Clinical Gastroenterology and Hepatology 2015;13:115–121
Performance of New Thresholds of the GlasgowBlatchford Score in Managing Patients With UpperGastrointestinal Bleeding
Stig B. Laursen,* Harry R. Dalton,‡ Iain A. Murray,§ Nick Michell,‡ Matt R. Johnston,k
Michael Schultz,¶ Jane M. Hansen,* Ove B. Schaffalitzky de Muckadell,* Oliver Blatchford,#
and Adrian J. Stanley,** on behalf of the Upper Gastrointestinal Hemorrhage InternationalConsortium
*Department of Medical Gastroenterology, Odense University Hospital, Odense, Denmark; ‡Gastrointestinal Unit, RoyalCornwall Hospital, Cornwall, United Kingdom; §Gastrointestinal Unit, Dunedin Public Hospital, Dunedin, New Zealand;kDunedin School of Medicine, ¶Department of Medicine, University of Otago, Dunedin, New Zealand; #Public HealthDepartment, University of Glasgow, Glasgow, United Kingdom; and **Department of Gastroenterology, Glasgow RoyalInfirmary, Glasgow, United Kingdom
BACKGROUND & AIMS:
Upper gastrointestinal hemorrhage (UGIH) is a common cause of hospital admission. TheGlasgow Blatchford score (GBS) is an accurate determinant of patients’ risk for hospital-basedintervention or death. Patients with a GBS of 0 are at low risk for poor outcome and could bemanaged as outpatients. Some investigators therefore have proposed extending the definitionof low-risk patients by using a higher GBS cut-off value, possibly with an age adjustment. Wecompared 3 thresholds of the GBS and 2 age-adjusted modifications to identify the optimal cut-off value or modification.
METHODS:
We performed an observational study of 2305 consecutive patients presenting with UGIH at 4centers (Scotland, England, Denmark, and New Zealand). The performance of each thresholdand modification was evaluated based on sensitivity and specificity analyses, the proportion oflow-risk patients identified, and outcomes of patients classified as low risk.
RESULTS:
There were differences in age (P [ .0001), need for intervention (P < .0001), mortality (P <.015), and GBS (P[ .0001) among sites. All systems identified low-risk patients with high levelsof sensitivity (>97%). The GBS at cut-off values of £1 and £2, and both modifications, identifiedlow-risk patients with higher levels of specificity (40%–49%) than the GBS with a cut-off valueof 0 (22% specificity; P < .001). The GBS at a cut-off value of £2 had the highest specificity, but3% of patients classified as low-risk patients had adverse outcomes. All GBS cut-off values, andscore modifications, had low levels of specificity when tested in New Zealand (2.5%–11%).
CONCLUSIONS:
A GBS cut-off value of £1 and both GBS modifications identify almost twice as many low-riskpatients with UGIH as a GBS at a cut-off value of 0. Implementing a protocol for outpatientmanagement, based on one of these scores, could reduce hospital admissions by 15% to 20%.
Upper gastrointestinal hemorrhage (UGIH) remainsa common cause of admission to the hospital in
most countries. Several risk-scoring systems have beendeveloped for the assessment of patients presentingwith this condition.1–7 The Glasgow Blatchford score(GBS) has been shown to be accurate in identifying pa-tients’ risk of requiring hospital-based intervention(blood transfusion, endoscopic treatment, or surgery),or death.2,8–11 Several studies have found that patientswith a GBS of 0 have very low risk (<1%) of these out-comes.2,8–10,12–14 When using a cut-off value of 0, the GBS
is capable of identifying 5% to 22% of patients withUGIH who are at low risk of requiring hospital-basedintervention or death, and who potentially can bemanaged as outpatients.8,10,12,14 Utilization of the GBS
Hemoglobin level for men, g/L120–129 1100–119 3<100 6
Hemoglobin level for women, g/L100–119 1<100 6
Systolic blood pressure, mm Hg100–109 190–99 2<90 3
Other markersPulse �100/min 1Presentation with melena 1Presentation with syncope 2Hepatic diseasea 2Cardiac failureb 2
aKnown history, or clinical and laboratory evidence, of chronic or acute liverdisease.bKnown history, or clinical and echocardiographic evidence, of cardiac failure.
