Does This Patient Have a Severe Upper Gastrointestinal Bleed?

THE RATIONAL CLINICIAN’S CORNER CLINICAL EXAMINATION

F. Douglas Srygley, MD Context Emergency physicians must determine both the location and the severity Charles J. Gerardo, MD of acute gastrointestinal bleeding (GIB) to optimize the diagnostic and therapeutic ap- Tony Tran, MD proaches. Objectives To identify the historical features, symptoms, signs, bedside maneu- Deborah A. Fisher, MD, MHS vers, and basic laboratory test results that distinguish acute upper GIB (UGIB) from acute lower GIB (LGIB) and to risk stratify those patients with a UGIB least likely to CLINICAL SCENARIO have severe bleeding that necessitates an urgent intervention. Case 1 Data Sources A structured search of MEDLINE (1966-September 2011) and refer- A 62-year-old man presents to the emer- ence lists from retrieved articles, review articles, and physical examination textbooks. gency department after an episode of Study Selection High-quality studies were included of adult patients who were either syncope following several weeks of fa- admitted with GIB or evaluated in emergency departments with bedside evaluations tigue. He has a history of upper gastro- and/or routine laboratory tests, and studies that did not include endoscopic findings intestinal bleeding (UGIB) due to pep- in prediction models. The initial search yielded 2628 citations, of which 8 were re- tic ulcer disease. His blood pressure is tained that tested methods of identifying a UGIB and 18 that identified methods of 90/60 mm Hg, pulse of 105/min, and a determining the severity of UGIB. black, foul-smelling stool is found upon Data Extraction One author abstracted the data (prevalence, sensitivity, specific- rectal examination. Blood test results ity, and likelihood ratios [LRs]) and assessed methodological quality, with confirma- show a hemoglobin level of 6.5 g/dL, a tion by another author. Data were combined using random effects measures. creatinine level of 1.0 mg/dL, and a se- Data Synthesis The majority of patients (N=1776) had an acute UGIB (preva- rum urea nitrogen level of 55 mg/dL. lence, 63%; 95% CI, 51%-73%). Several clinical factors increase the likelihood that Would the results of a nasogastric la- a patient has a UGIB, including a patient-reported history of melena (LR range, 5.1- vage help determine between an up- 5.9), melenic stool on examination (LR, 25; 95% CI, 4-174), a nasogastric lavage per endoscopy or a colonoscopy as the with blood or coffee grounds (LR, 9.6; 95% CI, 4.0-23.0), and a serum urea nitro- test most likely to identify the bleed- gen:creatinine ratio of more than 30 (summary LR, 7.5; 95% CI, 2.8-12.0). Con- versely, the presence of blood clots in stool (LR, 0.05; 95% CI, 0.01-0.38) ing source? decreases the likelihood of a UGIB. Of the patients clinically diagnosed with acute UGIB, 36% (95% CI, 29%-44%) had severe bleeding. A nasogastric lavage with Case 2 red blood (summary LR, 3.1; 95% CI, 1.2-14.0), tachycardia (LR, 4.9; 95% CI, 3.2- A 45-year-old woman presents with 24 7.6), or a hemoglobin level of less than 8 g/dL (LR range, 4.5-6.2) increase the like- hours of diarrhea, nausea, and 2 epi- lihood of a severe UGIB requiring urgent intervention. A Blatchford score of 0 (sum- sodes of vomiting coffee ground mate- mary LR, 0.02; 95% CI, 0-0.05) decreases the likelihood that a UGIB requires rial. She takes no prescription medica- urgent intervention. tions other than those for hypertension Conclusions Melena, nasogastric lavage with blood or coffee grounds, or serum urea and she takes no over-the-counter nitrogen:creatinine ratio of more than 30 increase the likelihood of a UGIB. Blood clots medications. Physical examination re- in the stool make a UGIB much less likely. The Blatchford clinical prediction score, which veals blood pressure of 145/100 mm Hg, does not require nasogastric lavage, is very efficient for identifying patients who do pulse of 70/min, and rectal examina- not require urgent intervention. tion with liquid brown stool. Blood JAMA. 2012;307(10):1072-1079 www.jama.com test results reveal a hemoglobin level of 13.7 g/dL, a creatinine level of 1.1 Author Affiliations: Departments of Medicine (Drs Department of Medicine, Duke University Medical Srygley, Tran, and Fisher) and Surgery (Dr Gerardo), Center, PO Box 3913, Durham, NC 27710 (srygl001 Duke University Medical Center, Durham, North Caro- @mc.duke.edu). CME available online at lina; and Center for Health Services Research in Pri- The Rational Clinical Examination Section Editors: Da- www.jamaarchivescme.com mary Care, Durham Veterans Affairs Medical Cen- vid L. Simel, MD, MHS, Durham Veterans Affairs Medi- and questions on p 1091. ter, Durham, North Carolina (Dr Fisher). cal Center and Duke University Medical Center, Dur- Corresponding Author: F. Douglas Srygley, MD, ham, NC; Drummond Rennie, MD, Deputy Editor.

