ORIGINAL ARTICLE
A Risk Scoring System to Predict In-hospital Mortality in Patients With Cirrhosis Presenting With Upper Gastrointestinal Bleeding Thomas Lyles, MD,* Alan Elliott, MAS, MBA,w and Don C. Rockey, MDz Goals: We aimed to develop a simple and practical risk scoring system to predict in-hospital mortality in cirrhotics presenting with upper gastrointestinal (GI) bleeding. Study: Extensive clinical data were captured in patients with documented cirrhosis who underwent endoscopic evaluation for upper GI bleeding between January 1, 2003 and June 30, 2011 at Parkland Memorial Hospital. Predictors of mortality were identified by multivariate regression analysis. Results: A total of 884 patients with cirrhosis admitted for upper GI bleeding were identified; 809 patients survived and 75 died (8.4%). The etiology of bleeding was similar in both groups, with bleeding attributed to esophageal varices in 59% of survivors and 60% of non-survivors (ulcer disease and other etiologies of bleeding accounted for the other causes of bleeding). Mortality was 8.6% and 8.3% in patients with variceal bleeding and nonvariceal bleeding, respectively. While survivors and those who died were similarly matched with regard to gender, age, ethnicity and etiology of cirrhosis, patients who died had lower systolic blood pressures, higher pulse rates and lower mean arterial pressures at admission than patients who survived. Non-survivors were more likely to be Childs C (61% vs. 19%, P < 0.001). Multivariate regression analysis identified the following 4 predictors of in-hospital mortality: use of vasoactive pressors, number of packed red blood cells transfused, model for end-stage liver disease (MELD) score, and serum albumin. A receiver operating characteristic curve including these 4 variables yielded an area under the receiver operating characteristic (AUROC) curve of 0.94 (95% confidence interval, 0.91-0.98). Classification and Regression Tree analysis yielded similar results, identifying vasoactive pressors and then MELD > 21 as the most important decision nodes for predicting death. By comparison, using the Rockall scoring system in the same patients, the AUROC curve was 0.70 (95% confidence interval, 0.64-0.76 and the comparison of the University of Texas Southwestern model to the Rockall model revealed P < 0.0001). A validation set comprised of 150 unique admissions between July 1, 2011 and July 31, 2012, had
Received for publication April 13, 2013; accepted September 16, 2013. From the *Division of Digestive and Liver Diseases, UT-Southwestern Medical Center and the Parkland Health and Hospital System; wDepartment of Statistical Science, Southern Methodist University, Dallas, TX; and zDepartment of Internal Medicine, Medical University of South Carolina, Charleston, SC. T.L., A.E., and D.C.R. had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. T.L., D.C.R.: study concept and design, drafting of the manuscript, critical revision of the manuscript for important intellectual content, administrative, technical, or material support, study supervision; T.L., A.E., D.C.R.: analysis and interpretation of data, statistical analysis; T.L.: acquisition of data. The authors declare that they have nothing to disclose. Reprints: Don C. Rockey, MD, Department of Internal Medicine, Medical University of South Carolina, 96 Jonathan Lucas Street, Suite 803, MSC 623, Charleston, SC 29425 (e-mail: rockey@ musc.edu). Supplemental Digital Content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Website, www.jcge.com. Copyright r 2013 by Lippincott Williams & Wilkins
712 | www.jcge.com
an AUROC of 0.92, and the outcomes of 97% of the subjects in this set were accurately predicted by the risk score model. Conclusions: Use of vasoactive agents, packed red blood cell transfusion, albumin, and MELD score were highly predictive of in-hospital mortality in cirrhotics presenting with upper GI bleeding. These variables were used to formulate a clinical risk scoring system for in-hospital mortality, which is available at: http://medweb.musc.edu/LogisticModelPredictor. Key Words: varices, ulcer, outcome, prediction, endoscopy, model
(J Clin Gastroenterol 2014;48:712–720)
U
pper gastrointestinal (GI) bleeding is an important and common clinical problem in the United States, accounting for approximately 300,000 hospitalizations annually.1–3 Outcomes for cirrhotic patients with upper GI bleeding are worse than for those who do not have cirrhosis.3 Estimation of prognosis is an essential element of managing patients with cirrhosis admitted to a hospital. Longitudinal studies have previously found that stage of cirrhosis, defined by consensus as stage I to IV based on presence of clinical complications and global liver function [Child-Pugh class, model for end-stage liver disease (MELD) score] are important prognosticators of a patient’s overall mortality risk.4–8 By definition, patients with cirrhosis and GI bleeding are classified as stage IV; it has been reported that patients within this category have a 1-year mortality risk of 57%.9 Moreover, half of these patients will die within the 6 weeks after their admission for GI bleeding.3,5,6,10,11 Although several risk-stratification tools have been formulated for each episode of acute GI bleeding and for estimation of 6-week mortality after an episode of variceal bleeding in cirrhosis,3–5,12,13 a stratification model to estimate in-hospital mortality in all cirrhotics admitted with acute upper GI bleeding is not currently available. This seems to be in part due to the perceived distinction between variceal bleeding and nonvariceal bleeding. Although variceal bleeding has historically carried the higher mortality rate, this may not be the case.4,5,12 Prediction of outcome in patients with acute upper GI bleeding seems to be feasible.14–17 Thus, we have taken advantage of a large and robust cohort of patients with cirrhosis and upper GI bleeding to focus on in-hospital outcomes. Our goal was to develop a simple, and practical clinical tool specifically to help estimate in-hospital mortality risk in patients with cirrhosis and all forms of acute upper GI bleeding, whether from variceal or nonvariceal upper GI bleeding.
MATERIALS AND METHODS This cohort study focused on predictors of in-hospital mortality in cirrhotics who were admitted to the hospital
J Clin Gastroenterol
Volume 48, Number 8, September 2014
J Clin Gastroenterol
Volume 48, Number 8, September 2014
with a diagnosis of upper GI bleeding. The study included consecutive patients admitted to Parkland Memorial Hospital (Dallas, TX), a University of Texas Southwestern (UTSW) teaching hospital, from January 1, 2003 through June 30, 2011. Inclusion required a definitive diagnosis of cirrhosis and documented acute upper GI bleeding. Patient data were captured through the unit’s GI Bleeding Healthcare Registry, which prospectively collects data on patients admitted to Parkland Memorial Hospital with any form of gastrointestinal bleeding and through the unit’s Chronic Liver Disease Registry, which collects data on consecutive cirrhotic patients admitted to our institution. Data captured in the GI Bleeding Healthcare Registry includes multiple clinical and historical features, including the American Society of Anesthesiologists score on physical status, medications, laboratory and endoscopic data (time to endoscopy), and clinical features associated with GI bleeding. Primary hemostasis rates, treatment failures, and 30-day rebleeding events are also collected. Outcomes collected include rebleeding, overall length of stay, and mortality. Rebleeding is defined as visualization of vomited red blood, a drop in hematocrit of Z9 points (or hemoglobin 3 g/dL) after endoscopy or by development of hypotension (SBPr90) more than 2 hours after endoscopy. By design, a bleeding lesion or a lesion with stigmata of recent bleeding in any given case is designated as the primary diagnosis. When >1 lesion/diagnosis is present in addition to a primary lesion, it is considered a secondary lesion, but not deemed to be the cause of hemorrhage. Primary bleeding lesions are assigned by a 3-panel group to one of the 18 specific primary diagnoses (or other/no source identified). Causes of death for all patients are classified into 8 different groups, also adjudicated by the study group, which include the following: gastrointestinal bleeding, cardiorespiratory failure, renal failure, liver failure, sepsis, multiorgan system dysfunction, malignancy, unknown, or others. The Chronic Liver Disease Registry similarly collects extensive clinical data, including the same and additional features highlighted above, such as patient demographics, medical history and clinical data, hospital course data [intensive care unit (ICU) stay, intubation, use of vasoactive agents such as dopamine, dobutamine, norepinephrine, etc.], medications (including b blockers at admission and discharge, antibiotics), laboratory data, physical examination findings relevant to cirrhosis (ascites, portosystemic encephalopathy, etc.), endoscopic features, and outcomes. To compare co-morbidities between survivors and nonsurvivors, the Charlson Comorbidity Index18,19 was calculated on each patient in the test set. Laboratory data were captured at the time of admission. MELD and ChildPugh score were calculated upon admission with current laboratory markers and physical examination findings. For varices, additional data were collected such as the number and grade of varices, use of band ligation as hemostasis, and number of bands placed. Ascites were divided into 2 categories: patients without ascites or ascites only detectable by imaging were grouped together and those with clinically evident ascites were grouped together.20 Patients with diuretic controlled ascites were coded in the former group. Portosystemic encephalopathy was determined upon admission by the admitting physician or hepatology consultant and reported in accordance with the West Haven criteria.21 Cirrhosis was defined as follows: a history consistent with chronic liver disease, plus the presence of clinical r
2013 Lippincott Williams & Wilkins
A RS System to Predict In-hospital Mortality
