Original Contribution

Early risk stratification with simple clinical parameters for cirrhotic patients with Acute upper gastrointestinal bleeding

Yao-Chun Hsu MD^a,b, Jyh-Ming Liou MD^c, Chen-Shuan Chung MD^c, Cheng-Hao Tseng MD^d, Tzu-Ling Lin MD^c, Chieh-Chang Chen MD^e,

Ming-Shiang Wu MD, PhD^c, Hsiu-Po Wang MD^c,e,?

^aDivision of Gastroenterology and Hepatology, Department of Internal Medicine, Lotung Poh-Ai Hospital, Yilan, Taiwan

^bGraduate Institute of Clinical Medicine, National Taiwan University College of Medicine, Taipei, Taiwan

^cDepartment of Internal Medicine, National Taiwan University Hospital, National Taiwan University College of Medicine,

Taipei, Taiwan

^dDepartment of Internal Medicine, Tao Yuan General Hospital, Tao-Yuan, Taiwan

^eDepartment of Internal Medicine, National Taiwan University Hospital Yun-Lin Branch, Yun-Lin, Taiwan

Received 5 April 2009; revised 24 April 2009; accepted 24 April 2009

Abstract

Objective: This study aimed to identify pre-endoscopic clinical parameters independently associated with 6-week mortality and to develop a prognostic model in cirrhotic patients with acute upper gastrointestinal (UGI) bleeding.

Methods: A total of 542 consecutive admissions of 389 cirrhotic patients with acute UGI bleeding were retrospectively investigated. Pertinent clinical data obtained at the emergency department were analyzed. Multivariate logistic regression analysis was performed to determine risk factors for 6-week mortality and to develop a predictive model.

Results: Forty-four patients (8.12%) died within 6 weeks. The 6-week mortality was independently associated with male sex, hepatocellular carcinoma, non-hepatocellular carcinoma malignancy, hypoxemia with peripheral oxygen saturation less than 95%, serum bilirubin, and prothrombin time. A predictive model consisting of these 6 simple parameters was built. The c statistic of our model was 0.84, significantly superior to that (0.71) of the model for end-stage liver disease score (P = .002).

Conclusions: Simple pre-endoscopic clinical parameters are valuable for early risk stratification in cirrhotic patients with acute UGI bleeding. Our prognostic model warrants prospective validation by further studies.

(C) 2010

Introduction

* Corresponding author. Department of Internal Medicine, National Taiwan University Hospital, Taipei 100, Taiwan. Tel.: +886 2

23123456×5695; fax: +886 2 23947899.

E-mail address: [email protected] (H.-P. Wang).

Acute upper gastrointestinal (UGI) bleeding is a life- threatening emergency in patients with liver cirrhosis with a mortality rate as high as 15% to 20% [1,2]. Because these patients are usually managed initially at the emergency department (ED), it is important for ED physicians to rapidly

0735-6757/$ - see front matter (C) 2010 doi:10.1016/j.ajem.2009.04.037

stratify patients at different risks of mortality. An ideal risk stratification system in the ED is built on simple, objective, and readily available parameters without compromising its accuracy. Several risk stratification schemes have been developed and validated for general patients with acute UGI hemorrhage [3-5], but they were not dedicated specifically to those with cirrhosis.

prognostic factors.for cirrhotic patients with all sources of acute UGI bleeding have been sparsely studied. Previous studies have focused on acute variceal hemor- rhage (AVH), the most common etiology of acute UGI bleeding in cirrhotic patients, which accounted for 70% to 80% of the bleeding episodes [1,6]. Hematocrit, serum aminotransferase, serum D-dimer, renal function, hepato- cellular carcinoma (HCC), bacterial infection, hepatic vein pressure gradient, bleeding at endoscopy, portal vein thrombosis, Child-Turcotte-Pugh (CTP) class, and model for end-stage liver disease score have been demonstrated as predictors of clinical outcomes in patients with AVH [1,7-12]. However, these prognostic factors may not be conveniently applied in the ED, because not all patients bleed from varices and source of bleeding sometimes cannot be ascertained by endoscopy [6,13]. Previous predictive models for all sources of UGI bleeding in cirrhotic patients required endoscopic or radiographic criteria and allowed subjective discretion [1,13]. A risk stratification model consisting of simple, objective, and routinely available pre-endoscopic para- meters has not been studied.

