe HE at the time of LT, which suggested that HE was exaggerated by transplant centres, when CTP score was used for liver allocation. Furthermore, there was a decrease in new registrations and in mortality of patients on the waiting list and an increase in the number of cadaveric transplants for almost all aetiologies.
Despite its particular and specific development in the area of liver donor allocation, MELD has also been reported as the novel and better prognostic score to CTP for many complications or therapies in cirrhosis. Is this enthusiasm justified particularly as MELD was developed solely to evaluate 3-month mortality to patients already listed for LT? In this review, we examine the evidence for the prognostic utility of MELD compared with CTP score in non-transplant settings.