116 Laursen et al Clinical Gastroenterology and Hepatology Vol. 13, No. 1
with a cut-off value of 0 in the emergency departmenthas been shown to safely reduce admissions amonglow-risk patients with this condition.8 Some investigatorshave suggested that the proportion of identified low-riskpatients could be increased by using a higher GBS cut-offvalue,9,12,13 or modifying the GBS with an adjustment forage.10,15 However, using a higher cut-off value oradjusted GBS also may lead to reduced sensitivity, result-ing in an increased risk of misclassification of patientswho actually require intervention, or die.
The aim of this study was to compare 3 thresholds ofGBS and 2 score modifications to identify the optimalcut-off value or modification using an internationalmulticenter analysis. We also examined if the perfor-mance of the GBS depended on the location of the patientcohort.
Methods
Data Collection
Data were collected prospectively from consecutivepatients presenting with UGIH for 24 months at boththe Royal Cornwall Hospital (Truro, England) and theOdense University Hospital (Odense, Denmark), pro-spectively for 18 months at Glasgow Royal Infirmary(Glasgow, Scotland), and retrospectively for 58 monthsat Dunedin Hospital (Dunedin, New Zealand). UGIH wasdefined as a history of hematemesis, coffee-groundvomit, or melena. Patients experiencing UGIH whilealready inpatients for another reason were excluded.
Data were collected by junior doctors (non-specialists),medical students, or research nurses, depending on thesite. Patient characteristics; GBS; endoscopic findings (ifperformed); interventions in the form of blood trans-fusion, endoscopic treatment, surgery, or arterial embo-lization; and mortality were recorded at all sites. Alladmitted patients were followed up until discharge, butall patients who were not admitted for any reason werefollowed up for a minimum of 30 days to assess outcome.
Some of the data from Glasgow, Truro, and Odensehave been reported in earlier studies.8,10,16 The localhospitals recognized this as a service evaluation, ratherthan research, because no randomization or allocationto an intervention was performed, therefore ethicalapproval was unnecessary. The study was conductedaccording to the Declaration of Helsinki.17 Our reportfollows the Strengthening the Reporting of ObservationalStudies in Epidemiology (STROBE) guidelines.
Evaluated Scoring Systems
The GBS was calculated according to the criteriastated in the original article (Table 1).2 The ability of theGBS to predict low risk of need for intervention or deathwas evaluated at the following cut-off values: 0, �1, and�2. Two previously reported modifications of the GBS
also were examined: first, an age-modified versiondefining low risk as follows: GBS �2 and age <70 years15
(amGBS1); and, second, an age-modified version defininglow risk as follows: GBS �2 when age between 60 and69 years was assigned 1 extra point and age �70 yearswas assigned 3 points10 (amGBS2).
Definition of Hemostatic Intervention
Hemostatic intervention was defined as endoscopictherapy, surgery, or interventional radiology performedduring the index hospital stay.
Definition of Low-Risk Patients
Low-risk status was defined as patients who did notneed a blood transfusion or hemostatic intervention, anddid not die during the index admission.
Treatment of Patients
The general treatment of patients followed the na-tional guidelines of the British Society of GastroenterologyEndoscopy Committee18 (Glasgow, Truro), the DanishSociety of Gastroenterology and Hepatology19 (Odense),and local guidelines (all centers). Patients were admittedto the Departments of Gastroenterology (Glasgow, Truro,Dunedin), or a specialized gastrointestinal bleeding unit(Odense). Critically ill patients were transferred tointensive care units based on clinical judgment and localresources. The majority of patients were endoscoped
January 2015 Optimal Use of the Glasgow Blatchford Score 117
within 24 hours from presentation to the hospital. Pre-endoscopy treatment with proton pump inhibitors(PPIs) was used in most patients in Truro and Dunedin. Incontrast, pre-endoscopy PPIs were not used in Glasgowand Odense. Ulcers with stigmata of recent hemorrhage(active hemorrhage, visible nonhemorrhaging vessel, oran adherent clot) were treated with a minimum of 2endoscopic modalities (injection of diluted adrenaline,thermal hemostatic devices, or hemoclips) followed byinfusion of a PPI for 72 hours in all centers. Patientstreated with transcatheter arterial embolization weretreated in most cases by insertion of coils in the bleedingvessel. In patients with no active extravasation of contrast,the radiologist managed patients on a case-by-case basis,however, the majority had selective embolization guidedby a previously placed endoscopic hemoclip.