We aimed to investigate pre-endoscopic clinical para- meters that were readily available in the first few hours of hospital stay to identify risk factors of 6-week mortality in cirrhotic patients with acute UGI bleeding. We set to develop a prognostic model consisting of these simple covariates and to compare the Predictive performance of our model with that of the MELD score.

Methods

Study design and setting

This was a retrospective single-center study of consecu- tive cirrhotic patients presenting with acute UGI bleeding to the ED. We conducted this study by reviewing medical records and computerized database. The institutional review board of the hospital (National Taiwan University Hospital) approved this study.

This study was conducted in a Tertiary medical center with 2000 beds serving a metropolitan area in northern Taiwan (National Taiwan University Hospital, Taipei, Taiwan). In this hospital, ED physicians collaborated with consultant gastroenterologists in managing cirrhotic patients with acute UGI bleeding. Generally, cirrhotic patients with AVH were managed with intravenous

vasoactive pharmacotherapy (either terlipressin or soma- tostatin) and endoscopic band ligation or injection of tissue adhesives (n-butyl-2-cyanoacrylate). In addition, we routi- nely administered intravenous proton pump inhibitor (omeprazole) to those who bled from Peptic ulcers and performed endoscopic hemostasis for ulcers with high-risk stigmata (active bleeding, visible exposed vessel, or adherent blood clot).

Selection of participants

We reviewed the medical records of all ED patients with Initial diagnosis of acute UGI hemorrhage between July 2004 and July 2007. Eligible patients were those who fulfilled the following inclusion criteria: (1) older than 20 years, (2) diagnosis of liver cirrhosis before the index admission, and (3) confirmed diagnosis of acute UGI bleeding. We verified the diagnosis of acute UGI hemorrhage if the presentation was hematemesis or bloody aspirates from a nasogastric tube. In patients presenting with melena, hematochezia, or compatible symptoms with anemia (eg, syncope, coma, delirium, and dyspnea), information on UGI endoscopy was required to confirm the source of hemorrhage. Patients were excluded from analysis if UGI tract could not be ascertained as the level of bleeding.

Methods of measurement and outcome measures

Three investigators (C.S.C., C.H.T., and T.L.L.) manu- ally abstracted the following pertinent demographic and laboratory data. Clinical variables included age, sex, blood type, presentation of bleeding, etiology of cirrhosis, presence of HCC, severe comorbid conditions (congestive heart failure, end-stage renal disease, respiratory failure, debilitating stroke, Malignant diseases other than HCC, and organ transplantation), admission at off-hours (defined as 6 PM-7 AM of the next day during weekdays and all day on weekend and national holidays), blood pressure, heart rate, peripheral oxygen saturation (determined by the Pulse oximeter), and body temperature. laboratory variables obtained at the first hour of ED stay were also abstracted. Reliability of the collected data was determined by random re-abstraction of one third of the eligible patients by another investigator (Y.C.H.), who was blinded to the data collected by other investigators.

Data analysis

All statistical analysis was performed using commercial software (Stata, version 8.2; Stata Corp, College Station, TX). Continuous variables were expressed with median and interquartile range (IQR) because most variables were not normally distributed. Mann-Whitney U test was used to compare sets of continuous variables between the 6-

Fig. 1 Flowchart of patient identification and enrollment.

week survivors and the mortality cases. ?2 Test was used for univariate analysis of proportions, and Fisher exact test was applied in the case of expectant value below 10. For the purpose of this study, each admission was considered as 1 patient in the analysis.

We used stepwise multivariate logistic regression analysis to identify the risk factors independently associated with 6-week mortality. Likelihood ratio test was used to determine whether an added covariate improved the nested model. Adjusted odds ratio (aOR) was calculated for the respective risk factor. We built a predictive model consisting of these simple clinical parameters. Hosmer- Lemeshow test was used to assess the goodness of fit of our model. The c statistic of the developed model for 6-week mortality was determined by calculating the area under the receiver operating characteristic curve and was compared

with that of the MELD score. We estimated the sensitivity, specificity, positive predictive value, and negative predic- tive value of our model when probability of .5 (ie, the score of our model was 0) was chosen as the cutoff value to predict outcomes.