1072 JAMA, March 14, 2012—Vol 307, No. 10 ©2012 American Medical Association. All rights reserved. SEVERE UPPER GASTROINTESTINAL BLEEDING mg/dL, and a serum urea nitrogen level UGIB. Once the location of the GIB is bleeding source in which there is more of 12 mg/dL. A nasogastric lavage has determined to be from an upper source, time for enzymatic breakdown to trans- clear fluid. What is the probability that management will reflect the likeli- form blood to melena and produce the this patient is having a severe UGIB? hood of severe bleeding. We con- characteristic pungent odor that an ex- ducted a systematic review to deter- perienced practitioner can often iden- WHY ARE THESE QUESTIONS mine the accuracy of historical features, tify immediately. In clinical experi- IMPORTANT? symptoms, signs, and combinations of ments, placing as little as 50 mL of blood findings that might help assess pain the stomach can cause melena.10 Al- In the United States, an estimated tients with GIB. though gastric acid may play a role in its 390 000 hospitalizations annually oc- formation, blood inserted into the small cur with a principal diagnosis of gastro- VISUAL EVIDENCE OF GIB bowel or cecum can still create melenic intestinal bleeding (GIB) and an addi- Melena stool. Melena appears to depend primar- tional 600 000 hospitalizations occur Although patients will recognize he- ily on the length of time between instal- 1 with a secondary diagnosis of GIB. Some matemesis or hematochezia, they may lation of the blood and the patient hav- patients will have severe bleeding and re- not appreciate and therefore may not re- ing a bowel movement.11,12 For example, quire urgent intervention to reduce the port the passage of black, tarry stools that 1 L of blood placed in the stomach can potential for morbidity and mortality; characterize melena. However, not all result in bright red blood per rectum however, many patients with minor dark stools are melenic. Iron- or bismuth- within 4 hours, which implies that red bleeding can be managed effectively by containing medications can produce dark blood or clots per rectum are due to arranging an endoscopy in the outpa- stools that mimic the appearance of true either a very rapid blood loss or a distal tient setting, which can spare the pa- melena.9 Melena suggests a proximal source of bleeding.10,13 tient an urgent intervention and hospi- talization.2 Urgent interventions for the Figure 1. Anatomical Landmarks and Location of Gastrointestinal Bleeding patient with severe bleeding include endoscopic therapy, blood transfusion, radiological intervention, or surgery. Pa- tients hospitalized with a UGIB have a DIAPHRAGM mortality of 4.5% to 8.2%,3-5 and simi- Endoscope lar patients with a lower GIB (LGIB) have a mortality of 3.0% to 8.8%.4,6,7 Identification of a UGIB (proximal to Inferior vena cava the ligament of Treitz) vs LGIB (distal to the ligament of Treitz) is critical to performing an effective, efficient evalua- Celiac trunk of aorta tion and may be difficult even for an ex- perienced clinician (FIGURE 1). In 1 study,8 more than a third of patients Ligament of Treitz without hematemesis presenting with an acute GIB underwent diagnostic test- ing of both the upper and lower gastrointestinal tract. Although all pa- Duodenojejunal junction tients with GIB require assessment of hemodynamics and blood cell count, the remaining evaluations and potential interventions for a UGIB and LGIB differ considerably. For example, evaluation of a likely LGIB will include colonoscopy, flexible sigmoidoscopy, tagged red blood cell scan, or an angiogram. Conversely, if a patient has a likely Endoscope UGIB, the first diagnostic study is typi- (distal extent) cally an upper endoscopy. Addition- ally, certain medical therapies such as proton pump inhibitors and octreo- Bleeding sources proximal to the duodenojejunal junction, as approximated by the ligament of Treitz, are con- tide would only be appropriate for a sidered upper gastrointestinal bleeds.

Gastric Contents mum of 100 to 200 mL of room tem- assessment of GIB, patients being ad- Nasogastric lavage provides a method for perature water or normal saline is in- mitted to the hospital for GIB, or both). sampling contents from the stomach. Al- serted via the tube into the stomach and We defined a severe UGIB as one with though the routine use of nasogastric la- then aspirated to evaluate the color of either high-risk endoscopic stigmata be- vage in patients with suspected GIB re- return. Bright red blood indicates fresh cause intervention is recommended22 or mains controversial,14,15 it is frequently blood. Aspirate that looks like wet cof- if an intervention was performed, in- used in the United States and Canada. A fee grounds indicates blood that has cluding endoscopic therapy, blood study performed in Los Angeles revealed been partially degraded.13,15,20 Many cli- transfusion, radiological intervention, that 60% (n=632) of patients underwent nicians infer that yellow-green tinge in- or surgery. Previous studies have used nasogastric lavage for UGIB,16 and 28% dicates bile from contents beyond the similar definitions as any of the above (n=1869) of patients underwent the pro- pylorus, but visual determination of bile outcomes would necessitate an imme- cedure in a study from Canada.17 A recent may not be accurate.21 diate intervention by the clinician.2,23 survey of gastroenterologists’ opinions The search was conducted using a about the role of nasogastric lavage for METHODS similar strategy developed for the Ra- UGIB showed they were uncertain of its Search Strategy and Study tional Clinical Examination series appropriateness (quantified with the Selection (eMethods, http://www.jama.com). We RAND appropriateness scale) and also We searched MEDLINE (1966- included studies that evaluated the di- showed “extreme variation” (quantified September 2011) for articles on the re- agnostic accuracy or reliability of some with a disagreement index) in their opin- liability and diagnostic accuracy for element of the history, physical exami- ions about its role.18 Some clinicians con- components of the clinical examina- nation, bedside maneuvers (nasogas- sider the presence of varices a relative tion and routine investigations for dis- tric lavage), or routine investigations contraindication for the use of nasogas- tinguishing acute UGIB from LGIB and (defined as complete blood cell count, tric lavage. However, there are no pub- for determining the severity of a UGIB. electrolytes, creatinine level, serum urea lished trials to suggest that nasogastric Our strategy was intentionally broad to nitrogen level, prothrombin time, par- lavageworsensbleeding19 inpatientswith minimize the chance of overlooking a tial thromboplastin time, bilirubin, or without varices. relevant article. We defined acute by the transaminases, alkaline phosphatase) Once a nasogastric tube has been setting (defined as patients having pre- for determining either the presence of properly inserted (FIGURE 2), a mini- sented to an emergency department for a UGIB in a patient presenting with a

Figure 2. Key Steps in Nasogastric (NG) Tube Insertion

A Insert the NG tube along the base B Continue to advance the NG tube through C Confirm proper position of the NG tube. of the nares parallel to the ground. the oropharynx.

NG tube

Floor off the Epiglottis Epiglottis nasopharynxarynx

Trachea To ease insertion past the trachea, advance the tube while the patient Esophagus swallows sips of water from a straw, which closes the epiglottis.

Prior to insertion of the NG tube, inspect the nares to confirm the absence of any obstruction. If the nares are symmetrical, insert the tube in the side with the larger passageway. Most experts suggest applying topical lidocaine in the nares for patient comfort during insertion. A, With the patient seated and head level, insert the first 10 cm of a lubricated NG tube along the floor of the nasopharynx (parallel to the ground). B, Advance the tube through the oropharynx past the epiglottis and trachea into the esophagus and stomach. C, Confirm proper location of the NG tube by aspiration for stomach contents, auscultation over the stomach after pushing air through the tube with a syringe, or by an abdominal radiograph.