Statistical Analyses
The predictive ability of the GBS at the 3 cut-offvalues, and the 2 modified versions of the GBS, wereevaluated by calculation of sensitivity, specificity, posi-tive predictive values, and negative predictive values.Furthermore, we calculated the proportion of identifiedlow-risk patients, and the rate of transfusion, hemostaticintervention, and in-hospital mortality for that low-riskgroup. The Pearson chi-square test and the Fisher exacttest were used to compare proportions. The Man-n–Whitney U test and the Kruskal–Wallis equality-of-populations rank test were used to compare medians.A 2-tailed significance level of 5% was used in all com-parisons. Missing data were handled using list-wisedeletion. We used Stata (version 11; StataCorp, CollegeStation, TX) for data analysis.
Results
A total of 2305 patients from the 4 centers wereincluded. Patient characteristics are shown in Table 2.
Table 2. Characteristics of Included Patients
Glasgow(n ¼ 701)
Truro(n ¼ 540)
Median age [range], y 51.0 [13–95] 65.0 [14–99]Male sex 456 (65) 304 (57)Median GBS [range] 3 [0–18] 5 [0–19]GBS of 0a 141 (20) 73 (14)Actual low riskb 537 (77) 330 (61)Transfusionc 118 (17) 178 (33)Hemostatic interventiond 85 (12) 110 (21)Mortality 30 (4.3) 17 (3.1)
NOTE. Data refer to the number of patients (%), unless otherwise stated. There wreceived blood transfusions (n ¼ 3).aNumber of patients with a GBS of 0.bNumber of patients without need of intervention.cNumber of patients treated with a blood transfusion.dNumber of patients treated with endoscopic therapy, interventional radiology, o
As illustrated in Table 2, patient age varied betweencenters (P ¼ .0001). Patients were oldest in Dunedin(median age, 74.3 y), and youngest in Glasgow (medianage, 51 y). There were also major differences in the GBSdepending on the geographic site (P ¼ .0001). The GBSwas highest among patients from Dunedin (medianGBS, 9), and lowest among patients from Glasgow (me-dian GBS, 3) (Table 2).
Seventy-seven percent of patients overall wereendoscoped. Findings at endoscopy are shown in Table 3.Peptic ulcer bleeding and esophagitis, gastritis, or duo-denitis were the most common findings in all centers.Patients not endoscoped had a median GBS of 1 (range,0–8). The overall mortality rate among patients notendoscoped was 5.1%. The mortality rate among pa-tients not endoscoped with a GBS of �1 was 0.8%.
The proportion of patients requiring hemostaticintervention or transfusion varied between centers (P <.0001). The rate of hemostatic intervention or trans-fusion was highest in Odense (52%), and lowest inGlasgow (23%). There were also significant differencesin mortality rates (P < .015), with mortality highest inOdense (7%) and lowest in Truro (3%) (Table 2).
The predictive abilities of the different versions of theGBS, based on data from all centers, are shown inTable 4. The GBS at a cut-off value of �1 and �2, and theage-extended GBS versions, all identified a higher pro-portion of low-risk patients compared with the GBS at acut-off value of 0 (P < .001). All of the systems werecharacterized by high sensitivities. The GBS at a cut-offvalue of �2 had the highest specificity (48.9%)(Table 4), but patients classified as low risk using thiscut-off value had a higher need for hemostatic inter-vention, transfusion, or mortality (22 of 704) comparedwith the GBS at a cut-off value of 0 (3 of 313; P ¼ .047),GBS value of �1 (7 of 562; P ¼ .036), the amGBS1 (6 of588; P ¼ .012), and the amGBS2 (4 of 569; P ¼ .002). TheGBS at a cut-off value of �1 identified as many low-riskpatients as the age-extended versions of the GBS(Table 4). There were no differences between the GBS at
NOTE. Data are shown as the number of patients (%).