Results

Demographic and laboratory characteristics of cirrhotic patients with acute UGI bleeding

In total, 542 consecutive ED admissions of 389 cirrhotic patients were identified and enrolled into the analysis (Fig. 1). Most patients were men (77.86%) in their forties to sixties (median age, 54 years; IQR, 48-65 years), with chronic viral hepatitis as the most common etiology of cirrhosis (48.89% with viral hepatitis B and 22.88% with Hepatitis C). The median MELD score of this cohort was 11.58 (IQR, 8.54- 14.95). Demographic, clinical, and laboratory variables are summarized in Table 1.

Table 1 Clinical and laboratory characteristics in the 542 episodes of acute UGI bleeding in cirrhotic patients

Patient characteristics

Age (y) 54 (48-65) Male sex 422 (77.86) Hematemesis 306 (56.46) ED visit at off-hour 102 (18.82) Systolic blood pressure (mm Hg) 119 (103-135) Hypotension a 41 (7.56) Heart rate (beats/min) 99 (84-114) Hypoxemia b 31 (5.72) Etiology Hepatitis B 265 (48.89) Hepatitis C 124 (22.88) Alcoholic 96 (17.71) HCC 230 (42.44) Non-HCC malignancy 29 (5.35) Hemoglobin (g/dL) 8.9 (7.6-10.4) Severe anemia c 36 (6.64) Leukocyte (per uL) 7030 (4830-9910) Platelet (k/per mm3) 96.5 (64-144) INR 1.3 (1.2-1.47) Bilirubin (mg/dL) 1.635 (1.02-3.04) Creatinine (mg/dL) 1 (0.8-1.2) MELD score 11.58 (8.54-14.95) Hospital stay (d) 6 (3-12) 6-wk mortality 44 (8.12)

Data are presented as no. (%) or median (IQR), where appropriate. a Hypotension is defined as systolic blood pressure less than 90 mm Hg. b Hypoxemia is defined as peripheral oxygen saturation less than 95%. c Severe anemia is defined as hemoglobin level less than 6 g/dL.

Variables	6-wk mortality case (n = 44)	6-wk	survivor (n = 498)	P
Age (y)	52 (44.5-63.5)	54	(49-65)	.18
Male sex	41 (93.18)	381	(76.51)	.008
Hematemesis	22 (50)	284	(57.03)	.367
ED visit at off-hour	12 (27.27)	90	(18.07)	.16
Systolic blood pressure (mm Hg)	113 (100.5-126)	119	(104-135)	.16
Hypotension a	2 (4.55)	39	(7.83)	.56
Heart rate (beats/min)	103 (86.5-115)	99	(83-113)	.32
Hypoxemia b	14 (31.82)	17	(3.41)	b.001
Etiology
Hepatitis B	22 (50)	243	(48.8)	.88
Hepatitis C	8 (18.18)	116	(23.29)	.57
Alcoholic	9 (20.45)	87	(17.47)	.68
HCC	25 (56.82)	205	(41.16)	.044
Non-HCC malignancy	6 (13.64)	23	(4.62)	.023
Hemoglobin (g/dL)	8.7 (7.65-10.8)	8.9	(7.6-10.3)	.73
Severe anemia c	2 (4.55)	34	(6.83)	.76
Leukocyte (per uL)	8430 (6055-11530)	6870	(4710-9590)	.01
Platelet (k/per mm3)	93 (56.5-161.5)	96.5	(64-141)	.95
INR	1.5 (1.28-1.925)	1.3	(1.19-1.44)	b.001
Bilirubin (mg/dL)	3.38 (1.93-9.28)	1.53	(1.01-2.74)	b.001
Creatinine (mg/dL)	1.1 (0.85-1.6)	1	(0.8-1.2)	.07
MELD score	17.28 (10.75-24.77)	11.35	(8.50-14.65)	b.001
Data are presented as no. (%) or median (IQR), where appropriate. a Hypotension is defined as systolic blood pressure less than 90 mm Hg. b Hypoxemia is defined as peripheral oxygen saturation less than 95%. c Severe anemia is defined as hemoglobin level less than 6 g/dL.