1074 JAMA, March 14, 2012—Vol 307, No. 10 ©2012 American Medical Association. All rights reserved. SEVERE UPPER GASTROINTESTINAL BLEEDING clinically apparent GIB or for deter- prospective studies of severity of UGIB Historical Factors. Patients with a mining the need for an urgent evalua- included in our review showed that 36% UGIB are much more likely to have had tion of a patient presenting with a UGIB. (95% CI, 29%-44%) of patients who a prior history of UGIB (LR, 6.2; 95% To convert creatinine level to micro- present with a UGIB require immedi- CI, 2.8-14.0). Conversely, a prior his- moles per liter, multiply by 88.4; and ate intervention (blood transfusion, ur- tory of an LGIB makes an upper source serum urea nitrogen level to milli- gent endoscopic therapy, radiological less likely (LR, 0.17; 95% CI, 0.09- moles per liter, multiply by 0.357. intervention, or surgery). 0.35). Patients with UGIB are more likely to be younger than 50 years (LR, Statistical Analysis Accuracy of Findings That 3.5; 95% CI, 2.0-6.1). Although war- For each study, we calculated the preva- Distinguish UGIB From LGIB farin use was twice as likely in pa- lence, sensitivity, specificity, likeli- The factors of the history, physical ex- tients with a UGIB rather than an LGIB hood ratios (LRs), and 95% CIs from amination, and basic laboratory test re- (LR, 2.3; 95% CI, 1.1-5.0), all other his- 2ϫ2 tables (for studies with a 0 cell sults with positive LR of 2.0 or more torical features with an LR of more than value, 0.5 was added to each cell to cal- or negative LR of 0.5 or less that dis- 2 had a CI that included 1. The use of culate the LR CI). When possible, we criminate UGIB from LGIB are shown nonsteroidal anti-inflammatory drugs, reanalyzed the reported results from in- in TABLE 1 (eTable 3 includes com- aspirin, and alcohol had LRs approach- dividual studies that allowed us to es- plete data for other findings with less ing 1 and could not differentiate be- timate the stratum-specific LR (serial useful LR and eTable 4 includes re- tween UGIB and LGIB (eTable 3). LR) for increasing levels of abnormal- sults of individual studies presented as Symptoms. A history of passing black ity from scoring systems. Findings that summary measures).8,27-33 stool (LR range, 5.1-5.9) or tarry stool were evaluated in only 2 studies are summarized with ranges and those findings in only 3 studies were summa- Table 1. Clinical Factors of the Bleeding Location for the Evaluation of UGIB From the History and Clinical Examination With Positive LR of Ն2.0 or Negative LR of Յ0.5 rized with univariate random effects Sensitivity, % Specificity, % Positive LR Negative LR measures (Comprehensive Meta- Clinical Factors (95% CI) (95% CI) (95% CI) (95% CI) analysis, version 2.2.057; Biostat). We Demographic and attempted to use bivariate random ef- historical features Prior history of 22 (18-25) 96 (94-98) 6.2 (2.8-14.0) 0.81 (0.74-0.89) fects for findings reported in 4 or more UGIB8 studies, but when results did not con- Age Ͻ50 y 8 27 (22-31) 92 (89-95) 3.5 (2.0-6.1) 0.80 (0.71-0.89) verge on a solution we used univariate Cirrhosis8 5 (3-6) 99 (97-99.4) 3.1 (0.78-12.0) 0.97 (0.93-1.00) 24,25 measures. Heterogeneity for find- Warfarin use8 12 (8-15) 95 (93-97) 2.3 (1.1-5.0) 0.93 (0.87-1.00) ings reported in 3 or more studies was Iron use8 6 (3-8) 98 (96-99) 2.2 (0.7-6.6) 0.97 (0.93-1.00) assessed with a derSimonian-Laird pro- History of LGIB8 6 (3-11) 64 (62-67) 0.17 (0.09-0.35) 1.5 (1.3-1.6) cedure using Comprehensive Meta- Symptoms analysis.26 Black stool history 77-95 81-87 5.1-5.9 0.06-0.27 (melena)8,27, a RESULTS Epigastric pain8 17 (12-21) 93 (90-95) 2.3 (1.2-4.4) 0.90 (0.82-0.98) Signs Search Results Melenic stool on 49 (45-50) 98 (91-99.6) 25 (4-174) 0.52 (0.42-0.64) A total of 2628 citations were identi- examination27 fied in our literature search. Of these, Nasogastric lavage 44 (39-48) 95 (90-98) 9.6 (4.0-23.0) 0.58 (0.49-0.70) with blood or 2516 were excluded after review of their coffee grounds28 abstracts and titles. The remaining stud- Clots in stool8 15 (14-15) 99.2 (96.0-99.9) 0.05 (0.01-0.38) 1.2 (1.1-1.2) ies (n=112) were reviewed in detail, Laboratory findings with 25 studies meeting inclusion cri- Serum urea nitrogen: 51 (26 to 75) 93 (87 to 99) 7.5 (2.8-12.0) 0.53 (0.28-0.78) creatinine ratio teria (eMethods, eFigure, eTable 1, and Ͼ308,29-33, b eTable 2). Hematocrit, %8, c Յ20 NA NA 2.6 (1.4-4.6) Prevalence of UGIB Hemorrhage 21-29 NA NA 1.9 (1.4-2.5) A summary prevalence of 8 stud- 30-39 NA NA 0.46 (0.32-0.65) ies8,27-33 in our review that differenti- Ն40 NA NA 0.26 (0.10-0.67) Abbreviations: LGIB, lower gastrointestinal bleeding; LR, likelihood ratio; NA, not applicable; UGIB, upper gastrointestinal ated UGIB from LGIB found that 63% bleeding. (95% CI, 51%-73%) of patients pre- aSummary estimate provided as a range because the finding was evaluated in only 2 studies. bSummary estimates calculated with bivariate random effects. For serum urea nitrogen:creatinine ratio of more than 30, senting with gastrointestinal hemor- the positive LR test for homogeneity, I2=17%; P=.31; and the negative LR test for homogeneity, I 2=94%; PϽ.001. rhage have an upper gastrointestinal cSerial LR (95% CI) is a multilevel LR that can be used to give LR for each different threshold as opposed to choosing a single cutoff and then giving a dichotomous LR. source. A summary prevalence from 8