features consistent with cirrhosis, including clinical findings of cirrhosis and portal hypertension (spider angiomata, gynecomastia, splenomegaly, thrombocytopenia) or a documented complication of chronic liver disease (ie, ascites, varices, hepatic encephalopathy), and/or imaging consistent with cirrhosis, and/or liver histology consistent with cirrhosis. Upper GI bleeding was defined as reported or witnessed melena, hematemesis, coffee ground emesis, or hematochezia (with a documented upper GI tract lesion) in the setting of at least a 4-point drop in hematocrit from baseline or lower than normal. It is a standard practice at our institution to place patients on continuous protonpump inhibitor infusion, octreotide infusion, and antibiotics upon admission and to perform esophagogastroduodenoscopy in patients with any form of upper GI bleeding as defined above unless contraindications to endoscopy exist. In the event of esophageal variceal bleeding, variceal ligation was preferentially used during the study period. Sclerotherapy and transjugular intrahepatic portosystemic shunt were used if bleeding could not be controlled with variceal ligation. Treatment of nonvariceal bleeding was left to the discretion of the endoscopist and included the use of epinephrine injection therapy, endoclip hemostasis, bipolar coagulation therapy, and combinations of the aforementioned modalities. Notably, our institution’s management of upper GI bleeding (including pharmacologic and endoscopic therapy) did not significantly change over the course of the study period, although it should be emphasized that efforts to improve standardization have been ongoing. With regard to packed red blood cell transfusion, the practice in our institution is that the hemoglobin is typically targeted to be 8 to 9 g/dL. However, in our experience, the cutoff for ideal transfusion varies among practitioners, and depends on individual patients and co-morbidities. Because of this variability, our institution currently does not attempt to impose specific transfusion practices. Although the reference Hgb level for transfusion is recommended to be 8 to 9 g/dL, the use of packed red blood cell transfusion and number of units transfused was left to the discretion of the admitting physician in conjunction with the gastroenterology consult service. Patients were excluded from this analysis if they had a form of GI bleeding other than an upper GI source, were under 18 years old, did not have endoscopy during their hospitalization, or were pregnant. Patients were not excluded if they had hepatocellular carcinoma, previous transjugular intrahepatic portosystemic shunt/surgical shunt, or bleeding from gastrointestinal malignancy. For patients who had multiple hospitalizations for bleeding during the study period, rebleeding after 42 days from the index bleed was considered a new bleed. A readmission within 42 days from the index bleed was considered a rebleed and this readmission data were excluded from the risk score analysis. The study was approved by the University of Texas Southwestern Institutional Review Board and met all criteria for good clinical practice.22 A second validation cohort was compiled from cirrhotic patients admitted to our institution between July 1, 2011 and July 31, 2012. The aforementioned definitions were applied to this validation cohort to ensure uniformity between the 2 groups. After formulating the in-hospital mortality risk score, this second set was used to validate the model. www.jcge.com |
713
Lyles et al
Descriptive statistics were used to present demographic and clinical characteristics of patients in the study cohort. The primary outcome of this study was in-hospital mortality. We specifically aimed to identify prognostic variables of in-hospital mortality in cirrhotics with upper GI bleeding, and to incorporate those variables into a clinically relevant risk-stratification score. Patients were divided into 2 groups, those surviving hospitalization and those who did not survive. Group characteristics and demographics were compared between groups using Student t test for continuous variables or w2 for categorical variables where appropriate. To examine variables for their predictive relationship to mortality, univariate logistic regression was utilized to examine possible predictors. Variables that were at least marginally predictive (P < 0.2) in the univariate case were considered for a multivariate logistic model. A final logistic regression model included only those variables whose multivariate combination was most predictive for the binary outcome variable mortality. Model fit was assessed using the area under the receiver operating characteristic (AUROC) curve and a HosmerLemeshow goodness-of-fit analysis. A second analysis was performed using the Classification and Regression Tree (CART) technique. This method uses recursive partitioning to select variables most predictive of mortality, and creates a tree-structured model that can be analyzed to assess probability of survival. All P-values in the analyses presented are 2 sided and considered significant when Pr0.05 unless otherwise stated. No adjustments for multiple testing were made. The data were analyzed using the SAS V 9.2 (SAS Inc., Cary, NC) statistical package and CART software (version 6.0; Salford Systems, San Diego, CA).
J Clin Gastroenterol
Volume 48, Number 8, September 2014
TABLE 1. Demographic and Clinical Characteristics
Mean ± SD/n (%)
Age* (y) Sex, female (%) Ethnicity Hispanic White AfricanAmerican Others Cirrhosis etiology HCV Alcohol Cryptogenic NAFLD HCV + HBV HBV Others Systolic blood pressure Mean arterial pressure Pulse Presence of ascites Presence of PSE Charlson score CTP class A B C
Survivors (n = 809)
Nonsurvivors (n = 75)
51 (± 9) 217 (27)
51 (± 10) 25 (33)
413 (51) 256 (32) 114 (14)
33 (44) 21 (28) 17 (23)
26 (3)
4 (5)
P 0.514 0.226 0.152
0.21 362 315 32 33 22 22 23 122
(45) (39) (4) (4) (3) (3) (2) (± 23)
88 (± 17) 96 464 95 3
(± 20) (57) (12) (2)
265 (33) 387 (48) 157 (19)
42 24 5 0 2 1 1 112
(56) (32) (7) (0) (3) (1) (1) (± 23)
< 0.001
82 (± 17)
0.001
103 34 23 4
(± 17) (45) (31) (2)
0.003 0.045 < 0.001 0.426 < 0.001
8 (11) 21 (28) 46 (61)
*Data provided are those upon the patient’s admission to the hospital. CTP indicates Child-Turcotte-Pugh; HBV, hepatitis B virus; HCV, hepatitis C virus; NAFLD, nonalcoholic fatty liver disease; PSE, portosystemic encephalopathy.
RESULTS
FIGURE 1. Flow diagram of the patient cohort. A total of 2548 patients were admitted to our institution with acute GI bleeding in the study period. After excluding patients with lower GI bleeding, those who were not cirrhotic and who did not undergo esophagogastroduodenoscopy, the study cohort was composed of 884 patients. EGD indicates esophagogastroduodenoscopy; GI, gastrointestinal.
714 | www.jcge.com
We identified a total of 884 unique admissions of patients with cirrhosis and upper GI bleeding (Fig. 1). This cohort included 521 patients with variceal bleeding and 363 patients with nonvariceal bleeding. Seventy-five patients (8.4%) died during hospitalization. The cohort included patients with typical causes of cirrhosis, was predominantly male, was relatively young with a mean age of 50 years, and included a high proportion of patients of Hispanic ethnicity (Table 1). Although survivors and those who died were similarly matched with regard to sex, age, ethnicity, and etiology of cirrhosis, there were several key differences. Patients who died had lower systolic blood pressures, higher pulse rates, and lower mean arterial pressure than patients who survived. In addition, patients who died more frequently had ascites, and when present it was more advanced than survivors. Patients who died also had worse global liver dysfunction than survivors, as evidenced by a greater proportion with Child class C and higher MELD score (Tables 1, 2). Notably, patients who died were less likely to be on b blockers at the time of their admission (62% vs. 76%, P = 0.026). There were a number of notable differences in laboratory abnormalities among survivors and nonsurvivors (Table 2). Nonsurvivors had higher white blood cell counts, higher blood urea nitrogen, lower sodium, higher creatinine, higher international normalized ratio, higher bilirubin, lower albumin, and had significantly higher MELD r
2013 Lippincott Williams & Wilkins
J Clin Gastroenterol
Volume 48, Number 8, September 2014
TABLE 2. Laboratory Data
Mean ± SD Survivors (n = 809) WBC ( 109/L) Hematocrit (%) Platelets ( 109/L) BUN (mg/dL) Sodium (mmol/L) Creatinine (mg/dL) INR Total bilirubin (mg/dL) Albumin (g/dL) MELD MELD-Na Positive blood cultures [n (%)]
Nonsurvivors (n = 75)
P
8±5 28 ± 8 117 ± 70 25 ± 16 135 ± 5 1.0 ± 0.9 1.5 ± 0.7 2.6 ± 3.8
12 ± 6 26 ± 7 119 ± 67 33 ± 22 132 ± 6 1.9 ± 1.4 2.1 ± 0.8 8.7 ± 10.6
< 0.001 0.044 0.743 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001
2.9 ± 0.6 14 ± 6 17 ± 6 23 (3)
2.3 ± 0.6 25 ± 9 27 ± 8 19 (25)
< 0.001 < 0.001 < 0.001 < 0.001
BUN indicates blood urea nitrogen; INR, international normalized ratio; MELD, model for end-stage liver disease; Na, sodium; WBC, white blood cells.
scores. Patients who did not survive were more likely to have a positive blood culture (3% for survivors vs. 25% for nonsurvivors, P < 0.001). Hematocrit and platelet levels were similar in the 2 groups. The etiology of upper GI bleeding was similar in both groups, with bleeding in each group attributed to esophageal varices approximately 60% of the time (Table 3), consistent with previous reports.3,4,23 Esophageal varices were identified in 739 patients overall, and were the culprit lesion in 521 patients. Our institution’s practice is to preferentially treat bleeding esophageal varices with band ligation therapy; sclerotherapy and transjugular intrahepatic portosystemic shunt are performed if bleeding cannot be controlled with band ligation and/or culprit lesions are not amenable to banding (ie, significant esophageal scarring from prior banding sessions). Outcomes among the 2 groups were also significantly different (Table 4). Those who died had longer hospitalizations, were more likely to have active bleeding at time of endoscopy, require blood transfusion, and when transfused, received a greater number of packed red blood cells.