Univariate association between 6-week mortality and clinical parameters

Table 2 Clinical features of 542 episodes of acute UGI bleeding in cirrhotic patients, partitioned by 6-week mortality

Forty-four patients died (8.12%) within 6 weeks. Different clinical parameters between those who died and those who survived are given in Table 2. The mortality cases were characterized by higher prevalence of male sex (93.18% vs 76.51%; P = .008), peripheral oxygen saturation less than 95% (31.82% vs 3.41%; P b.001), HCC (56.82% vs 41.16%;

P = .044), and non-HCC malignancy (13.64% vs 4.62%; P =

.023). The medians of the peripheral leukocyte count (8430 vs 6870; P = .01), serum total bilirubin (3.38 vs 1.53; P b .001),

Table 3 Statistical parameters of the multivariate logistic regression model

and International normalized ratio (1.5 vs 1.3; P b.001) were all significantly higher in those who died. A trend of higher serum creatinine was observed in the mortality cases, nonetheless without statistic significance (P = .07).

Independent risk factors and multivariate regression models predictive of mortality

Multivariate logistic regression analysis revealed the independent association of 6-week mortality with male sex (P = .032), HCC (P = .024), non-HCC malignancy (P =

.006), hypoxemia at presentation (P b .001), serum bilirubin

Coefficient (?)		Standard error of	?	95% CI of ?	P
Constant	-6.40	0.98	-8.32 to -4.49		b.001
Male sex	1.47	0.68	0.13 to 2.81		.032
Hypoxemia a	2.24	0.48	1.29 to 3.19		b.001
HCC	0.84	0.37	0.11 to 1.56		.024
Non-HCC malignancy	1.55	0.57	0.44 to 2.66		.006
Bilirubin (mg/dL)	0.07	0.03	0.02 to 0.12		.012
INR	1.06	0.42	0.24 to 1.87		.011
The regression model was -6.40 + 1.47 x male sex + 2.24 x hypoxemia + 0.84 x HCC + 1.55 x another cancer + 0.07 x serum bilirubin + 1.06 x INR. a Hypoxemia is defined as peripheral oxygen saturation less than 95%.

Group	Total	Death			Survival
		Observed	Expected		Observed	Expected

1	54	1	0.36	53	53.64
2	54	0	0.84	54	53.16
3	54	0	1.42	54	52.58
4	54	1	1.64	53	52.36
5	54	3	2.01	51	51.99
6	54	2	2.81	52	51.19
7	54	3	3.38	51	50.62
8	54	4	4.12	50	49.88
9	54	11	6.33	43	47.67
10	56	19	21.08	37	34.92
* P = .366.

(P = .012), and INR for prothrombin time (P = .011). A predictive model composing of these 6 covariates was built (Table 3):

Table 4 Partition of study samples for the Hosmer and Lemeshow test ?

Our model = Ln(Probability of death/1

-probability of death) = — 6.40 + 1.47

xmale sex + 2.24 x hypoxemia + 0.84

xHCC + 1.55 x non-HCC malignancy

+ 0.07 x serum bilirubin (mg/dL) + 1.06

xINR; Ln denotes natural logarithm

The goodness of fit of using 6 variables in our model was verified using the Hosmer-Lemeshow test (Table 4) [14]. The receiver operating characteristic curve of our predictive model (model of National Taiwan University Hospital, NTUH model) to predict mortality was drawn and was compared with that of the MELD score (Fig. 2). The c statistic of our model was 0.84 (95% confidence interval [CI], 0.78-0.90), and that of the MELD score was 0.71

Fig. 2 Receiver operating characteristic curve of our model versus MELD score to predict in-hospital mortality.

Note. NTUH, National Taiwan University Hospital; MELD, Model for End-stage Liver Disease; P = .002 between two models.

(95% CI, 0.61-0.80). Our model was significantly superior to the MELD score in predicting 6-week mortality (P =

.002). Adjusted odds ratio was calculated for each risk factor (Table 5). If cutoff value for the probability of death was set at .5 to classify patients, the sensitivity, specificity, positive predictive value, and negative predictive value of our model were 22.73%, 99.80%, 90.91%, and 93.60% respectively, and this model correctly classified 93.54% of the study subjects (Table 6).