©2012 American Medical Association. All rights reserved. JAMA, March 14, 2012—Vol 307, No. 10 1075 SEVERE UPPER GASTROINTESTINAL BLEEDING makes a UGIB much more likely. Epi- UGIB less likely (LR, 0.58; 95% CI, (LR, 2.6; 95% CI, 1.4-4.6), but those gastric discomfort also favors a UGIB 0.49-0.70). with no anemia are less likely to have (LR, 2.3; 95% CI, 1.2-4.4). Laboratory Findings. An increased a UGIB (hematocrit Ն40%; LR, 0.26; Signs. The physician’s examination serum urea nitrogen:creatinine ratio sug- 95% CI, 0.10-0.67). finding of melena (either observed as gests a UGIB. Although several differ- a passed stool or stool obtained via recent thresholds have been evaluated, com- Accuracy of Factors to Identify tal examination) was the most useful bining all results in which the serum urea Need for Urgent Evaluation of UGIB finding for identifying a UGIB (LR, 25; nitrogen:creatinine ratio is more than 30 In patients presenting with symptoms 95% CI, 4-174). In patients without a results in a homogeneous diagnostic odds of UGIB, the factors obtained from the history of hematemesis, nasogastric la- ratio (I2=17%, P=.31), with a summary history, physical examination, and ba- vage with clear evidence of blood in the LR of 7.5 (95% CI, 2.8-12.0). Serum urea sic laboratory test results predictive of aspirate (blood or coffee grounds nitrogen:creatinine ratios of 30 or less de- a severe UGIB with a positive LR of 2.0 grossly present) suggests a UGIB with crease the odds of a UGIB with a sum- or more or a negative LR of 0.5 or less an LR of 9.6 (95% CI, 4.0-23.0). mary LR of 0.53 (95% CI, 0.28-0.78), al- are shown in TABLE 2 (eTable 5 in- The presence of clots in the stool though there is heterogeneity in the cludes complete data for other find- makes a UGIB much less likely (LR, results (I2=94%, PϽ.001). ings with less useful LR and eTable 6 0.05; 95% CI, 0.01-0.38). Nasogas- Increasing severity of anemia in- includes results of individual studies tric lavage without evidence of blood creases the likelihood of UGIB. Pa- presented as summary measures) and in the aspirate (no blood or coffee tients with severe anemia (hematocrit TABLE 3. Single-study clinical predic- grounds grossly present) makes a Յ20%) are more likely to have a UGIB tion rules are shown in eTable 7. Historical Factors, Symptoms, and Signs. Once a patient is identified as hav- Table 2. Clinical Factors of Severity of Bleeding From the History and Clinical Examination Used to Determine the Need for Urgent Evaluation of UGIB With Positive LR of Ն2.0 or ing a UGIB, patients with history of cir- Negative LR of Յ0.5a rhosisormalignancy(LR,3.7;95%CI,1.6- Sensitivity, Specificity, Positive LR Negative LR 8.8), syncope (LR, 3.0; 95% CI, 1.7-5.4), Clinical Factors % (95% CI) % (95% CI) (95% CI) (95% CI) or those using analgesics (LR, 2.6; 95% Demographic and CI, 1.3-5.2) are more likely to have severe historical features History of malignancy 22 (14-28) 94 (92-96) 3.7 (1.6-8.8) 0.83 (0.72-0.97) bleeding.Theimportanceofanalgesicuse or cirrhosis34 requires confirmation because the use of Cirrhosis35 15 (12-18) 95 (94-97) 3.2 (2.1-4.9) 0.89 (0.85-0.94) nonsteroidalanti-inflammatorydrugs(LR, Syncope35 8 (6-10) 98 (97-98) 3.0 (1.7-5.4) 0.95 (0.91-0.98) 1.8; 95% CI, 1.2-2.6) was less important Analgesic use36 13 (8-19) 95 (94-96) 2.6 (1.3-5.2) 0.92 (0.84-0.99) for identifying patients with severe bleed- Coffee ground 7 (4-10) 83 (82-84) 0.41 (0.26-0.64) 1.1 (1.1-1.2) 35 ing (eTable 5). vomiting The clinical use of the nasogastric la- Hematochezia35 2 (1-4) 92 (91-93) 0.22 (0.09-0.53) 1.1 (1.0-1.1) vage depends on the color of the re- Signs Pulse rate 71 (60-79) 86 (82-89) 4.9 (3.2-7.6) 0.34 (0.22-0.53) turned fluid. The highest likelihood of Ͼ100/min34 a severe UGIB is with an aspirate of red Nasogastric lavage, red 77 (57-90) 76 (32-95) 3.1 (1.2-14.0) 0.32 (0.17-0.57) blood (summary LR, 3.1; 95% CI, 1.2- blood16,17,21,34, b 17,34, c,d 14.0). A similar result was found in Shock 78 (56-90) 71 (46-88) 2.8 (1.1-7.2) 0.32 (0.10-0.96) studies that considered either red blood Nasogastric lavage, red 81 (67-89) 55 (19-87) 2.0 (1.0-4.0) 0.40 (0.20-0.81) blood or coffee or coffee grounds as positive for UGIB grounds17,21,28, c (summary LR, 2.0; 95% CI, 1.0-4.0). A Hypotension34,36, e 55-59 53-89 1.2-4.8 0.51-0.78 nasogastric lavage without bright red Laboratory findings blood (summary LR, 0.32; 95% CI, Hemoglobin level 65-68 86-89 4.5-6.2 0.36-0.41 Ͻ8 g/dL34, e 0.17-0.57) makes severe bleeding less Serum urea nitrogen level 63 (52-72) 83 (79-86) 3.6 (2.4-5.5) 0.45 (0.31-0.65) likely. As opposed to visible blood, the Ͼ90 mg/dL34 presence of occult blood is not diag- White blood cell count 61 (50-71) 82 (78-86) 3.4 (2.2-5.1) 0.48 (0.34-0.68) nostic of UGIB needing urgent inter- Ͼ 9 34 12 x10 /L vention, and the absence of occult blood Abbreviations: LR, likelihood ratio; UGIB, upper gastrointestinal bleeding. SI conversion: To convert serum urea nitrogen level to mmol/L, multiply by 0.357. has a broad CI and thus its use is un- aSummary ranges and LR are presented for factors evaluated in multiple studies. bHeterogeneity for bivariate random effects (positive LR, I 2=90%; PϽ.001; negative LR, I 2=81%; PϽ.001). certain (eTable 5). cHeterogeneity for univariate random effects (shock: positive LR, I 2=96%; PϽ.001; negative LR, I 2=91%; PϽ.001; red Hemodynamic signs associated with blood or coffee grounds: positive LR, I 2=95%; PϽ.001; negative LR, I 2=67%; P=.047). dSystolic blood pressure of less than 100 mm Hg or pulse of more than 100/min or orthostatic decrease in systolic blood volume loss, such as tachycardia (LR, 4.9; pressure by more than 10% or orthostatic increase in pulse of more than 10% between sitting and supine position. 95% CI, 3.2-7.6), shock (summary LR, eRange given because finding evaluated in only 2 studies. 2.8; 95% CI, 1.1-7.2), and hypotension

