Prognostic Role of MELD-Lactate in Cirrhotic Patients' Short- and Long-Term Prognosis, Stratified by Causes of Cirrhosis

Objectives Recently, model for end-stage liver disease-lactate (MELD-LA) proved to be a superior predicting factor of inpatient mortality in patients with chronic liver disease. The study's objective was to evaluate the ability of MELD-LA to predict both short- and long-term mortality in critically ill cirrhotic patients stratified by causes of cirrhosis. Materials and Methods This was a retrospective observational research of 469 cirrhotic patients entering intensive care unit. Clinical parameters and prognostic scores were measured and collected in the first 24 hours after entering intensive care unit. Follow-up duration was at least 5 years. Independent relationship between MELD-LA and mortality was evaluated by multivariate logistic regression analyses. Discrimination of scoring system was evaluated by the area under the receiver operating characteristic curve. Calibration of the score was evaluated by Hosmer-Lemeshow goodness of fit test for significance. Results The MELD-LA score (odds ratio: 1.179, 95% confidence interval: 1.112–1.250, P < 0.001) was an independent risk factor for 15-day mortality. The area under the curve of MELD-LA was the highest (0.808, 95% confidence interval: 0.765–0.852) in predicting 15-day mortality and it had superior calibration. We found MELD-LA showed the best discrimination ability in cirrhotic patients caused by both alcohol and hepatitis (0.783, 95% confidence interval: 0.651–0.915) or alcohol alone (0.805, 95% confidence interval: 0.743–0.867). Conclusions MELD-LA performs better for predicting short-term prognosis in critically ill cirrhotic patients, especially caused by both alcohol and hepatitis or alcohol alone.


Introduction
Chronic liver diseases can finally develop into liver cirrhosis and the process is irreversible [1]. Long-term liver fibrosis can lead to severe liver dysfunction and portal hypertension [2]. Patients with decompensated cirrhosis are usually accompanied by single or multiple organ failure [3][4][5]. e health burden of cirrhosis is increasing worldwide [6]. In compensated cirrhotic patients, unstable clinical conditions are often caused by life-threatening complications such as variceal bleeding, hepatic encephalopathy, hepato-renal syndrome, ascites, infection, and sepsis, which require entering an intensive care unit (ICU) [7,8]. Mortality rates for cirrhotic patients in ICUs can range from 34% to 86%. erefore, in order to select the most appropriate treatment as soon as possible and improve critically ill patients' prognosis, there is urgent need to find simple and practical forecasting methods that may help evaluate the severity of cirrhotic patients.
Model for end-stage liver disease-lactate (MELD-LA), a recently-developed clinical score, is a much better predicting factor of in-hospital mortality in cirrhotic patients than MELD alone [9]. MELD-LA performs significantly better compared with MELD for predicting in-hospital mortality in patients hospitalized for infection [10]. However, little was known about the ability of MELD-LA to predict short-and long-term prognosis in critically ill cirrhotic patients. As previously described, alcohol may be a coetiology in patients with viral (HBV and HCV)-related chronic liver diseases and ethanol intake is an independent predictor of cirrhosis in subjects with a chronic viral infection and an independent predictor of death in subjects with either HCV or HBV infection [11]. But, it is unclear if there are classes of cirrhotic patients with different causes (e.g., alcohol and hepatitis) where MELD-LA may be more useful or less. To address these problems, we aimed to evaluate the value of the MELD-LA score in predicting critically ill cirrhotic patients' short-and long-term mortality, stratified by causes of cirrhosis, in comparison to chronic liver failure-sequential organ failure assessment (CLIF-SOFA) [12], SOFA [13], Child-Pugh system [14], and MELD score [15].

Study Population.
Patient dataset was extracted from the Medical Information Mart for Intensive Care III, an openaccess database [16]. We completed "Protecting Human Research Participants" which was the National Institute of Health's training course (certificate number: 36072928) and then obtained access to the database. e database had been approved to establish by the institutional review boards of the Massachusetts Institute of Technology, in which all information associated with patients was anonymous, and this study we conducted was a retrospective observational research.
us, ethics committee approval and informed consent were not necessary. We selected patients aged at least 18 years and stayed more than 24 h. Diagnostic criteria for cirrhosis were as follows: (i) clinical symptoms such as jaundice and ascites, (ii) laboratory test results such as prothrombin time prolongation and albumin reduction, (iii) imaging features, and (iv) histopathology [17,18]. Patients were excluded for the following reasons: (i) malignancy, (ii) previous liver transplantation surgery, (iii) human immunodeficiency virus infection, and (iv) total bilirubin, international normalized ratio, lactate, or creatinine data lost. e study population included 469 patients after applying exclusion criteria.

Data Collection.
Clinical parameters were collected including demographic information, vital signs, and laboratory parameters such as glucose, international normalized ratio, hemoglobin, prothrombin time, platelet count, arterial blood lactate, total bilirubin, albumin, and creatinine. e calculations of scoring systems including CLIF-SOFA, MELD, SOFA, Child-Pugh, and MELD-LA scores used the mean value of each clinical parameter within the first 24 h after entering ICU, using the formulae published [12][13][14][15]. e MELD-LA score was calculated as follows: 0.251 + 5.5257 × sqrt (lactate) + 0.338 × MELD [9]. Followup began the day entering ICU and lasted for at least 5 years. 15-day and 5-year all-cause mortality were main outcomes.