Discussion

This study demonstrates that risk stratification of cirrhotic patients with acute UGI hemorrhage may be achieved by a set of 6 simple clinical parameters readily obtained in the first hour at the ED stay. Six-week mortality is independently associated with male sex, hypoxemia, HCC, non-HCC malignancy, serum bilirubin, and prothrombin time. A predictive model consisting of these covariates is superior to the MELD score in predicting 6-week mortality, with c statistic of 0.84 and 0.71, respectively (P = .002).

Our study confirmed that the severity of the underlying liver cirrhosis was a major determinant of clinical outcomes in cirrhotic patients with acute UGI hemorrhage. We demonstrated that 6-week mortality was independently

Table 5 Independent risk factors of 6-week mortality in cirrhotic patients with acute UGI hemorrhage, determined by multivariate logistic regression model

aOR		95% CI
Male sex	4.35	1.14-16.62
Hypoxemia a	9.42	3.65-24.30
HCC	2.31	1.12-4.78
Non-HCC malignancy	4.70	1.55-14.26
Bilirubin (per mg/dL)	1.07	1.02-1.13
INR (per unit)	2.88	1.28-6.51
a Hypoxemia is defined as peripheral oxygen saturation less than 95%.

Table 6 Predictive performance of our model with cutoff value for the probability of 6-week mortality set at 0.5 Sensitivity Specificity PPV NPV Correctly classified
Our model	22.73%	99.80%	90.91%	93.60%	93.54%
MELD score	4.55%	99.40%	40.00%	92.18%	91.70%
PPV indicates positive predictive value; NPV, negative predictive value.

associated with prothrombin time (aOR, 2.88 for each unit of INR; 95% CI, 1.28-6.51) and serum bilirubin (aOR, 1.07 for each milligram per deciliter; 95% CI, 1.02-1.13), both of which were indicators of Hepatic function. Our study kept in line with most of previous studies that reported an independent association between mortality and severity of Hepatic dysfunction in cirrhotic patients with acute UGI bleeding or AVH [1,7,9-13]. We also revealed that the comorbidity of malignant disease was another important poor prognostic factor. In our study, previous diagnosis of HCC was associated with an aOR of 2.31 (95% CI, 1.12- 4.78), and coexistence of another cancer other than HCC was independently associated with an additional aOR of

4.70 (95% CI, 1.55-14.26). Our finding was consistent with that of D’Amico and De Franchis [1] and Gatta et al [13], who reported HCC as an independent risk factor for short- term mortality. Nevertheless, the association between 6- week mortality and a non-HCC malignant disease has not been reported before. Furthermore, we confirmed that risk of death was higher in the male patients, a finding that has been reported previously [13]. Although male sex has been found to indicate poorer prognosis in cirrhotic patients, independent of stage of the underlying liver disease, age, and comorbidity of HCC [15-17], the exact mechanism remains undetermined. Finally, we demonstrated that peripheral hypoxemia (defined as oxygen saturation less than 95% by pulse oximeter) was a strong predictor of short-term mortality.

Previous prognostic models of acute UGI hemorrhage in cirrhotic patients comprised not only objective clinical variables but also endoscopic findings, radiographic criteria, and subjective discretion. Gatta el al [13] developed a prognostic index by stepwise logistic regression analysis, which required endoscopy to confirm varices as the definite or probable bleeding source. Their model also necessitated determination of ascites. Other covariates of the model of Gatta et al were serum creatinine, serum bilirubin, pro- thrombin index, diagnosis of HCC, male sex, and presenta- tion of hematemesis. In the prognostic models by D’Amico and De Franchis [1], poor outcomes were predicted by active bleeding on endoscopy, initial hematocrit, serum aspartate aminotransferase, portal vein thrombosis, CTP class, alco- holic etiology, HCC, serum bilirubin, serum albumin, Hepatic encephalopathy, and Blood transfusion requirements. Although UGI endoscopy is indicated in bleeding cirrhotic patients, endoscopic data cannot be available to emergency physicians until several hours later. Moreover, endoscopy is frequently delayed in those patients presenting with unstable

hemodynamics. As a result, endoscopic criteria may not be convenient to emergency physicians. We also consider subjective judgment unreliable, particularly relevant in the busy setting of ED. Diagnosis of ascites and hepatic encephalopathy depend heavily on the skills of examiners and cannot be free of interobserver variation [18,19]. Furthermore, the criterion of portal vein thrombosis cannot be routinely applied because computed tomography is not always performed and sonography is operator dependent.