118 Laursen et al Clinical Gastroenterology and Hepatology Vol. 13, No. 1
a cut-off value of 0, the GBS at a cut-off value of �1, andthe age-extended versions of the GBS with respect tothe proportion of classified low-risk patients needinghemostatic intervention, transfusion, or dying duringadmission (P ¼ .83). The performance of the GBS at thedescribed thresholds and the score modifications in eachcenter is available in Supplementary Tables 1–5.
Two patients (0.4%) with a GBS of �1 received ablood transfusion. In Glasgow, Truro, and Dunedin, noneof the patients classified as low risk by a GBS of �1needed endoscopic therapy, surgery, radiologic inter-vention, or died during admission. Among patientsincluded from Odense, 4 patients with a GBS of �1 weretreated with endoscopic therapy. In 2 of these cases theendoscopist chose to inject diluted adrenaline becausethere was uncertainty about the presence of an under-lying vessel. In both cases, the endoscopist concludedthat no vessel was present. The third patient had anesophageal laceration with an adherent clot. The fourthpatient had a Dieulafoy’s lesion with minor bleeding.Therefore, across all sites, only 2 patients with a GBS of�1 (0.4%) de facto needed full endoscopic therapy. Noneneeded surgery. A total of 2 patients (0.4%) with a GBSof �1 died. Both patients were older than 85 years andwere admitted with coffee-ground vomiting. One patient
Table 4. Predictive Abilities of the Different Versions of the GBS
NPV, negative predictive value; PPV, positive predictive value.aNumber (%) of classified low-risk patients according to risk-scoring system.bNumber (%) of patients needing transfusion among patients classified as low riscNumber (%) of patients needing endoscopic therapy, surgery, or arterial embolizdNumber (%) of patients dying during admission among patients classified as low
had a cerebrovascular insult and died the following day.The other patient had urinary sepsis and died after 3days. None of these patients had evidence of significantUGIH during admission, and therefore did not undergoendoscopy. In neither case was the cause of death relatedto gastrointestinal bleeding. Overall, 1.2% of patientswith a GBS of �1 required transfusion, hemostaticintervention, or died. Fourteen percent of patients with aGBS of �1 were older than 70 years.
Among patients classified as low risk according to theamGBS1, 2 patients (0.3%) required transfusion, 3 pa-tients (0.5%) needed endoscopic therapy, and 1 patient(0.2%) died of unknown causes. Therefore, a total of1.0% of low-risk patients identified using amGBS1required transfusion, hemostatic intervention, or died.Among patients classified as low risk using the amGBS2,2 patients (0.4%) required transfusion, 1 patient (0.2%)needed endoscopic therapy, and 1 patient (0.2%) diedof unknown causes. Therefore, a total of 0.7% of low-riskpatients identified by amGBS2 required transfusion,hemostatic intervention, or died. Of patients with a GBSof �2, there were 6 patients (0.9%) who requiredtransfusion, 14 patients (2%) who needed endoscopictherapy, and 4 patients (0.6%) who died of unknowncauses. Therefore, a total of 3% of patients with a GBS of
A major challenge in the assessment of patientspresenting with UGIH is the accurate identification ofa low-risk group who will not require hospital-basedintervention or die. Several studies have shown thatthe GBS can identify a cohort of low-risk patientssuitable for outpatient management.2,8–10,12,13 At theaccepted low-risk definition of a GBS of 0,20 5% to 22%of patients could be offered outpatient care,8,14 unlesshospital admission is required for other reasons. How-ever, this accounts for only a minority of patients whodo not require intervention and survive. The presentstudy illustrates that by increasing the cut-off value to�1, or by use of an age-modified version of the GBS, thespecificity is increased from 22% to 40% to 42%. Thismeans that it is possible, at presentation, to identifyalmost half of the patients who will not need hospital-based treatment for UGIH or die. Because 61% of allpatients presenting with UGIH do not need intervention(Table 2) this corresponds to possible outpatient man-agement of approximately 25% of all patients present-ing with UGIH. Many of these patients, however, willneed admission for other conditions. A previous studyfound that outpatient management was possible in 68%of patients evaluated as low risk using a GBS of 0.8
Although generally higher-risk patients may be morelikely to need admission for other reasons, it is possiblethat use of a GBS of �1, or an age-modified version, mayfacilitate outpatient management of 15% to 20% of allpatients presenting with UGIH. Use of these GBS varia-tions is not associated with a higher risk of need ofhospital-based intervention or death among classifiedlow-risk patients when compared with the GBS at a cut-off value of 0.