Statistical Analysis.
Quantitative variables were presented median (interquartile range (IQR)) and compared by Mann-Whitney U test. Categorical variables were expressed absolute numbers (frequencies) and compared by chi-square test or Fisher's exact test. Independent relationship between parameters and mortality was assessed by multivariate logistic regression analyses. Odds ratio (OR) was reported with 95% confidence interval (CI). Calibration of the score was assessed by Hosmer-Lemeshow goodness of fit test for significance (p > 0.05). Discrimination of the score was assessed by the area under the receiver operating characteristic curve. DeLong test was used to perform the comparison between area under the curve (AUC) [19]. Sensitivity and specificity at the optimal cut-off value were compared in various scoring systems, and we stratified patients into three groups (alcoholic and hepatitis; alcoholic; alcoholic) by causes of cirrhosis. Besides, patients included were regrouped into two classes (relatively low and high risk) by MELD-LA score's optimal cut-off value. All the tests were two sided. P value <0.05 indicated statistical significance. All statistical analyses used STATA (version 14.0; StataCorp, State of Texas, USA).

Baseline Characteristics.
Our study included a total of 469 patients. e majority of the study population were male and white. Alcohol was the primary cause of cirrhosis. ere were three patients that had three causes and 60 patients that had two. e median age of all the participants was 55.6 years. e most common causes of ICU hospitalization were infection and sepsis. e survivors and nonsurvivors were not significantly different in glucose, age, body mass index, sex, ethnicity, causes of cirrhosis, comorbidity, albumin, and length of ICU stay (P ≥ 0.05). Nonsurvivors had higher scores than survivors in MELD-LA, CLIF-SOFA, SOFA, MELD, and Child-Pugh scores (all P < 0.001). e details of the patients' baseline characteristics in both survivor and nonsurvivor cohorts are presented in Table 1.

Discussion
Our study evaluated for the first time the ability of MELD-LA to predict both short-and long-term mortality in critically ill cirrhotic patients stratified by causes of cirrhosis. e research showed that cirrhotic patients admitted to ICU still had high mortality despite aggressive medical interventions, as has been reported before [3][4][5]20]. erefore, it is very important for risk assessment, optimal treatment selection, prolonging survival time, and improving survival quality, to have early, accurate, and objective predicting tools of mortality with accessible variables.
Recently, the MELD-LA score proved to be an early and objective predicting factor of in-hospital mortality in cirrhotic patients [9,10]. We further investigated its value in predicting critically ill cirrhotic patients' short-and longterm prognosis, stratified by different etiologies. Our study confirmed that the MELD-LA score showed optimal discrimination value in predicting critically ill cirrhotic patients' short-term prognosis, especially caused by both alcohol and hepatitis or alcohol alone. However, for predicting long-term prognosis, MELD performed better. e prediction of critically ill cirrhotic patients' short-term mortality can be enhanced by lactate while the value of lactate in predicting long-term mortality requires further research [21]. MELD-LA score was related to lactate, may resulting in the ability of MELD-LA to predict long-term mortality worse than short-term mortality. Besides, further study is needed to carry out on the reasons for the poor value of MELD-LA in predicting short-term prognosis in critically ill cirrhotic patients caused by hepatitis alone as compared to patients due to other etiologies. Another clinically relevant study result is the statistically significant difference between survivors and nonsurvivors in terms of platelet count. is may be due to different bleeding risks caused by different platelet counts, while there are other studies that show that platelet count does not predict unprovoked major or minor bleeding in cirrhotic patients [22]. us, further research is needed. Some potentially clinical applications of MELD-LA are described as follows. With lactate being associated with acute hepatic impairment [23][24][25][26] and used to assess the disease's severity in critically ill patients [27][28][29][30][31], the inclusion of lactate can more precisely show systemic lesions occurring in cirrhotic patients. MELD-LA score at admission can promptly and accurately assess the severity of cirrhosis and may be useful for stratifying patients that require higher levels of care or earlier interventions. Cirrhotic patients with an MELD-LA score >15 have extremely severe hepatic failure and higher short-term risk of death. An MELD-LA    score >15 may indicate that the patient need liver transplantation. MELD-LA scores during hospitalization may help to identify patients that are not responding well to current treatment, which may allow for discussing treatment adjustment or palliative care earlier.
However, our research has limitations. is was a retrospective research conducted in a single institution. A future prospective multicentered study is needed. Besides, part of classic clinical scores were included, but others were excluded. In addition, the mortality was defined as all-cause mortality so it may be affected by other causes of death. Finally, therapeutic measures were not taken into account such as anticoagulant treatment, which may affect prognosis of patients. Anticoagulant treatment to treat portal vein thrombosis potentially improved the survival of patients with cirrhosis and such a complication [32]. We will conduct studies to solve these problems in the future.

Conclusions
e MELD-LA score, a recently-developed scoring system, has significantly superior performance in predicting shortterm prognosis in critically ill cirrhotic patients, especially caused by both alcohol and hepatitis or alcohol alone. For predicting long-term prognosis, MELD performs better. Moreover, the MELD-LA score's potentially clinical applications need our further consideration and exploration.

Data Availability
e data used to support the findings of this study are available from the author upon request.

Conflicts of Interest
e author declares that there are no conflicts of interest regarding the publication of this paper.

Authors' Contributions
Xiao-Fu Chen put forward the research plan; was involved in data collection and analysis; wrote the manuscript; and was involved in critical review and revision of the manuscript. e author agreed to submit the manuscript.