TABLE 3. Etiology of Upper Gastrointestinal Bleeding
N (%) Survivors (n = 809) Esophageal varices Gastro-duodenal ulcer Esophagitis* Othersw Portal hypertensive gastropathy Mallory-Weiss tear Banding ulcer No clear source
476 77 48 39 82
(59) (10) (6) (4) (10)
24 (3) 9 (1) 63 (8)
Nonsurvivors (n = 75) 45 8 8 10 3
(60) (11) (11) (13) (4)
1 (1) 0 (0) 0 (0)
P 0.90 0.54 0.13 0.006 0.10 0.72 1.0 1.0
*Esophagitis + GE junction ulcer + esophageal ulcer + banding ulcer. wGastric mass, hemobilia, oropharyngeal bleeding, coagulopathic oozing, GAVE, Dieulafoy, arteriovenous malformation, gastritis, gastric/duodenal erosions.
r
2013 Lippincott Williams & Wilkins
A RS System to Predict In-hospital Mortality
Patients who died were also more likely to be admitted to the ICU, require vasoactive agents and intubation and mechanical ventilation. Of the 53 patients requiring vasoactive agents, only 7 survived the hospitalization. One hundred twenty-six patients were intubated and of these, 69 survived their hospitalization. Overall, 49 patients required both vasoactive agents and intubation; of these, 5 lived. We found that the mortality rate in the 2 cohorts was identical, with an in-hospital mortality of 8.6% in patients with variceal bleeding and 8.3% in patients nonvariceal bleeding (Supplementary Table 1, Supplemental Digital Content 2, http://links.lww.com/JCG/A98). Because the 2 groups had similar in-hospital mortality rates, regression analysis and CART were focused on detecting significant variables in the overall population of cirrhotics with upper GI bleeding, with in-hospital mortality as the endpoint. Univariate and multivariate logistic regression analysis were performed. We included 57 different clinical variables in the univariate analysis. Those with a P-value 21 as the second decision node for predicting inhospital mortality (Fig. 3). We next sought to validate the primary logistic regression model using a prospectively collected cohort (July 1, 2011 and July 31, 2012). This set was composed of 150 unique admissions and had an in-hospital mortality rate of 6.7%. Clinical characteristics of this group were similar to patients in the primary group (Supplementary Table 3, Supplemental Digital Content 2, http://links. lww.com/JCG/A98). The outcomes of 97% of the subjects in the validation set were correctly identified. However, 3 predicted deaths out of 10 were incorrect and 2 predicted survivors out of 140 were incorrect, which indicates a sensitivity of 97.8% and specificity of 78% (positive predictive value of 98.7% and a negative predictive value of 70%). The AUROC for the validation set is 0.9164 (95% CI, 0.870.95; Fig. 2C). When plotted on respective histograms, the training data set and validation data set had good
716 | www.jcge.com
vasoactive agents, and endotracheal intubation, and the subsequent development of clinical complications including renal insufficiency, cardiopulmonary complications, and infection. Considerable effort has been directed at prediction of outcome in patients with acute upper GI bleeding14,15 and previously identified multiple potential risk factors for mortality such as increasing age, male sex, being uninsured, physical examination findings (systolic blood pressure 18, it was a smaller overall study of 102 patients8 and there were other notable differences including a retrospective design, use of terlipressin from 2006 to 2008 in their study, no hospital course data including ICU-related data, and an AUROC of 0.76. Their findings, however, correlate with and bolster 2 of the variables identified by the multivariate logistic regression analysis performed on our study group. Our model incorporates MELD, which is used in organ allocation for transplant and more widely used as a marker of the severity of a patient’s underlying liver disease. The original MELD paper reported a c-statistic of 0.87 for predicting death within 3 months for hospitalized patients.44 Our models incorporate MELD and also focus on in-hospital mortality, with a higher c-statistic (of 0.94)
718 | www.jcge.com
J Clin Gastroenterol
Volume 48, Number 8, September 2014
than the original MELD paper. When examined in our cohort, MELD alone as a predictor of in-hospital mortality had an AUROC of 0.84. This is an important consideration as our model does incorporate several elements (use of vasoactive agents, packed red blood cell transfusion) that may not be present upon or just after admission; thus, MELD alone, upon admission, could be viewed as an inhospital mortality predictor and can also inform clinician’s clinical judgment upon admission to the hospital, before endoscopic therapy. Although our study included several clear strengths, including a large patient cohort, extensive clinical data, and the inclusion of a large number of patients with variceal and nonvariceal bleeding, and was carried out in an era in which practice management has been uniform (our institution’s practice is to treat cirrhotic patients with upper GI bleeding with continuous proton-pump inhibitor infusion, octreotide infusion, and antibiotics upon admission)45 we recognize potential limitations. First, our study was performed at a large urban hospital that may not be representative of the demographics of the other groups, and thus could potentially limit the generalizability of the study. However, we feel that this potential weakness is also a strength because this cohort is likely to be more representative of a global population than has been examined in other studies. Our study was also performed at a single-site, and thus biases in the severity of clinical disease, or the nature of clinical practice at this institution could be present. However, we suspect this also to be unlikely, as outcomes were consistent with other published studies.4,5,13 Although variables such as MELD and packed red blood cell transfusion can change as a clinical course becomes more complicated, our model accounts for that dynamic change by being able to be recalculated as the patient’s clinical course becomes more involved. The retrospective nature of our study also posed several other limitations. Among these, active alcohol use was inconsistently documented. Further, as previously noted, nonsurvivors were less likely to be on b blockers (62% vs. 76%, P = 0.026). Antibiotic use was lower than might be expected; although antibiotic use in our cohort mirrored those in other institutions,46,47 its use was noted to progressively increase over time, starting at 51% in 2003 and rising to 84% in 2011. An important potential advantage of this model is that as mortality in variceal bleeding and nonvariceal bleeding seems to be similar, this model may free the clinician from relying on endoscopic data to formulate a comprehensive treatment plan. It is important to emphasize, however, that the risk score was developed in a setting in which it is a routine practice to pursue endoscopy aggressively. Thus, use of the risk score should not preclude performance of endoscopy and therapy if indicated. It is possible that the risk score could help identify patients in whom intervention is futile, but this would need to be validated in further studies, including in patients not undergoing endoscopic therapy. Development of a robust clinical risk scoring system for cirrhotics with upper GI bleeding is important for patients with cirrhosis for several reasons. First, in-hospital mortality in patients with cirrhosis and upper GI bleeding is higher than that seen in patients without cirrhosis who present with upper GI bleeding.26 Therefore, this patient population requires accurate triage and specific management and we recognize several potential applications of such a scoring system. For example, in patients who may be transplant candidates, recognizing a potentially poor r
2013 Lippincott Williams & Wilkins
J Clin Gastroenterol
Volume 48, Number 8, September 2014