(LR range, 1.2-4.8), may be helpful for Table 3. Well-Validated Clinical Prediction Rules Used to Determine the Need for Urgent identifying patients with severe UGIB, Evaluation of UGIB but have broad CIs. The absence of tachy- Specificity, cardia (LR, 0.34; 95% CI, 0.22-0.53) was Sensitivity, % % Positive LR Negative LR the most useful sign for decreasing the Clinical Score Threshold (95% CI) (95% CI) (95% CI) (95% CI) Blatchford Յ2a 98 (92-99) 27 (11-53) 1.4 (1.1-1.8) 0.08 (0.01-0.41) likelihood of severe UGIB. score37-39 Laboratory Findings. A hemoglobin Blatchford 0b 99.6 (99.0-100.0) 15 (5-25) 1.2 (1.0-1.3) 0.02 (0-0.05) level of less than 8 g/dL (LR range, 4.5- score2,23,35,37-41 6.2), a serum urea nitrogen level of more Preendoscopic 0c 91 (88-95) 21 (9-34) 1.20 (0.97-1.30) 0.41 (0.12-0.70) Rockall than90mg/dL(LR,3.6;95%CI,2.4-5.5), score2,38,40-42 or a white blood cell count of more than Modified Յ1 95 (93-96) 12 (11-13) 1.1 (1.0-1.1) 0.45 (0.31-0.66) 12ϫ109/L (LR, 3.4; 95% CI, 2.2-5.1) in- Blatchford crease the likelihood of severe UGIB. A score43 Abbreviations: LR, likelihood ratio; UGIB, upper gastrointestinal bleeding. hemoglobin level of 8 g/dL or higher (LR aHeterogeneity for univariate random effects (positive LR, I 2=86%; PϽ.001; negative LR, I 2=45%; P=.16). range, 0.36-0.41), a serum urea nitro- bBecause sensitivity was 99% to 100% in every study, sensitivity was treated as a fixed effect. Heterogeneity (positive LR, I 2=97%; PϽ.001; negative LR, I 2=0%; P=.59). gen level of 90 mg/dL or less (LR, 0.45; cHeterogeneity (positive LR, I 2=92%; PϽ.001; negative LR, I 2=70%; PϽ.001). 95% CI, 0.31-0.65), and a white blood cell count of 12ϫ109/L or less (LR, 0.48;

95% CI, 0.34-0.68) decrease the likeli- location of a GIB is problematic. Often, 23 hood of severe UGIB. a patient may have either no definitive Table 4. The Blatchford Score Clinical Prediction Models. Com- source or multiple possible sources of a Variable Score Serum urea nitrogen, mg/dL binations of findings have been evalu- GIB. For example, a recent study of LGIB Ͻ18.2 0 ated to better guide clinical decision found that in patients who received stan- 18.2-Ͻ22.4 2 making (TABLE 4) (eTable 8, eTable 9, dard care, only 22% had a definitive 22.4-Ͻ28.0 3 eTable 10, and eTable 11 include de- source of bleeding and only 76% had 28.0-Ͻ70.0 4 tailed descriptions of the algorithms). either a definitive or a presumptive Ն70.0 6 23 44 The Blatchford score (Table 4) has the source of bleeding. Thus, all studies Hemoglobin for men, g/dL most extensive validation and the best identifying factors associated with source Ͼ13.0 0 accuracy for identifying patients who of bleeding may be limited by ascertain- 12.0-Ͻ13.0 1 present with UGIB who do not need ur- ment bias. Other limitations are due to 10.0-Ͻ12.0 3 gent intervention (Table 3). A Blatch- the paucity of published investigations Ͻ10.0 6 ford score of 0 occurs in up to 22% of for certain predictive factors. For ex- Hemoglobin for women, g/dL Ͼ12.0 0 patients with UGIB2,23 and identifies pa- ample, few data exist regarding the as- 10.0-Ͻ12.0 1 tients with a low likelihood of severe sociation of hematemesis and GIB loca- Ͻ10.0 6 bleeding (summary LR, 0.02; 95% CI, tion because presumably all patients with Systolic blood pressure, mm Hg 0-0.05). Blatchford scores at a thresh- hematemesis have a UGIB. In accor- Ͼ109 0 old of 2 or less perform similarly, al- dance with the format and focus of the 100-109 1 though the CI is broader (summary LR, Rational Clinical Examination series, our 90-99 2 0.08; 95% CI, 0.01-0.41). The compo- review was designed to systematically ex- Ͻ90 3 nents of the full Rockall score that can amine the clinical factors that may as- Pulse rate Ͼ100/min 1 be obtained before endoscopy pro- sist only in the diagnosis of GIB. Our goal Presentation with melena 1 duce the preendoscopic Rockall score. was to identify both the location of bleed- Presentation with syncope 2 A preendoscopic Rockall score of more ing (ie, upper vs lower intestinal tract) Hepatic disease 2 than 0 does not increase the likeli- and in those patients with UGIB the se- Cardiac failure 2 hood of severe bleeding (summary LR, verity of bleeding as defined by those SI conversion: To convert serum urea nitrogen to mmol/L, multiply by 0.357. 1.20; 95% CI, 0.97-1.30), but a score likely to require an intervention. The re- equal to 0 decreases the likelihood of view does not, however, discuss the role severe bleeding (summary LR, 0.41; and data for specific interventions to treat changing practice patterns over time. 95% CI, 0.12-0.70) (Table 3). patients with UGIB. Both the diagnostic process and likely Heterogeneity in results may come even underlying pathophysiology (in COMMENT from differences in study populations the proton pump inhibitor era) have Our review and meta-analysis have limi- (eg, severity of comorbid illnesses), lo- changed dramatically in the last 25 tations inherent in the particular study cal practice patterns (eg, admission rates years, leading to differences in results questions. The reference standard for a for patients with UGIB), prevalence of that may manifest as wide CIs for sum- study that seeks to establish a definitive different etiologies of UGIB, and even mary LRs. Heterogeneity may also arise