Prognostic value by simple clinical variables has been reported in a study by Abraldes and colleagues [20], who demonstrated that a model using CTP classification, systolic blood pressure, and etiology of cirrhosis was as accurate as the invasive measurement of hepatic venous pressure gradient in the prediction of poor outcomes. Their study corroborated that simple clinical parameters alone may successfully predict the short-term prognosis in cirrhotic patients with AVH. Nevertheless, their studies were limited to those with Variceal hemorrhage and thus could not be generalized to all cirrhotic patients with acute UGI bleeding. Moreover, their model allowed subjective variation by using CTP classification.

Short-term mortality rate in cirrhotic patients with acute UGI hemorrhage has improved significantly over the past 3 decades, from 42% in the 1970s [21] to 20% in the late 1990s

[1] to 15% in the early 2000s [2]. In a recent Korean study [22], the short-term mortality of gastric variceal hemorrhage was 11.6%. The 6-week mortality rate of 8.12% noted in our study may reflect a continuing advancement. Probable reasons for the improvement include administration of vasoactive agents, endoscopic therapy, and antibiotic pro- phylaxis [1,2]. In fact, with widespread application of effective therapeutic approaches into routine clinical practice, the impact of the treatment discrepancy on clinical outcomes may be limited. The success of using only pre-endoscopic clinical variables to stratify patients at different risks of death in our and others’ studies [20,22] implied that therapeutic variables might not decisively affect clinical outcomes.

Our model correctly classified 93.54% of study subjects by applying probability of death set at .5 as the cutoff value to classify patients (Table 6), but trade-off between sensitivity and specificity could not be avoided. Because 6-week mortality rate was relatively uncommon in our study (8.12%), a cutoff value with higher sensitivity (at the expense of specificity) would considerably reduce positive predictive value while only marginally improving negative predictive value. Therefore, the specificity of this model was as high as 99.8% and sensitivity was as low as 22.73% to maximize the

proportion of correct classification. In fact, the selection of cutoff point for a predictive model depends on what the users care for the most. One may choose to augment sensitivity and thereby reduce false-negative prediction.

Merits of this study include a large sample size. We analyzed a total of 542 episodes of acute UGI hemorrhage in cirrhotic patients and, therefore, were empowered to identify subtle differences that might be overlooked in a smaller sample. For example, although the prevalence of hypoxemia and that of non-HCC malignancy were only 5.72% and 5.35%, respectively, we were able to demonstrate their independent association with 6-week mortality. Second, our model consists of only simple, objective, and easily obtainable clinical variables that can be conveniently introduced into everyday practice. Third, the relatively short period (3 years) of the study reflects current clinical practice and minimizes practice variation over time.

However, several limitations of this study should be noted. First, the retrospective design may raise concerns of selection bias and uncovered confounding factors. To minimize the influence of this limitation, we enrolled consecutive patients fulfilling clearly defined inclusion criteria. More importantly, none of these 6 identified predictive risk factors was missed in any study subject. Second, this investigation was conducted in single tertiary referral center in a metropolitan area. As a result, our study subjects might not represent those treated at a general or local hospital in terms of patient characteristics and management strategies. Therefore, whether our result may be generalized to other dissimilar settings warrants further studies for validation. Third, because we focused on early predictive factors, this study did not include therapeutic variables into analysis. However, we felt confident that it did not influence our results. The management principles specified in the “Methods” section of this article were applied to all enrolled patients. Furthermore, the success of our predictive model using only simple clinical covariates lent strong support that the management issue might not be a decisive factor to affect our result.

In conclusion, the 6-week mortality of acute UGI bleeding in patients with liver cirrhosis is independently associated with 6 simple clinical parameters, including male sex, HCC, non-HCC malignancy, hypoxemia, serum bilirubin, and prothrombin time. A prognostic model built on these 6 covariates is superior to the MELD score in predicting short- term mortality. As these clinical variables are readily available in the ED, our model may be valuable for early risk stratification of cirrhotic patients with acute UGI hemorrhage. Our results warrant more research for prospec- tive validation.