outcome in the absence of transplant is imperative for expediting transplant work-up and subsequent listing. Alternatively, in patients who are not transplant candidates, such a scoring system may serve as a highly accurate measure of medical futility. Thus, we propose that a simple clinical risk score can help clinicians further appreciate the severity of their patient’s illness and identify those patients at high risk of mortality. REFERENCES 1. Gilbert DA. Epidemiology of upper gastrointestinal bleeding. Gastrointest Endosc. 1990;36:S8–S13. 2. Johanson JF. Curbing the costs of GI bleeding. Am J Gastroenterol. 1998;93:1384–1385. 3. Lecleire S, Di Fiore F, Merle V, et al. Acute upper gastrointestinal bleeding in patients with liver cirrhosis and in noncirrhotic patients: epidemiology and predictive factors of mortality in a prospective multicenter population-based study. J Clin Gastroenterol. 2005;39:321–327. 4. D’Amico G, De Franchis R. Upper digestive bleeding in cirrhosis. Post-therapeutic outcome and prognostic indicators. Hepatology. 2003;38:599–612. 5. Seo YS, Kim YH, Ahn SH, et al. Clinical features and treatment outcomes of upper gastrointestinal bleeding in patients with cirrhosis. J Korean Med Sci. 2008;23:635–643. 6. D’Amico G, Garcia-Tsao G, Pagliaro L. Natural history and prognostic indicators of survival in cirrhosis: a systematic review of 118 studies. J Hepatol. 2006;44:217–231. 7. D’Amico G, Morabito A, Pagliaro L, et al. Survival and prognostic indicators in compensated and decompensated cirrhosis. Dig Dis Sci. 1986;31:468–475. 8. Cerqueira RM, Andrade L, Correia MR, et al. Risk factors for in-hospital mortality in cirrhotic patients with oesophageal variceal bleeding. Eur J Gastroenterol Hepatol. 2012;24: 551–557. 9. de Franchis R. Evolving consensus in portal hypertension. Report of the Baveno IV consensus workshop on methodology of diagnosis and therapy in portal hypertension. J Hepatol. 2005;43:167–176. 10. Carbonell N, Pauwels A, Serfaty L, et al. Improved survival after variceal bleeding in patients with cirrhosis over the past two decades. Hepatology. 2004;40:652–659. 11. El-Serag HB, Everhart JE. Improved survival after variceal hemorrhage over an 11-year period in the Department of Veterans Affairs. Am J Gastroenterol. 2000;95:3566–3573. 12. Gonzalez-Gonzalez JA, Garcia-Compean D, VazquezElizondo G, et al. Nonvariceal upper gastrointestinal bleeding in patients with liver cirrhosis. Clinical features, outcomes and predictors of in-hospital mortality. A prospective study. Ann Hepatol. 2011;10:287–295. 13. Hsu YC, Liou JM, Chung CS, et al. Early risk stratification with simple clinical parameters for cirrhotic patients with acute upper gastrointestinal bleeding. Am J Emerg Med. 2010; 28:884–890. 14. Saltzman JR, Tabak YP, Hyett BH, et al. A simple risk score accurately predicts in-hospital mortality, length of stay, and cost in acute upper GI bleeding. Gastrointest Endosc. 2011; 74:1215–1224. 15. Imperiale TF, Dominitz JA, Provenzale DT, et al. Predicting poor outcome from acute upper gastrointestinal hemorrhage. Arch Intern Med. 2007;167:1291–1296. 16. Blatchford O, Murray WR, Blatchford M. A risk score to predict need for treatment for upper-gastrointestinal haemorrhage. Lancet. 2000;356:1318–1321. 17. Rockall TA, Logan RF, Devlin HB, et al. Risk assessment after acute upper gastrointestinal haemorrhage. Gut. 1996; 38:316–321. 18. Charlson ME, Pompei P, Ales KL, et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40:373–383. r
2013 Lippincott Williams & Wilkins
A RS System to Predict In-hospital Mortality
19. Charlson M, Szatrowski TP, Peterson J, et al. Validation of a combined comorbidity index. J Clin Epidemiol. 1994;47: 1245–1251. 20. Runyon BA. Management of adult patients with ascites due to cirrhosis: an update. Hepatology. 2009;49:2087–2107. 21. Ferenci P, Lockwood A, Mullen K, et al. Hepatic encephalopathy—definition, nomenclature, diagnosis, and quantification: final report of the working party at the 11th World Congresses of Gastroenterology, Vienna, 1998. Hepatology. 2002;35:716–721. 22. World Medical Association declaration of Helsinki. Recommendations guiding physicians in biomedical research involving human subjects. JAMA. 1997;277:925–926. 23. del Olmo JA, Pena A, Serra MA, et al. Predictors of morbidity and mortality after the first episode of upper gastrointestinal bleeding in liver cirrhosis. J Hepatol. 2000;32:19–24. 24. Hosmer DW, Hosmer T, Le Cessie S, et al. A comparison of goodness-of-fit tests for the logistic regression model. Stat Med. 1997;16:965–980. 25. Sanders DS, Carter MJ, Goodchap RJ, et al. Prospective validation of the Rockall risk scoring system for upper GI hemorrhage in subgroups of patients with varices and peptic ulcers. Am J Gastroenterol. 2002;97:630–635. 26. Ananthakrishnan AN, McGinley EL, Saeian K. Higher hospital volume predicts endoscopy but not the in-hospital mortality rate in patients with acute variceal hemorrhage. Gastrointest Endosc. 2009;69:221–229. 27. Chalasani N, Kahi C, Francois F, et al. Improved patient survival after acute variceal bleeding: a multicenter, cohort study. Am J Gastroenterol. 2003;98:653–659. 28. Hsu YC, Chung CS, Tseng CH, et al. Delayed endoscopy as a risk factor for in-hospital mortality in cirrhotic patients with acute variceal hemorrhage. J Gastroenterol Hepatol. 2009;24: 1294–1299. 29. Cheung J, Soo I, Bastiampillai R, et al. Urgent vs. non-urgent endoscopy in stable acute variceal bleeding. Am J Gastroenterol. 2009;104:1125–1129. 30. Thomopoulos K, Theocharis G, Mimidis K, et al. Improved survival of patients presenting with acute variceal bleeding. Prognostic indicators of short- and long-term mortality. Dig Liver Dis. 2006;38:899–904. 31. Ismail FW, Mumtaz K, Shah HA, et al. Factors predicting inhospital mortality in patients with cirrhosis hospitalized with gastro-esophageal variceal hemorrhage. Indian J Gastroenterol. 2006;25:240–243. 32. Myers RP, Papay KD, Shaheen AA, et al. Relationship between hospital volume and outcomes of esophageal variceal bleeding in the United States. Clin Gastroenterol Hepatol. 2008;6:789–798. 33. Jamal MM, Samarasena JB, Hashemzadeh M. Decreasing inhospital mortality for oesophageal variceal hemorrhage in the USA. Eur J Gastroenterol Hepatol. 2008;20:947–955. 34. 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;48:229–236. 35. Flores-Rendon AR, Gonzalez-Gonzalez JA, Garcia-Compean D, et al. Model for end stage of liver disease (MELD) is better than the Child-Pugh score for predicting in-hospital mortality related to esophageal variceal bleeding. Ann Hepatol. 2008;7: 230–234. 36. Zhao C, Chen SB, Zhou JP, et al. Prognosis of hepatic cirrhosis patients with esophageal or gastric variceal hemorrhage: multivariate analysis. Hepatobiliary Pancreat Dis Int. 2002;1:416–419. 37. Piqueras B, Banares R, Rincon D, et al. Predictive factors of the mortality of digestive hemorrhage caused by esophageal varices in elderly patients. Gastroenterol Hepatol. 2001;24:51–55. 38. Amitrano L, Guardascione MA, Manguso F, et al. The effectiveness of current acute variceal bleed treatments in unselected cirrhotic patients: refining short-term prognosis and risk factors. Am J Gastroenterol. 2012;107:1872–1878.
www.jcge.com |
719
Lyles et al
39. Lahiff C, Shields W, Cretu I, et al. Upper gastrointestinal bleeding: predictors of risk in a mixed patient group including variceal and nonvariceal haemorrhage. Eur J Gastroenterol Hepatol. 2012;24:149–154. 40. Sarwar S, Dilshad A, Khan AA, et al. Predictive value of Rockall score for rebleeding and mortality in patients with variceal bleeding. J Coll Physicians Surg Pak. 2007;17:253–256. 41. Vreeburg EM, Terwee CB, Snel P, et al. Validation of the Rockall risk scoring system in upper gastrointestinal bleeding. Gut. 1999;44:331–335. 42. Stanley AJ, Dalton HR, Blatchford O, et al. Multicentre comparison of the Glasgow Blatchford and Rockall scores in the prediction of clinical end-points after upper gastrointestinal haemorrhage. Aliment Pharmacol Ther. 2011;34:470–475. 43. Laursen SB, Hansen JM, Schaffalitzky de Muckadell OB. The Glasgow Blatchford score is the most accurate assessment of
720 | www.jcge.com
J Clin Gastroenterol
44. 45. 46.
47.