©2012 American Medical Association. All rights reserved. JAMA, March 14, 2012—Vol 307, No. 10 1077 SEVERE UPPER GASTROINTESTINAL BLEEDING from differences in case definitions for in” severe bleeding. Much more fre- physiological processes. Inspection of bleeding source or severity. Despite quently, the physician is attempting to the stool is the most important factor these potential differences, we chose to “rule out” a bleed requiring urgent in- for identifying the bleeding source. Me- combine studies whenever possible to tervention. Unfortunately, a negative na- lenic stool on physical examination (LR, provide an overall assessment of the sogastric lavage has an LR that provides 25) provides compelling evidence for usefulness of each clinical feature or little or no assistance in “ruling out” se- a UGIB, although red blood and clots predictive rule, while also providing ac- vere bleeding and is unlikely to change are unlikely to come from a UGIB (LR, companying CIs to allow the appropri- the clinical determination for urgent en- 0.05). Nasogastric lavage for blood or ate interpretation of the complete data. doscopy. coffee grounds (LR, 9.6), a history of a Nasogastric aspiration has advan- prior UGIB (LR, 6.2), and a serum urea tages as a diagnostic bedside maneuver SCENARIO RESOLUTION nitrogen:creatinine ratio of more than to evaluate GIB due to its availability, low Case 1 30 (LR, 7.5) are other findings that sug- cost, and very low risk of complica- This older man has multiple factors in- gest a UGIB. tions.45 Additional benefits include re- creasing the likelihood for an upper Tachycardia (pulse rate of Ͼ100/ moval of excess fluid, blood, and clots, source of blood loss with a history of min; LR, 4.9), a history of cirrhosis or which may improve visualization and prior UGIB (LR, 6.2), melena (LR range, malignancy (LR, 3.7), hemoglobin level decrease the risk of aspiration if endos- 5.1-5.9), low hemoglobin level (LR, of less than 8 g/dL (LR range, 4.5-6.2), copy is performed.46 Generally patients 2.6), and an increased serum urea ni- or a nasogastric lavage with red blood with a positive result go on to upper en- trogen:creatinine ratio (LR, 7.5). With (LR, 3.1) increase the likelihood of se- doscopy. However, patients with nega- a pretest probability of 63% for a UGIB, vere bleeding. All patients with a UGIB tive results may still require upper en- these findings increase the probability should have a Blatchford score, which doscopy instead of an examination for of a UGIB to at least 82%. Likewise, this does not require a nasogastric lavage, LGIB because the negative LR is insuf- patient has several factors that in- to help assess the severity (Blatchford ficient to rule out a UGIB. Therefore, its crease the likelihood of a severe UGIB score=0; LR, 0.02 for identifying pa- role as a diagnostic test to differentiate with signs of tachycardia (LR, 4.9), low tients requiring urgent evaluation). UGIB and LGIB should be ques- hemoglobin level (LR range, 4.5-6.2), When negative, prediction rules tioned.14,15 Furthermore, nasogastric in- and a Blatchford score of more than 1 combining symptoms, signs, and rou- sertion is among the most uncomfort- (LR, 1.4). Based only on these factors, tine laboratory test results almost able bedside procedures in the this patient should receive an urgent up- definitively rule out severe UGIB, emergency department and patients rate per endoscopy either in the emer- thereby identifying at least some pa- its discomfort comparable with frac- gency department or as an inpatient, tients who can be safely evaluated as an ture reduction or abscess drainage.47 and a diagnostic nasogastric lavage is outpatient. Often the clinical challenge is not to unnecessary to differentiate a UGIB Author Contributions: Drs Srygley and Fisher had full identify the source of bleeding, but rather from an LGIB. access to all of the data in the study and take respon- to assess the severity of a suspected UGIB. sibility for the integrity of the data and the accuracy of the data analysis. Although clinical situations such as sus- Case 2 Study concept and design: Srygley, Fisher. picion of variceal bleeding in a patient This woman has multiple factors that de- Acquisition of data: Srygley, Gerardo, Tran, Fisher. Analysis and interpretation of data: Srygley, Gerardo, with cirrhosis or of an aortoenteric fis- crease the likelihood of a severe UGIB. Fisher. tula require urgent evaluation,48,49 most She has no tachycardia (LR, 0.34), a nor- Drafting of the manuscript: Srygley. situations are not as clear. Some gastro- Critical revision of the manuscript for important in- mal hemoglobin level (range LR, 0.36- tellectual content: Srygley, Gerardo, Tran, Fisher. enterologists use the presence of bright 0.41), a clear nasogastric lavage (LR, Statistical analysis: Srygley. red blood on nasogastric lavage as a cri- 0.40), and a Blatchford score of 0 (LR, Administrative, technical, or material support: Gerardo, Tran. terion for performing endoscopy within 0.02). Based on a pretest probability of Study supervision: Gerardo, Fisher. 6 to 12 hours, although there is no clear 30% for a severe UGIB among all pa- Conflict of Interest Disclosures: All authors have com- pleted and submitted the ICMJE Form for Disclosure evidence that rapid endoscopy pro- tients with gastrointestinal hemor- of Potential Conflicts of Interest. Dr Fisher reported vides clinical benefit in nonvariceal rhage, the best individual finding low- receiving consultancy money from Epigenomics Inc. 50,51 No other authors reported any financial disclosures. bleeding and consensus guidelines ers the likelihood of a severe hemorrhage Funding/Support: Dr Fisher is supported in part by a only recommend that endoscopy be per- to 13% or less. However, the combina- Veterans Affairs Health Services Research and Devel- formed within 24 hours in patients with tion of findings from the Blatchford score opment Career Development Transition Award (RCD 52 03-174). UGIB. Although a nasogastric lavage make a severe GIB unlikely (probabil- Role of the Sponsor: The Veterans Affairs had no role with a bloody result has a significant LR ity Ͻ1%) in this patient. in the design and conduct of the study, in the collec- tion, analysis, and interpretation of the data, or in the that can further help “rule in” a UGIB re- preparation, review, or approval of the manuscript. quiring urgent endoscopy, seldom is a BOTTOM LINE Online-Only Material: The eMethods, eFigure, and eTables 1 through 11 are available at http://www clinician encountered with a situation in Gastrointestinal hemorrhage is the com- .jama.com. which nasogastric lavage is used to “rule mon pathway of myriad different patho- Additional Contributions: Cara L. O’Brien, MD (Duke