According to our results, the GBS at a cut-off value of�1 and the age-modified versions perform similarly withrespect to the proportion of identified low-risk patientsand the risk of adverse outcome among patients classi-fied as low risk. Because 14% of patients with a GBS of�1 are older than age 70 years it is possible that clini-cians might be cautious about outpatient management ofthis subgroup. This potentially could reduce the clinicallyutility of the GBS of �1 compared with the age-modifiedversions of the GBS. However, because our results indi-cate that patients with a GBS of �1, independent of age,have a very low risk of requiring transfusion, hemostaticintervention, or death we do not believe that thissignificantly would impact the clinically utility of a GBSof �1. There were no differences in the performance ofthe 2 age-modified versions of the GBS. The amGBS1 maybe best because it is less complicated. Our results indi-cate that identification of patients suitable for outpatientmanagement using a GBS of �2 is less safe because 3%
of classified low-risk patients needed hospital-basedintervention.
Our findings are in accordance with a prospectivestudy from England where use of the GBS at a cut-offvalue of �1 correctly identified 39% of low-risk pa-tients and 99% of patients needing intervention.12 Thesafety of using a cut-off value of �1 also has been re-ported in smaller retrospective studies from England9
and Japan.21 Use of screening methods for identifica-tion of low-risk patients will always be associated with aminor risk of a false-negative result. For example, pa-tients classified as being at low risk of having deep-veinthrombosis using the Wells score have a 3% risk ofhaving a deep-vein thrombosis.22
Previous studies from the United Kingdom,9,12 TheNetherlands,13 and Japan21 have found that a GBS of �2has 97% to 100% specificity. These studies all wereretrospective. Studies reporting a specificity of 100%were characterized by very small sample sizes (n ¼93–174).9,21 Our data support the findings of Le Jeuneet al,12 indicating that the specificity of a GBS of �2 is97% to 98%. Although the specificity of a GBS of �2 ishigh, many clinicians may find a 3% rate of adverseevents among patients classified as low risk using thiscut-off value is too high for use in clinical practice.
The present study shows that the performance of theGBS is dependent on geographic location. Although weobserved differences in patient age (51–68 y), medianGBS (3–7), need for intervention (23%–52%), and mor-tality rate (3.1%–7.0%) among patients from Scotland,England, and Denmark, the GBS performed relativelyconsistently in these European populations. This in-dicates that use of the GBS appears valid in countrieswith similar systems of health care delivery.
Patients from Dunedin were older than from othercenters but the need for hemostatic intervention andmortality were similar to the European centers. Despitethe fact that 53% of patients included in Dunedin did notneed intervention, only 1.3% of patients had a GBS of0 and 3.6% had a GBS of �1. The cause of this is unclear,but may be owing to retrospective data collection. Thedifference in the proportion of low-risk patients alsomight be a result of the large rural catchment area ofDunedin and health care delivery there. In some cases,patients needed to travel more than 4 hours to anendoscopy center. It is likely that some low-risk patientswere managed by their general practitioner, or in thelocal rural hospitals, without being transferred to Dun-edin for a gastroscopy. There is also a no-fault compen-sation system in New Zealand that may have affected thenumber of lower-risk patients who were referred to thehospital. Our results indicate that the use of this score issafe, but not particularly useful in the Dunedin popula-tion because of low specificity.