Volume 48, Number 8, September 2014
patients with upper gastrointestinal hemorrhage. Clin Gastroenterol Hepatol. 2012;10:1130–1135, e1. Kamath PS, Wiesner RH, Malinchoc M, et al. A model to predict survival in patients with end-stage liver disease. Hepatology. 2001;33:464–470. Garcia-Tsao G, Sanyal AJ, Grace ND, et al. Prevention and management of gastroesophageal varices and variceal hemorrhage in cirrhosis. Hepatology. 2007;46:922–938. Schlansky B, Lee B, Hartwell L, et al. Guideline adherence and outcomes in esophageal variceal hemorrhage: comparison of tertiary care and non-tertiary care settings. J Clin Gastroenterol. 2012;46:235–242. Cheung J, Wong W, Zandieh I, et al. Acute management and secondary prophylaxis of esophageal variceal bleeding: a western Canadian survey. Can J Gastroenterol. 2006;20: 531–534.
r
2013 Lippincott Williams & Wilkins
A Risk Scoring System to Predict In-hospital Mortality in Patients With Cirrhosis Presenting With Upper Gastrointestinal Bleeding Thomas Lyles, MD,* Alan Elliott, MAS, MBA,w and Don C. Rockey, MDz Goals: We aimed to develop a simple and practical risk scoring system to predict in-hospital mortality in cirrhotics presenting with upper gastrointestinal (GI) bleeding. Study: Extensive clinical data were captured in patients with documented cirrhosis who underwent endoscopic evaluation for upper GI bleeding between January 1, 2003 and June 30, 2011 at Parkland Memorial Hospital. Predictors of mortality were identified by multivariate regression analysis. Results: A total of 884 patients with cirrhosis admitted for upper GI bleeding were identified; 809 patients survived and 75 died (8.4%). The etiology of bleeding was similar in both groups, with bleeding attributed to esophageal varices in 59% of survivors and 60% of non-survivors (ulcer disease and other etiologies of bleeding accounted for the other causes of bleeding). Mortality was 8.6% and 8.3% in patients with variceal bleeding and nonvariceal bleeding, respectively. While survivors and those who died were similarly matched with regard to gender, age, ethnicity and etiology of cirrhosis, patients who died had lower systolic blood pressures, higher pulse rates and lower mean arterial pressures at admission than patients who survived. Non-survivors were more likely to be Childs C (61% vs. 19%, P < 0.001). Multivariate regression analysis identified the following 4 predictors of in-hospital mortality: use of vasoactive pressors, number of packed red blood cells transfused, model for end-stage liver disease (MELD) score, and serum albumin. A receiver operating characteristic curve including these 4 variables yielded an area under the receiver operating characteristic (AUROC) curve of 0.94 (95% confidence interval, 0.91-0.98). Classification and Regression Tree analysis yielded similar results, identifying vasoactive pressors and then MELD > 21 as the most important decision nodes for predicting death. By comparison, using the Rockall scoring system in the same patients, the AUROC curve was 0.70 (95% confidence interval, 0.64-0.76 and the comparison of the University of Texas Southwestern model to the Rockall model revealed P < 0.0001). A validation set comprised of 150 unique admissions between July 1, 2011 and July 31, 2012, had
Received for publication April 13, 2013; accepted September 16, 2013. From the *Division of Digestive and Liver Diseases, UT-Southwestern Medical Center and the Parkland Health and Hospital System; wDepartment of Statistical Science, Southern Methodist University, Dallas, TX; and zDepartment of Internal Medicine, Medical University of South Carolina, Charleston, SC. T.L., A.E., and D.C.R. had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. T.L., D.C.R.: study concept and design, drafting of the manuscript, critical revision of the manuscript for important intellectual content, administrative, technical, or material support, study supervision; T.L., A.E., D.C.R.: analysis and interpretation of data, statistical analysis; T.L.: acquisition of data. The authors declare that they have nothing to disclose. Reprints: Don C. Rockey, MD, Department of Internal Medicine, Medical University of South Carolina, 96 Jonathan Lucas Street, Suite 803, MSC 623, Charleston, SC 29425 (e-mail: rockey@ musc.edu). Supplemental Digital Content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Website, www.jcge.com. Copyright r 2013 by Lippincott Williams & Wilkins
712 | www.jcge.com
an AUROC of 0.92, and the outcomes of 97% of the subjects in this set were accurately predicted by the risk score model. Conclusions: Use of vasoactive agents, packed red blood cell transfusion, albumin, and MELD score were highly predictive of in-hospital mortality in cirrhotics presenting with upper GI bleeding. These variables were used to formulate a clinical risk scoring system for in-hospital mortality, which is available at: http://medweb.musc.edu/LogisticModelPredictor. Key Words: varices, ulcer, outcome, prediction, endoscopy, model
(J Clin Gastroenterol 2014;48:712–720)
U
pper gastrointestinal (GI) bleeding is an important and common clinical problem in the United States, accounting for approximately 300,000 hospitalizations annually.1–3 Outcomes for cirrhotic patients with upper GI bleeding are worse than for those who do not have cirrhosis.3 Estimation of prognosis is an essential element of managing patients with cirrhosis admitted to a hospital. Longitudinal studies have previously found that stage of cirrhosis, defined by consensus as stage I to IV based on presence of clinical complications and global liver function [Child-Pugh class, model for end-stage liver disease (MELD) score] are important prognosticators of a patient’s overall mortality risk.4–8 By definition, patients with cirrhosis and GI bleeding are classified as stage IV; it has been reported that patients within this category have a 1-year mortality risk of 57%.9 Moreover, half of these patients will die within the 6 weeks after their admission for GI bleeding.3,5,6,10,11 Although several risk-stratification tools have been formulated for each episode of acute GI bleeding and for estimation of 6-week mortality after an episode of variceal bleeding in cirrhosis,3–5,12,13 a stratification model to estimate in-hospital mortality in all cirrhotics admitted with acute upper GI bleeding is not currently available. This seems to be in part due to the perceived distinction between variceal bleeding and nonvariceal bleeding. Although variceal bleeding has historically carried the higher mortality rate, this may not be the case.4,5,12 Prediction of outcome in patients with acute upper GI bleeding seems to be feasible.14–17 Thus, we have taken advantage of a large and robust cohort of patients with cirrhosis and upper GI bleeding to focus on in-hospital outcomes. Our goal was to develop a simple, and practical clinical tool specifically to help estimate in-hospital mortality risk in patients with cirrhosis and all forms of acute upper GI bleeding, whether from variceal or nonvariceal upper GI bleeding.
MATERIALS AND METHODS This cohort study focused on predictors of in-hospital mortality in cirrhotics who were admitted to the hospital
J Clin Gastroenterol
Volume 48, Number 8, September 2014
J Clin Gastroenterol
Volume 48, Number 8, September 2014
with a diagnosis of upper GI bleeding. The study included consecutive patients admitted to Parkland Memorial Hospital (Dallas, TX), a University of Texas Southwestern (UTSW) teaching hospital, from January 1, 2003 through June 30, 2011. Inclusion required a definitive diagnosis of cirrhosis and documented acute upper GI bleeding. Patient data were captured through the unit’s GI Bleeding Healthcare Registry, which prospectively collects data on patients admitted to Parkland Memorial Hospital with any form of gastrointestinal bleeding and through the unit’s Chronic Liver Disease Registry, which collects data on consecutive cirrhotic patients admitted to our institution. Data captured in the GI Bleeding Healthcare Registry includes multiple clinical and historical features, including the American Society of Anesthesiologists score on physical status, medications, laboratory and endoscopic data (time to endoscopy), and clinical features associated with GI bleeding. Primary hemostasis rates, treatment failures, and 30-day rebleeding events are also collected. Outcomes collected include rebleeding, overall length of stay, and mortality. Rebleeding is defined as visualization of vomited red blood, a drop in hematocrit of Z9 points (or hemoglobin 3 g/dL) after endoscopy or by development of hypotension (SBPr90) more than 2 hours after endoscopy. By design, a bleeding lesion or a lesion with stigmata of recent bleeding in any given case is designated as the primary diagnosis. When >1 lesion/diagnosis is present in addition to a primary lesion, it is considered a secondary lesion, but not deemed to be the cause of hemorrhage. Primary bleeding lesions are assigned by a 3-panel group to one of the 18 specific primary diagnoses (or other/no source identified). Causes of death for all patients are classified into 8 different groups, also adjudicated by the study group, which include the following: gastrointestinal bleeding, cardiorespiratory failure, renal failure, liver failure, sepsis, multiorgan system dysfunction, malignancy, unknown, or others. The Chronic Liver Disease Registry similarly collects extensive clinical data, including the same and additional features highlighted above, such as patient demographics, medical history and clinical data, hospital course data [intensive care unit (ICU) stay, intubation, use of vasoactive agents such as dopamine, dobutamine, norepinephrine, etc.], medications (including b blockers at admission and discharge, antibiotics), laboratory data, physical examination findings relevant to cirrhosis (ascites, portosystemic encephalopathy, etc.), endoscopic features, and outcomes. To compare co-morbidities between survivors and nonsurvivors, the Charlson Comorbidity Index18,19 was calculated on each patient in the test set. Laboratory data were captured at the time of admission. MELD and ChildPugh score were calculated upon admission with current laboratory markers and physical examination findings. For varices, additional data were collected such as the number and grade of varices, use of band ligation as hemostasis, and number of bands placed. Ascites were divided into 2 categories: patients without ascites or ascites only detectable by imaging were grouped together and those with clinically evident ascites were grouped together.20 Patients with diuretic controlled ascites were coded in the former group. Portosystemic encephalopathy was determined upon admission by the admitting physician or hepatology consultant and reported in accordance with the West Haven criteria.21 Cirrhosis was defined as follows: a history consistent with chronic liver disease, plus the presence of clinical r