University), Daniel Nishijima, MD (UC Davis Medical 17. Aljebreen AM, Fallone CA, Barkun AN. Nasogas- endoscopic Rockall score in predicting the need for en- Center), and Russell Yang, MD, PhD (Southern Illi- tric aspirate predicts high-risk endoscopic lesions in pa- doscopic therapy in patients with upper GI hemorrhage. nois University School of Medicine), provided advice tients with acute upper-GI bleeding. Gastrointest Gastrointest Endosc. 2010;71(7):1134-1140. on earlier versions of the manuscript. Drs O’Brien, Nishi- Endosc. 2004;59(2):172-178. 36. Sto¨ ltzing H, Ohmann C, Krick M, Thon K. Diag- jima, and Yang received no compensation for their 18. Esrailian E, Gralnek IM, Jensen D, et al. Evaluat- nostic emergency endoscopy in upper gastrointesti- contribution. ing the process of care in nonvariceal upper gastro- nal bleeding: do we have any decision aids for pa- intestinal haemorrhage: a survey of expert vs. non- tient selection? Hepatogastroenterology. 1991; expert gastroenterologists. Aliment Pharmacol Ther. 38(3):224-227. REFERENCES 2008;28(10):1199-1208. 37. Masaoka T, Suzuki H, Hori S, Aikawa N, Hibi T. 19. Thomsen TW, Shaffer RW, Setnik GS. Videos in Blatchford scoring system is a useful scoring system 1. Healthcare Cost and Utilization Project (HCUP). clinical medicine: nasogastric intubation. N Engl J Med. for detecting patients with upper gastrointestinal bleed- HCUP Clinical Classifications Software (CCS) for 2006;354(17):e16. ing who do not need endoscopic intervention. J Gas- ICD-9-CM. http://www.hcup-us.ahrq.gov 20. Best C. Nasogastric tube insertion in adults who troenterol Hepatol. 2007;22(9):1404-1408. /toolssoftware/ccs/ccs.jsp. Accessed December 11, require enteral feeding. Nurs Stand. 2007;21(40): 38. Srirajaskanthan R, Conn R, Bulwer C, Irving P. The 2011. 39-43. Glasgow Blatchford scoring system enables accurate 2. Stanley AJ, Ashley D, Dalton HR, et al. Outpatient 21. Cuellar RE, Gavaler JS, Alexander JA, et al. Gas- risk stratification of patients with upper gastrointes- management of patients with low-risk upper- trointestinal tract hemorrhage: the value of a naso- tinal haemorrhage. Int J Clin Pract. 2010;64(7): gastrointestinal haemorrhage: multicentre validation gastric aspirate. Arch Intern Med. 1990;150(7): 868-874. and prospective evaluation. Lancet. 2009;373(9657): 1381-1384. 39. Chandra S, Hess EP, Agarwal D, et al. External vali- 42-47. 22. Barkun AN, Bardou M, Kuipers EJ, et al; Interna- dation of the Glasgow-Blatchford Bleeding Score and 3. Marmo R, Koch M, Cipolletta L, et al. Predictive tional Consensus Upper Gastrointestinal Bleeding Con- the Rockall Score in the US setting. Am J Emerg Med. factors of mortality from nonvariceal upper gastroin- ference Group. International consensus recommen- 2011. testinal hemorrhage: a multicenter study. Am J dations on the management of patients with 40. Chen IC, Hung MS, Chiu TF, Chen JC, Hsiao CT. Gastroenterol. 2008;103(7):1639-1647. nonvariceal upper gastrointestinal bleeding. Ann In- Risk scoring systems to predict need for clinical inter- 4. Lanas A, Garcı´a-Rodrı´guezLA, Polo-Toma´ s M, et al. tern Med. 2010;152(2):101-113. vention for patients with nonvariceal upper gastroin- Time trends and impact of upper and lower gastro- 23. Blatchford O, Murray WR, Blatchford M. A risk testinal tract bleeding. Am J Emerg Med. 2007; intestinal bleeding and perforation in clinical practice. score to predict need for treatment for upper- 25(7):774-779. Am J Gastroenterol. 2009;104(7):1633-1641. gastrointestinal haemorrhage. Lancet. 2000;356 41. Farooq FT, Lee MH, Das A, Dixit R, Wong RC. 5. Blatchford O, Davidson LA, Murray WR, Blatchford (9238):1318-1321. Clinical triage decision vs risk scores in predicting the M, Pell J. Acute upper gastrointestinal haemorrhage 24. Simel DL, Bossuyt PM. Differences between uni- need for endotherapy in upper gastrointestinal in west of Scotland: case ascertainment study. BMJ. variate and bivariate models for summarizing diag- bleeding. Am J Emerg Med. 2010;30(1):129-134. 1997;315(7107):510-514. nostic accuracy may not be large. J Clin Epidemiol. 42. Das A, Ben-Menachem T, Farooq FT, et al. Arti- 6. Das A, Ben-Menachem T, Cooper GS, et al. Pre- 2009;62(12):1292-1300. ficial neural network as a predictive instrument in pa- diction of outcome in acute lower-gastrointestinal 25. Menke J. Bivariate random-effects meta- tients with acute nonvariceal upper gastrointestinal haemorrhage based on an artificial neural network: analysis of sensitivity and specificity with SAS PROC hemorrhage. Gastroenterology. 2008;134(1):65- internal and external validation of a predictive model. GLIMMIX. Methods Inf Med. 2010;49(1):54-62. 74. Lancet. 2003;362(9392):1261-1266. 26. Borenstein M, Hedges L, Higgins J, Rothstein H. 43. Romagnuolo J, Barkun AN, Enns R, Armstrong 7. Schmulewitz N, Fisher DA, Rockey DC. Early colo- Introduction to Meta-Analysis. Chichester, England: D, Gregor J. Simple clinical predictors may obviate ur- noscopy for acute lower GI bleeding predicts shorter John Wiley & Sons; 2009. gent endoscopy in selected patients with nonvariceal hospital stay: a retrospective study of experience in a 27. Zuckerman GR, Trellis DR, Sherman TM, Clouse upper gastrointestinal tract bleeding. Arch Intern Med. single center. Gastrointest Endosc. 