Previous studies of the GBS also have indicated thatutilization of the GBS outside Europe is associated withlow specificity. A large prospective study from HongKong found a specificity of the GBS of 6.3% at a cut-off
120 Laursen et al Clinical Gastroenterology and Hepatology Vol. 13, No. 1
value of 0 for identification of patients needing endo-scopic therapy.14 A retrospective study from Japan re-ported a similar low specificity (4.3%–8.7%) foridentification of low-risk patients.21 These studies werebased only on patients who underwent endoscopy. Thesensitivity of the GBS was 100% in both studies, indi-cating that use of the GBS is safe.
The current evidence suggests that the GBS is lesseffective, but safe to use, in identification of low-riskpatients in countries outside Europe. The reason forthis difference is unclear but might be caused by differ-ences in population characteristics and health care de-livery. Therefore, the performance of the GBS needs to bevalidated further before implementation in countrieswith populations and health care delivery systemssignificantly different from the conditions in Europe.
Limitations of our study include the retrospectivecollection of data from Dunedin. This seems less impor-tant because the hospital protocol dictated hospitaladmission and performance of endoscopy in all patientswith UGIH. Based on this we believe that the risk ofhaving missed a significant proportion of patients islimited.
Pre-endoscopy PPIs were used in 2 centers (Truroand Dunedin). Use of pre-endoscopy PPIs is known toreduce the proportion of patients with high-risk stigmataof hemorrhage at index endoscopy and the proportion ofpatients needing endoscopic therapy.23 Therefore, thedifference in use of pre-endoscopy PPIs between centerscould have an impact on the proportion of patientsneeding intervention. Nevertheless, in the present studythe need for hemostatic intervention was lowest inGlasgow, where pre-endoscopy PPIs were not used. Thisindicates that the observed differences in the need forintervention were related to differences in the popula-tion characteristics and not use of pre-endoscopic PPIs.Endoscopic, surgical, radiologic, and post-therapy man-agement protocols for patients were similar in allcenters.
There were some differences between sites in themanagement of low-risk patients presenting withsymptoms of UGIH. After 6 months of the local audit, aguideline for nonadmission of patients with a GBS of0 was introduced in Glasgow. Patients who were notadmitted were offered outpatient endoscopy. All patientswere followed up for a minimum of 6 months. In Truroall patients were managed as inpatients in the first year.During the second year, 10% of patients were managedas outpatients using the amGBS1. Ninety-nine percent ofpatients had an endoscopy. Outpatients were followedup for 4 to 6 weeks. In Odense all patients were admittedfor clinical evaluation and the decision to performendoscopy was based on history, clinical findings, andblood tests. All patients were followed up for 1 monthafter admission. It was hospital policy in Dunedin toadmit all patients presenting with UGIH and to performendoscopy during admission. We believe that these dif-ferences in the management of low-risk patients did not
affect the main findings of the study because all patientswho were not admitted or endoscoped for whateverreason were followed up to assess whether they neededany intervention or died.
The original criteria for outpatient management usingthe amGBS1 also required that patients were accompa-nied at home, had a telephone and transport, had noactive significant comorbidities, were not takingwarfarin, and did not have suspected variceal bleeding.15
The present study shows that even without these criteriathe amGBS1 is good at identifying patients without needof hospital-based intervention.
In conclusion, we have shown that the GBS at a cut-offvalue of �1, the amGBS1, and the amGBS2 can be usedsafely and effectively to reduce unnecessary admissionsfor suspected UGIH. Because a GBS cut-off value of �1and the age-modified GBSs perform similarly with nosignificant differences in relevant end points, the simplerGBS cut-off value of �1 may be best for routine use.However, in areas with local clinical concern aboutavoiding admission of older patients, use of an age-modified GBS could be considered. By implementationof a protocol for nonadmission of UGIH-patients basedon one of these scores we expect that 15% to 20% of alladmissions can be avoided safely. Although the GBS ap-pears to perform consistently in European centers, ourdata suggest that further validation is needed before itswidespread use in other parts of the world with differenthealth care delivery systems.