2013 Lippincott Williams & Wilkins
A RS System to Predict In-hospital Mortality
features consistent with cirrhosis, including clinical findings of cirrhosis and portal hypertension (spider angiomata, gynecomastia, splenomegaly, thrombocytopenia) or a documented complication of chronic liver disease (ie, ascites, varices, hepatic encephalopathy), and/or imaging consistent with cirrhosis, and/or liver histology consistent with cirrhosis. Upper GI bleeding was defined as reported or witnessed melena, hematemesis, coffee ground emesis, or hematochezia (with a documented upper GI tract lesion) in the setting of at least a 4-point drop in hematocrit from baseline or lower than normal. It is a standard practice at our institution to place patients on continuous protonpump inhibitor infusion, octreotide infusion, and antibiotics upon admission and to perform esophagogastroduodenoscopy in patients with any form of upper GI bleeding as defined above unless contraindications to endoscopy exist. In the event of esophageal variceal bleeding, variceal ligation was preferentially used during the study period. Sclerotherapy and transjugular intrahepatic portosystemic shunt were used if bleeding could not be controlled with variceal ligation. Treatment of nonvariceal bleeding was left to the discretion of the endoscopist and included the use of epinephrine injection therapy, endoclip hemostasis, bipolar coagulation therapy, and combinations of the aforementioned modalities. Notably, our institution’s management of upper GI bleeding (including pharmacologic and endoscopic therapy) did not significantly change over the course of the study period, although it should be emphasized that efforts to improve standardization have been ongoing. With regard to packed red blood cell transfusion, the practice in our institution is that the hemoglobin is typically targeted to be 8 to 9 g/dL. However, in our experience, the cutoff for ideal transfusion varies among practitioners, and depends on individual patients and co-morbidities. Because of this variability, our institution currently does not attempt to impose specific transfusion practices. Although the reference Hgb level for transfusion is recommended to be 8 to 9 g/dL, the use of packed red blood cell transfusion and number of units transfused was left to the discretion of the admitting physician in conjunction with the gastroenterology consult service. Patients were excluded from this analysis if they had a form of GI bleeding other than an upper GI source, were under 18 years old, did not have endoscopy during their hospitalization, or were pregnant. Patients were not excluded if they had hepatocellular carcinoma, previous transjugular intrahepatic portosystemic shunt/surgical shunt, or bleeding from gastrointestinal malignancy. For patients who had multiple hospitalizations for bleeding during the study period, rebleeding after 42 days from the index bleed was considered a new bleed. A readmission within 42 days from the index bleed was considered a rebleed and this readmission data were excluded from the risk score analysis. The study was approved by the University of Texas Southwestern Institutional Review Board and met all criteria for good clinical practice.22 A second validation cohort was compiled from cirrhotic patients admitted to our institution between July 1, 2011 and July 31, 2012. The aforementioned definitions were applied to this validation cohort to ensure uniformity between the 2 groups. After formulating the in-hospital mortality risk score, this second set was used to validate the model. www.jcge.com |
713
Lyles et al
Descriptive statistics were used to present demographic and clinical characteristics of patients in the study cohort. The primary outcome of this study was in-hospital mortality. We specifically aimed to identify prognostic variables of in-hospital mortality in cirrhotics with upper GI bleeding, and to incorporate those variables into a clinically relevant risk-stratification score. Patients were divided into 2 groups, those surviving hospitalization and those who did not survive. Group characteristics and demographics were compared between groups using Student t test for continuous variables or w2 for categorical variables where appropriate. To examine variables for their predictive relationship to mortality, univariate logistic regression was utilized to examine possible predictors. Variables that were at least marginally predictive (P < 0.2) in the univariate case were considered for a multivariate logistic model. A final logistic regression model included only those variables whose multivariate combination was most predictive for the binary outcome variable mortality. Model fit was assessed using the area under the receiver operating characteristic (AUROC) curve and a HosmerLemeshow goodness-of-fit analysis. A second analysis was performed using the Classification and Regression Tree (CART) technique. This method uses recursive partitioning to select variables most predictive of mortality, and creates a tree-structured model that can be analyzed to assess probability of survival. All P-values in the analyses presented are 2 sided and considered significant when Pr0.05 unless otherwise stated. No adjustments for multiple testing were made. The data were analyzed using the SAS V 9.2 (SAS Inc., Cary, NC) statistical package and CART software (version 6.0; Salford Systems, San Diego, CA).
J Clin Gastroenterol
Volume 48, Number 8, September 2014
TABLE 1. Demographic and Clinical Characteristics
Mean ± SD/n (%)
Age* (y) Sex, female (%) Ethnicity Hispanic White AfricanAmerican Others Cirrhosis etiology HCV Alcohol Cryptogenic NAFLD HCV + HBV HBV Others Systolic blood pressure Mean arterial pressure Pulse Presence of ascites Presence of PSE Charlson score CTP class A B C
Survivors (n = 809)
Nonsurvivors (n = 75)
51 (± 9) 217 (27)
51 (± 10) 25 (33)
413 (51) 256 (32) 114 (14)
33 (44) 21 (28) 17 (23)
26 (3)
4 (5)
P 0.514 0.226 0.152
0.21 362 315 32 33 22 22 23 122
(45) (39) (4) (4) (3) (3) (2) (± 23)
88 (± 17) 96 464 95 3
(± 20) (57) (12) (2)
265 (33) 387 (48) 157 (19)
42 24 5 0 2 1 1 112
(56) (32) (7) (0) (3) (1) (1) (± 23)
< 0.001
82 (± 17)
0.001
103 34 23 4
(± 17) (45) (31) (2)
0.003 0.045 < 0.001 0.426 < 0.001
8 (11) 21 (28) 46 (61)
*Data provided are those upon the patient’s admission to the hospital. CTP indicates Child-Turcotte-Pugh; HBV, hepatitis B virus; HCV, hepatitis C virus; NAFLD, nonalcoholic fatty liver disease; PSE, portosystemic encephalopathy.
RESULTS
FIGURE 1. Flow diagram of the patient cohort. A total of 2548 patients were admitted to our institution with acute GI bleeding in the study period. After excluding patients with lower GI bleeding, those who were not cirrhotic and who did not undergo esophagogastroduodenoscopy, the study cohort was composed of 884 patients. EGD indicates esophagogastroduodenoscopy; GI, gastrointestinal.
714 | www.jcge.com
We identified a total of 884 unique admissions of patients with cirrhosis and upper GI bleeding (Fig. 1). This cohort included 521 patients with variceal bleeding and 363 patients with nonvariceal bleeding. Seventy-five patients (8.4%) died during hospitalization. The cohort included patients with typical causes of cirrhosis, was predominantly male, was relatively young with a mean age of 50 years, and included a high proportion of patients of Hispanic ethnicity (Table 1). Although survivors and those who died were similarly matched with regard to sex, age, ethnicity, and etiology of cirrhosis, there were several key differences. Patients who died had lower systolic blood pressures, higher pulse rates, and lower mean arterial pressure than patients who survived. In addition, patients who died more frequently had ascites, and when present it was more advanced than survivors. Patients who died also had worse global liver dysfunction than survivors, as evidenced by a greater proportion with Child class C and higher MELD score (Tables 1, 2). Notably, patients who died were less likely to be on b blockers at the time of their admission (62% vs. 76%, P = 0.026). There were a number of notable differences in laboratory abnormalities among survivors and nonsurvivors (Table 2). Nonsurvivors had higher white blood cell counts, higher blood urea nitrogen, lower sodium, higher creatinine, higher international normalized ratio, higher bilirubin, lower albumin, and had significantly higher MELD r
2013 Lippincott Williams & Wilkins
J Clin Gastroenterol
Volume 48, Number 8, September 2014
TABLE 2. Laboratory Data
Mean ± SD Survivors (n = 809) WBC ( 109/L) Hematocrit (%) Platelets ( 109/L) BUN (mg/dL) Sodium (mmol/L) Creatinine (mg/dL) INR Total bilirubin (mg/dL) Albumin (g/dL) MELD MELD-Na Positive blood cultures [n (%)]
Nonsurvivors (n = 75)
P
8±5 28 ± 8 117 ± 70 25 ± 16 135 ± 5 1.0 ± 0.9 1.5 ± 0.7 2.6 ± 3.8
12 ± 6 26 ± 7 119 ± 67 33 ± 22 132 ± 6 1.9 ± 1.4 2.1 ± 0.8 8.7 ± 10.6
< 0.001 0.044 0.743 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001
2.9 ± 0.6 14 ± 6 17 ± 6 23 (3)
2.3 ± 0.6 25 ± 9 27 ± 8 19 (25)
< 0.001 < 0.001 < 0.001 < 0.001
BUN indicates blood urea nitrogen; INR, international normalized ratio; MELD, model for end-stage liver disease; Na, sodium; WBC, white blood cells.
scores. Patients who did not survive were more likely to have a positive blood culture (3% for survivors vs. 25% for nonsurvivors, P < 0.001). Hematocrit and platelet levels were similar in the 2 groups. The etiology of upper GI bleeding was similar in both groups, with bleeding in each group attributed to esophageal varices approximately 60% of the time (Table 3), consistent with previous reports.3,4,23 Esophageal varices were identified in 739 patients overall, and were the culprit lesion in 521 patients. Our institution’s practice is to preferentially treat bleeding esophageal varices with band ligation therapy; sclerotherapy and transjugular intrahepatic portosystemic shunt are performed if bleeding cannot be controlled with band ligation and/or culprit lesions are not amenable to banding (ie, significant esophageal scarring from prior banding sessions). Outcomes among the 2 groups were also significantly different (Table 4). Those who died had longer hospitalizations, were more likely to have active bleeding at time of endoscopy, require blood transfusion, and when transfused, received a greater number of packed red blood cells.
