Predictive Indices of Morbidity and Mortality After Liver Resection

To the Editor:

Dr. Schroeder et al evaluated the predictive indices of the Model for End-Stage Liver Disease (MELD), Child-Turcotte-Pugh scores, and the American Society of Anesthesiology physical status classification on morbidity and mortality for patients after hepatic resection.¹ American Society of Anesthesiology and Child-Turcotte-Pugh scores were predictive of mortality but not morbidity, but MELD had no predictive value. Although we agree with the authors that the application of MELD as a prognostic tool for patients other than those awaiting hepatic transplantation requires further investigation before clinical application, we dispute the conclusion that MELD should not be used in the setting of elective hepatic resection.

MELD was specifically designed to predict liver failure specific mortality in patients with end-stage liver disease (cirrhosis) after transjugular intrahepatic portosystemic shunt.² The model has been repeatedly, independently validated in this clinical setting and has become the primary method for stratifying candidacy for hepatic transplantation.³ Importantly, MELD was developed as a reliable objective method to determine mortality risk in patients with cirrhosis only. The applicability of MELD in patients without cirrhosis, regardless of intervention, is unknown. Indeed, MELD would not be expected to stratify patients without end-stage or chronic established liver disease because the model was developed in patients with cirrhosis. Although the analysis performed by Schroeder et al was detailed extensively, we think that the conclusion regarding MELD cannot be drawn without data on the specific number of patients with cirrhosis and the exclusion of patients with chronic renal failure and anticoagulants that affect serum creatinine and INR (essential components of MELD score calculation). Indeed, analysis between patients with and without cirrhosis (CPT codes) is warranted, especially detailing the cause of death, liver failure, or otherwise.

Table 2 showed that 166 patients had primary liver malignancies and were at risk for cirrhosis. The remaining 361 patients with metastatic and benign liver tumors were at expectedly far less risk for cirrhosis and liver failure-related morbidity. Moreover, if the majority of patients did not have cirrhosis, their MELD score is likely to be low; therefore, MELD may not be discriminatory. Although the 50 patients who died were deleted from the distribution of tumor type, the sample size should provide basis to assess the predictive value of MELD after hepatic resection for patients with and without cirrhosis. In contrast to the authors’ findings, we have shown that MELD is predictive of postoperative mortality after hepatic resection for patients with cirrhosis and hepatocellular cancer.⁴ Others have also shown that MELD is predictive of perioperative mortality after other operations in patients with cirrhosis.^5–8

Tables 5 and 6 summarize the predictive accuracy of the indices for 30-day mortality and morbidity. None of the indices had a receiver operating area under the curve (ROC-AUC) greater than 0.7, the level above which an index has clinically useful predictive accuracy.⁹ Thus, the only clear conclusion that can be drawn from this study is that there is no index that is clinically useful in predicting operative risk, either mortality or morbidity, in this group of patients. The data also confirm that it is difficult to make meaningful conclusions from retrospective studies in heterogeneous groups of patients drawn from a large number of institutions, which may have varying levels of surgical expertise.

We think that MELD can be clinically useful in counseling patients and families on the risks of operative mortality after elective hepatic resection in patients with cirrhosis. Although the authors failed to show the predictive value of MELD in the setting of elective hepatic resection, we think that clarification of their data by the presence or absence of cirrhosis and exclusion of patients with chronic renal failure and systemic anticoagulation is necessary before the predictive utility of MELD is dismissed.

David M. Nagorney, MD
Patrick S. Kamath, MD