Acknowledgments

We are indebted to Dr FC Hu and his team of biostatistics for their assistance of statistical analysis.

References

1. D’Amico G, De Franchis R. Upper digestive bleeding in cirrhosis. Post-therapeutic outcome and prognostic indicators. Hepatology 2003; 38:599-612.
2. Carbonell N, Pauwels A, Serfaty L, et al. Improved survival after variceal bleeding in patients with cirrhosis over the past two decades. Hepatology 2004;40:652-9.
3. Rockall TA, Logan RF, Devlin HB, et al. Risk assessment after acute upper gastrointestinal haemorrhage. Gut 1996;38:316-21.
4. Blatchford O, Murray WR, Blatchford M. A risk score to predict need for treatment for upper-gastrointestinal haemorrhage. Lancet 2000; 356:1318-21.
5. Cameron EA, Pratap JN, Sims TJ, et al. Three-year prospective validation of a pre-endoscopic risk stratification in patients with acute upper-gastrointestinal haemorrhage. Eur J Gastroenterol Hepatol 2002; 14:497-501.
6. Teres J, Bordas JM, Bru C, et al. Upper gastrointestinal bleeding in cirrhosis: clinical and endoscopic correlations. Gut 1976;17:37-40.
7. Ben-Ari Z, Cardin F, McCormick AP, et al. A predictive model for failure to control bleeding during acute variceal haemorrhage. J Hepatol 1999;31:443-50.
8. Moitinho E, Escorsell A, Bandi JC, et al. Prognostic value of early measurements of portal pressure in acute variceal bleeding. Gastro- enterology 1999;117:626-31.
9. Bambha K, Kim WR, Pedersen R, et al. Predictors of early re-bleeding and mortality after acute variceal haemorrhage in patients with cirrhosis. Gut 2008;57:814-20.
10. Chalasani N, Kahi C, Francois F, et al. Model for end-stage liver disease (MELD) for predicting mortality in patients with acute variceal bleeding. Hepatology 2002;35:1282-4.
11. Amitrano L, Guardascione MA, Bennato R, et al. MELD score and hepatocellular carcinoma identify patients at different risk of short- term mortality among cirrhotics bleeding from esophageal varices. J Hepatol 2005;42:820-5.
12. Primignani M, Dell’Era A, Bucciarelli P, et al. High-D-dimer plasma levels predict poor outcome in esophageal variceal bleeding. Dig Liver Dis 2008;40:874-81.
13. Gatta A, Merkel C, Amodio P, et al. Development and validation of a prognostic index predicting death after upper gastrointestinal bleeding in patients with liver cirrhosis: a multicenter study. Am J Gastroenterol 1994;89:1528-36.
14. Hosmer DW, Hosmer T, Le Cessie S, et al. A comparison of goodness-of-Fit tests for the logistic regression model. Stat Med 1997; 16:965-80.
15. Schlichting P, Christensen E, Andersen PK, et al. Prognostic factors in cirrhosis identified by Cox’s regression model. Hepatology 1983; 3:889-95.
16. D’Amico G, Morabito A, Pagliaro L, et al. Survival and prognostic indicators in compensated and decompensated cirrhosis. Dig Dis Sci 1986;31:468-75.
17. Gines P, Quintero E, Arroyo V, et al. Compensated cirrhosis: natural history and prognostic factors. Hepatology 1987;7:122-8.
18. Conn HO. A peek at the Child-Turcotte classification. Hepatology 1981;1:673-6.
19. Malinchoc M, Kamath PS, Gordon FD, et al. A model to predict poor survival in patients undergoing transjugular intrahepatic portosystemic shunts. Hepatology 2000;31:864-71.
20. Abraldes JG, Villanueva C, Banares R, et al. Hepatic venous pressure gradient and prognosis in patients with acute variceal bleeding treated with pharmacologic and endoscopic therapy. J Hepatol 2008;48:229-36.
21. Graham DY, Smith JL. The course of patients after variceal hemorrhage. Gastroenterology 1981;80:800-9.
22. Paik CN, Kim SW, Lee IS, et al. The Therapeutic effect of cyanoacrylate on Gastric variceal bleeding and factors related to clinical outcome. J Clin Gastroenterol 2008;42:916-22.