TABLE 3. Etiology of Upper Gastrointestinal Bleeding
N (%) Survivors (n = 809) Esophageal varices Gastro-duodenal ulcer Esophagitis* Othersw Portal hypertensive gastropathy Mallory-Weiss tear Banding ulcer No clear source
476 77 48 39 82
(59) (10) (6) (4) (10)
24 (3) 9 (1) 63 (8)
Nonsurvivors (n = 75) 45 8 8 10 3
(60) (11) (11) (13) (4)
1 (1) 0 (0) 0 (0)
P 0.90 0.54 0.13 0.006 0.10 0.72 1.0 1.0
*Esophagitis + GE junction ulcer + esophageal ulcer + banding ulcer. wGastric mass, hemobilia, oropharyngeal bleeding, coagulopathic oozing, GAVE, Dieulafoy, arteriovenous malformation, gastritis, gastric/duodenal erosions.
r
2013 Lippincott Williams & Wilkins
A RS System to Predict In-hospital Mortality
Patients who died were also more likely to be admitted to the ICU, require vasoactive agents and intubation and mechanical ventilation. Of the 53 patients requiring vasoactive agents, only 7 survived the hospitalization. One hundred twenty-six patients were intubated and of these, 69 survived their hospitalization. Overall, 49 patients required both vasoactive agents and intubation; of these, 5 lived. We found that the mortality rate in the 2 cohorts was identical, with an in-hospital mortality of 8.6% in patients with variceal bleeding and 8.3% in patients nonvariceal bleeding (Supplementary Table 1, Supplemental Digital Content 2, http://links.lww.com/JCG/A98). Because the 2 groups had similar in-hospital mortality rates, regression analysis and CART were focused on detecting significant variables in the overall population of cirrhotics with upper GI bleeding, with in-hospital mortality as the endpoint. Univariate and multivariate logistic regression analysis were performed. We included 57 different clinical variables in the univariate analysis. Those with a P-value 21 as the second decision node for predicting inhospital mortality (Fig. 3). We next sought to validate the primary logistic regression model using a prospectively collected cohort (July 1, 2011 and July 31, 2012). This set was composed of 150 unique admissions and had an in-hospital mortality rate of 6.7%. Clinical characteristics of this group were similar to patients in the primary group (Supplementary Table 3, Supplemental Digital Content 2, http://links. lww.com/JCG/A98). The outcomes of 97% of the subjects in the validation set were correctly identified. However, 3 predicted deaths out of 10 were incorrect and 2 predicted survivors out of 140 were incorrect, which indicates a sensitivity of 97.8% and specificity of 78% (positive predictive value of 98.7% and a negative predictive value of 70%). The AUROC for the validation set is 0.9164 (95% CI, 0.870.95; Fig. 2C). When plotted on respective histograms, the training data set and validation data set had good
716 | www.jcge.com
vasoactive agents, and endotracheal intubation, and the subsequent development of clinical complications including renal insufficiency, cardiopulmonary complications, and infection. Considerable effort has been directed at prediction of outcome in patients with acute upper GI bleeding14,15 and previously identified multiple potential risk factors for mortality such as increasing age, male sex, being uninsured, physical examination findings (systolic blood pressure 18, it was a smaller overall study of 102 patients8 and there were other notable differences including a retrospective design, use of terlipressin from 2006 to 2008 in their study, no hospital course data including ICU-related data, and an AUROC of 0.76. Their findings, however, correlate with and bolster 2 of the variables identified by the multivariate logistic regression analysis performed on our study group. Our model incorporates MELD, which is used in organ allocation for transplant and more widely used as a marker of the severity of a patient’s underlying liver disease. The original MELD paper reported a c-statistic of 0.87 for predicting death within 3 months for hospitalized patients.44 Our models incorporate MELD and also focus on in-hospital mortality, with a higher c-statistic (of 0.94)
718 | www.jcge.com
J Clin Gastroenterol
Volume 48, Number 8, September 2014
than the original MELD paper. When examined in our cohort, MELD alone as a predictor of in-hospital mortality had an AUROC of 0.84. This is an important consideration as our model does incorporate several elements (use of vasoactive agents, packed red blood cell transfusion) that may not be present upon or just after admission; thus, MELD alone, upon admission, could be viewed as an inhospital mortality predictor and can also inform clinician’s clinical judgment upon admission to the hospital, before endoscopic therapy. Although our study included several clear strengths, including a large patient cohort, extensive clinical data, and the inclusion of a large number of patients with variceal and nonvariceal bleeding, and was carried out in an era in which practice management has been uniform (our institution’s practice is to treat cirrhotic patients with upper GI bleeding with continuous proton-pump inhibitor infusion, octreotide infusion, and antibiotics upon admission)45 we recognize potential limitations. First, our study was performed at a large urban hospital that may not be representative of the demographics of the other groups, and thus could potentially limit the generalizability of the study. However, we feel that this potential weakness is also a strength because this cohort is likely to be more representative of a global population than has been examined in other studies. Our study was also performed at a single-site, and thus biases in the severity of clinical disease, or the nature of clinical practice at this institution could be present. However, we suspect this also to be unlikely, as outcomes were consistent with other published studies.4,5,13 Although variables such as MELD and packed red blood cell transfusion can change as a clinical course becomes more complicated, our model accounts for that dynamic change by being able to be recalculated as the patient’s clinical course becomes more involved. The retrospective nature of our study also posed several other limitations. Among these, active alcohol use was inconsistently documented. Further, as previously noted, nonsurvivors were less likely to be on b blockers (62% vs. 76%, P = 0.026). Antibiotic use was lower than might be expected; although antibiotic use in our cohort mirrored those in other institutions,46,47 its use was noted to progressively increase over time, starting at 51% in 2003 and rising to 84% in 2011. An important potential advantage of this model is that as mortality in variceal bleeding and nonvariceal bleeding seems to be similar, this model may free the clinician from relying on endoscopic data to formulate a comprehensive treatment plan. It is important to emphasize, however, that the risk score was developed in a setting in which it is a routine practice to pursue endoscopy aggressively. Thus, use of the risk score should not preclude performance of endoscopy and therapy if indicated. It is possible that the risk score could help identify patients in whom intervention is futile, but this would need to be validated in further studies, including in patients not undergoing endoscopic therapy. Development of a robust clinical risk scoring system for cirrhotics with upper GI bleeding is important for patients with cirrhosis for several reasons. First, in-hospital mortality in patients with cirrhosis and upper GI bleeding is higher than that seen in patients without cirrhosis who present with upper GI bleeding.26 Therefore, this patient population requires accurate triage and specific management and we recognize several potential applications of such a scoring system. For example, in patients who may be transplant candidates, recognizing a potentially poor r
2013 Lippincott Williams & Wilkins
J Clin Gastroenterol
Volume 48, Number 8, September 2014