2003;58(6): RE. An objective measure of stool color for differen- 2007;167(3):265-270. 841-846. tiating upper from lower gastrointestinal bleeding. Dig 44. Green BT, Rockey DC, Portwood G, et al. Ur- 8. Witting MD, Magder L, Heins AE, Mattu A, Granja Dis Sci. 1995;40(8):1614-1621. gent colonoscopy for evaluation and management of CA, Baumgarten M. ED predictors of upper gastro- 28. Witting MD, Magder L, Heins AE, Mattu A, Granja acute lower gastrointestinal hemorrhage: a random- intestinal tract bleeding in patients without CA, Baumgarten M. Usefulness and validity of diag- ized controlled trial. Am J Gastroenterol. 2005; hematemesis. Am J Emerg Med. 2006;24(3):280- nostic nasogastric aspiration in patients without 100(11):2395-2402. 285. hematemesis. Ann Emerg Med. 2004;43(4):525- 45. Pillai JB, Vegas A, Brister S. Thoracic complica- 9. LeBlond RF, DeGowin RL, Brown DD. Chapter 9: 532. tions of nasogastric tube: review of safe practice. In- the abdomen, perineum, anus, and rectosigmoid. In: 29. Mortensen PB, Nøhr M, Møller-Petersen JF, Balslev teract Cardiovasc Thorac Surg. 2005;4(5):429- LeBlond RF, DeGowin RL, Brown DD, eds. DeGow- I. The diagnostic value of serum urea/creatinine ratio 433. in’s Diagnostic Examination. 9th ed. New York, NY: in distinguishing between upper and lower gastroin- 46. Lee SD, Kearney DJ. A randomized controlled trial McGraw-Hill; 2009. testinal bleeding: a prospective study. Dan Med Bull. of gastric lavage prior to endoscopy for acute upper 10. Schiff L, Stevens R, Shaprio N, Goodman S. Ob- 1994;41(2):237-240. gastrointestinal bleeding. J Clin Gastroenterol. 2004; servations on the oral administration of citrated blood 30. Snook JA, Holdstock GE, Bamforth J. Value of a 38(10):861-865. in man. Am J Med Sci. 1942;203:409-412. simple biochemical ratio in distinguishing upper and 47. Singer AJ, Richman PB, Kowalska A, Thode HC Jr. 11. Hilsman JH. The color of feces following the in- lower sites of gastrointestinal haemorrhage. Lancet. Comparison of patient and practitioner assessments stillation of citrated blood at various levels of the small 1986;1(8489):1064-1065. of pain from commonly performed emergency de- intestine. J Med Assoc Ga. 1950;39(10):402-405. 31. Chalasani N, Clark WS, Wilcox CM. Blood urea partment procedures. Ann Emerg Med. 1999;33 12. Luke RG, Lees W, Rudick J. Appearances of the nitrogen to creatinine concentration in gastrointesti- (6):652-658. stools after the introduction of blood into the caecum. nal bleeding: a reappraisal. Am J Gastroenterol. 1997; 48. Garcia-Tsao G, Sanyal AJ, Grace ND, et al. Pre- Gut. 1964;5:77-79. 92(10):1796-1799. vention and management of gastroesophageal vari- 13. Feldman M, Friedman LS, Brandt LJ, eds. Sleis- 32. Olsen LH, Andreassen KH. Stools containing al- ces and variceal hemorrhage in cirrhosis. Hepatology. inger and Fordtran’s Gastrointestinal and Liver Disease. tered blood-plasma urea: creatinine ratio as a simple 2007;46(3):922-938. Philadelphia, PA: Saunders; 2009:211-242. test for the source of bleeding. Br J Surg. 1991; 49. Champion MC, Sullivan SN, Coles JC, Goldbach 14. Pallin DJ, Saltzman JR. Is nasogastric tube lavage 78(1):71-73. M, Watson WC. Aortoenteric fistula: incidence, pre- in patients with acute upper GI bleeding indicated or 33. Richards RJ, Donica MB, Grayer D. Can the blood sentation recognition, and management. Ann Surg. antiquated? Gastrointest Endosc. 2011;74(5):981- urea nitrogen/creatinine ratio distinguish upper from 1982;195(3):314-317. 984. lower gastrointestinal bleeding? J Clin Gastroenterol. 50. Targownik LE, Murthy S, Keyvani L, Leeson S. The 15. Palamidessi N, Sinert R, Falzon L, Zehtabchi S. Na- 1990;12(5):500-504. role of rapid endoscopy for high-risk patients with acute sogastric aspiration and lavage in emergency depart- 34. Adamopoulos AB, Baibas NM, Efstathiou SP, et al. nonvariceal upper gastrointestinal bleeding. Can J ment patients with hematochezia or melena without Differentiation between patients with acute upper gas- Gastroenterol. 2007;21(7):425-429. hematemesis. Acad Emerg Med. 2010;17(2):126- trointestinal bleeding who need early urgent upper gas- 51. Spiegel BM. Endoscopy for acute upper GI tract 132. trointestinal endoscopy and those who do not: a pro- hemorrhage: sooner is better. Gastrointest Endosc. 16. Huang ES, Karsan S, Kanwal F, Singh I, Makhani spective study. Eur J Gastroenterol Hepatol. 2003; 2009;70(2):236-239. M, Spiegel BM. Impact of nasogastric lavage on out- 15(4):381-387. 52. Sung JJ, Chan FK, Chen M, et al. Asia-Pacific Work- comes in acute GI bleeding. Gastrointest Endosc. 2011; 35. Pang SH, Ching JY, Lau JY, Sung JJ, Graham DY, ing Group consensus on non-variceal upper gastro- 74(5):971-980. Chan FK. Comparing the Blatchford and pre- intestinal bleeding. Gut. 2011;60(9):1170-1177.