outcome in the absence of transplant is imperative for expediting transplant work-up and subsequent listing. Alternatively, in patients who are not transplant candidates, such a scoring system may serve as a highly accurate measure of medical futility. Thus, we propose that a simple clinical risk score can help clinicians further appreciate the severity of their patient’s illness and identify those patients at high risk of mortality. REFERENCES 1. Gilbert DA. Epidemiology of upper gastrointestinal bleeding. Gastrointest Endosc. 1990;36:S8–S13. 2. Johanson JF. Curbing the costs of GI bleeding. Am J Gastroenterol. 1998;93:1384–1385. 3. Lecleire S, Di Fiore F, Merle V, et al. Acute upper gastrointestinal bleeding in patients with liver cirrhosis and in noncirrhotic patients: epidemiology and predictive factors of mortality in a prospective multicenter population-based study. J Clin Gastroenterol. 2005;39:321–327. 4. D’Amico G, De Franchis R. Upper digestive bleeding in cirrhosis. Post-therapeutic outcome and prognostic indicators. Hepatology. 2003;38:599–612. 5. Seo YS, Kim YH, Ahn SH, et al. Clinical features and treatment outcomes of upper gastrointestinal bleeding in patients with cirrhosis. J Korean Med Sci. 2008;23:635–643. 6. D’Amico G, Garcia-Tsao G, Pagliaro L. Natural history and prognostic indicators of survival in cirrhosis: a systematic review of 118 studies. J Hepatol. 2006;44:217–231. 7. D’Amico G, Morabito A, Pagliaro L, et al. Survival and prognostic indicators in compensated and decompensated cirrhosis. Dig Dis Sci. 1986;31:468–475. 8. Cerqueira RM, Andrade L, Correia MR, et al. Risk factors for in-hospital mortality in cirrhotic patients with oesophageal variceal bleeding. Eur J Gastroenterol Hepatol. 2012;24: 551–557. 9. de Franchis R. Evolving consensus in portal hypertension. Report of the Baveno IV consensus workshop on methodology of diagnosis and therapy in portal hypertension. J Hepatol. 2005;43:167–176. 10. Carbonell N, Pauwels A, Serfaty L, et al. Improved survival after variceal bleeding in patients with cirrhosis over the past two decades. Hepatology. 2004;40:652–659. 11. El-Serag HB, Everhart JE. Improved survival after variceal hemorrhage over an 11-year period in the Department of Veterans Affairs. Am J Gastroenterol. 2000;95:3566–3573. 12. Gonzalez-Gonzalez JA, Garcia-Compean D, VazquezElizondo G, et al. Nonvariceal upper gastrointestinal bleeding in patients with liver cirrhosis. Clinical features, outcomes and predictors of in-hospital mortality. A prospective study. Ann Hepatol. 2011;10:287–295. 13. Hsu YC, Liou JM, Chung CS, et al. Early risk stratification with simple clinical parameters for cirrhotic patients with acute upper gastrointestinal bleeding. Am J Emerg Med. 2010; 28:884–890. 14. Saltzman JR, Tabak YP, Hyett BH, et al. A simple risk score accurately predicts in-hospital mortality, length of stay, and cost in acute upper GI bleeding. Gastrointest Endosc. 2011; 74:1215–1224. 15. Imperiale TF, Dominitz JA, Provenzale DT, et al. Predicting poor outcome from acute upper gastrointestinal hemorrhage. Arch Intern Med. 2007;167:1291–1296. 16. Blatchford O, Murray WR, Blatchford M. A risk score to predict need for treatment for upper-gastrointestinal haemorrhage. Lancet. 2000;356:1318–1321. 17. Rockall TA, Logan RF, Devlin HB, et al. Risk assessment after acute upper gastrointestinal haemorrhage. Gut. 1996; 38:316–321. 18. Charlson ME, Pompei P, Ales KL, et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40:373–383. r
2013 Lippincott Williams & Wilkins
A RS System to Predict In-hospital Mortality
19. Charlson M, Szatrowski TP, Peterson J, et al. Validation of a combined comorbidity index. J Clin Epidemiol. 1994;47: 1245–1251. 20. Runyon BA. Management of adult patients with ascites due to cirrhosis: an update. Hepatology. 2009;49:2087–2107. 21. Ferenci P, Lockwood A, Mullen K, et al. Hepatic encephalopathy—definition, nomenclature, diagnosis, and quantification: final report of the working party at the 11th World Congresses of Gastroenterology, Vienna, 1998. Hepatology. 2002;35:716–721. 22. World Medical Association declaration of Helsinki. Recommendations guiding physicians in biomedical research involving human subjects. JAMA. 1997;277:925–926. 23. del Olmo JA, Pena A, Serra MA, et al. Predictors of morbidity and mortality after the first episode of upper gastrointestinal bleeding in liver cirrhosis. J Hepatol. 2000;32:19–24. 24. Hosmer DW, Hosmer T, Le Cessie S, et al. A comparison of goodness-of-fit tests for the logistic regression model. Stat Med. 1997;16:965–980. 25. Sanders DS, Carter MJ, Goodchap RJ, et al. Prospective validation of the Rockall risk scoring system for upper GI hemorrhage in subgroups of patients with varices and peptic ulcers. Am J Gastroenterol. 2002;97:630–635. 26. Ananthakrishnan AN, McGinley EL, Saeian K. Higher hospital volume predicts endoscopy but not the in-hospital mortality rate in patients with acute variceal hemorrhage. Gastrointest Endosc. 2009;69:221–229. 27. Chalasani N, Kahi C, Francois F, et al. Improved patient survival after acute variceal bleeding: a multicenter, cohort study. Am J Gastroenterol. 2003;98:653–659. 28. Hsu YC, Chung CS, Tseng CH, et al. Delayed endoscopy as a risk factor for in-hospital mortality in cirrhotic patients with acute variceal hemorrhage. J Gastroenterol Hepatol. 2009;24: 1294–1299. 29. Cheung J, Soo I, Bastiampillai R, et al. Urgent vs. non-urgent endoscopy in stable acute variceal bleeding. Am J Gastroenterol. 2009;104:1125–1129. 30. Thomopoulos K, Theocharis G, Mimidis K, et al. Improved survival of patients presenting with acute variceal bleeding. Prognostic indicators of short- and long-term mortality. Dig Liver Dis. 2006;38:899–904. 31. Ismail FW, Mumtaz K, Shah HA, et al. Factors predicting inhospital mortality in patients with cirrhosis hospitalized with gastro-esophageal variceal hemorrhage. Indian J Gastroenterol. 2006;25:240–243. 32. Myers RP, Papay KD, Shaheen AA, et al. Relationship between hospital volume and outcomes of esophageal variceal bleeding in the United States. Clin Gastroenterol Hepatol. 2008;6:789–798. 33. Jamal MM, Samarasena JB, Hashemzadeh M. Decreasing inhospital mortality for oesophageal variceal hemorrhage in the USA. Eur J Gastroenterol Hepatol. 2008;20:947–955. 34. 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;48:229–236. 35. Flores-Rendon AR, Gonzalez-Gonzalez JA, Garcia-Compean D, et al. Model for end stage of liver disease (MELD) is better than the Child-Pugh score for predicting in-hospital mortality related to esophageal variceal bleeding. Ann Hepatol. 2008;7: 230–234. 36. Zhao C, Chen SB, Zhou JP, et al. Prognosis of hepatic cirrhosis patients with esophageal or gastric variceal hemorrhage: multivariate analysis. Hepatobiliary Pancreat Dis Int. 2002;1:416–419. 37. Piqueras B, Banares R, Rincon D, et al. Predictive factors of the mortality of digestive hemorrhage caused by esophageal varices in elderly patients. Gastroenterol Hepatol. 2001;24:51–55. 38. Amitrano L, Guardascione MA, Manguso F, et al. The effectiveness of current acute variceal bleed treatments in unselected cirrhotic patients: refining short-term prognosis and risk factors. Am J Gastroenterol. 2012;107:1872–1878.
www.jcge.com |
719
Lyles et al
39. Lahiff C, Shields W, Cretu I, et al. Upper gastrointestinal bleeding: predictors of risk in a mixed patient group including variceal and nonvariceal haemorrhage. Eur J Gastroenterol Hepatol. 2012;24:149–154. 40. Sarwar S, Dilshad A, Khan AA, et al. Predictive value of Rockall score for rebleeding and mortality in patients with variceal bleeding. J Coll Physicians Surg Pak. 2007;17:253–256. 41. Vreeburg EM, Terwee CB, Snel P, et al. Validation of the Rockall risk scoring system in upper gastrointestinal bleeding. Gut. 1999;44:331–335. 42. Stanley AJ, Dalton HR, Blatchford O, et al. Multicentre comparison of the Glasgow Blatchford and Rockall scores in the prediction of clinical end-points after upper gastrointestinal haemorrhage. Aliment Pharmacol Ther. 2011;34:470–475. 43. Laursen SB, Hansen JM, Schaffalitzky de Muckadell OB. The Glasgow Blatchford score is the most accurate assessment of
720 | www.jcge.com
J Clin Gastroenterol
44. 45. 46.
47.
Volume 48, Number 8, September 2014
patients with upper gastrointestinal hemorrhage. Clin Gastroenterol Hepatol. 2012;10:1130–1135, e1. Kamath PS, Wiesner RH, Malinchoc M, et al. A model to predict survival in patients with end-stage liver disease. Hepatology. 2001;33:464–470. Garcia-Tsao G, Sanyal AJ, Grace ND, et al. Prevention and management of gastroesophageal varices and variceal hemorrhage in cirrhosis. Hepatology. 2007;46:922–938. Schlansky B, Lee B, Hartwell L, et al. Guideline adherence and outcomes in esophageal variceal hemorrhage: comparison of tertiary care and non-tertiary care settings. J Clin Gastroenterol. 2012;46:235–242. Cheung J, Wong W, Zandieh I, et al. Acute management and secondary prophylaxis of esophageal variceal bleeding: a western Canadian survey. Can J Gastroenterol. 2006;20: 531–534.
r
2013 Lippincott Williams & Wilkins