Supplementary Material
Note: To access the supplementary material accom-panying this article, visit the online version of ClinicalGastroenterology and Hepatology at www.cghjournal.org,and at http://dx.doi.org/10.1016/j.cgh.2014.07.023.
References
1. Rockall TA, Logan RFA, Devlin HB, et al. Risk assessment after
acute upper gastrointestinal haemorrhage. Gut 1996;38:316–321.
2. Blatchford O, Murray WR, Blatchford M. A risk score to predictneed for treatment for upper gastrointestinal haemorrhage.Lancet 2000;356:1318–1321.
3. Saeed ZA, Winchester CB, Michaletz PA, et al. A scoring systemto predict rebleeding after endoscopic therapy of non-varicealupper gastrointestinal hemorrhage. Am J Gastroenterol 1993;88:1842–1849.
4. Hay JA, Lyubashevsky E, Elashoff J, et al. Upper gastrointestinalhemorrhage clinical guideline determining the optimal hospitallength of stay. Am J Med 1996;100:313–322.
5. Longstreth GF, Feitelberg SP. Successful outpatient manage-ment of acute upper gastrointestinal hemorrhage: use of prac-tice guidelines in a large patient series. Gastrointest Endosc1998;47:219–222.
6. Cipolletta L, Bianco MA, Rotondano G, et al. Outpatient man-agement for low-risk nonvariceal upper GI bleeding: a ran-domized controlled trial. Gastrointest Endosc 2002;55:1–5.
January 2015 Optimal Use of the Glasgow Blatchford Score 121
7. Cameron EA, Pratap JN, Sims TJ, et al. Three year prospectivevalidation of a pre-endoscopic risk stratification in patients withacute upper-gastrointestinal haemorrhage. Eur J GastroenterolHepatol 2002;14:497–501.
8. Stanley AJ, Ashley D, Dalton HR, et al. Outpatient managementof patients with low-risk upper-gastrointestinal haemorrhage:multicentre validation and prospective evaluation. Lancet 2009;373:42–47.
9. Srirajaskanthan R, Conn R, Bulwer C, et al. The GlasgowBlatchford scoring system enables accurate risk stratification ofpatients with upper gastrointestinal haemorrhage. Int J ClinPract 2010;64:868–874.
10. Laursen SB, Hansen JM, Schaffalitzky de Muckadell OB. TheGlasgow Blatchford score is the most accurate assessment ofpatients with upper gastrointestinal hemorrhage. Clin Gastro-enterol Hepatol 2012;10:1130–1135.
11. Cheng DW, Lu YW, Teller T, et al. A modified Glasgow Blatch-ford Score improves risk stratification in upper gastrointestinalbleed: a prospective comparison of scoring systems. AlimentPharmacol Ther 2012;36:782–789.
12. Le Jeune IR, Gordon AL, Farrugia D, et al. Safe discharge ofpatients with low-risk upper gastrointestinal bleeding (UGIB):can the use of Glasgow-Blatchford Bleeding Score beextended? Acute Med 2011;10:176–181.
13. Schiefer M, Aquarius M, Leffers P, et al. Predictive validity of theGlasgow Blatchford bleeding score in an unselected emergencydepartment population in continental Europe. Eur J Gastro-enterol Hepatol 2012;24:382–387.
14. Pang SH, Ching JY, Lau JY, et al. Comparing the Blatchford andpre-endoscopic Rockall score in predicting the need for endo-scopic therapy in patients with upper GI hemorrhage. Gastro-intest Endosc 2010;71:1134–1140.
15. Stephens JR, Hare NC, Warshow U, et al. Management of minorupper gastrointestinal haemorrhage in the community using theGlasgow Blatchford Score. Eur J Gastroenterol Hepatol 2009;21:1340–1346.
16. Stanley AJ, Dalton HR, Blatchford O, et al. Multicentre com-parison of the Glasgow Blatchford and Rockall scores in theprediction of clinical end-points after upper gastrointestinalhaemorrhage. Aliment Pharmacol Ther 2011;34:470–475.
17. World Medical Association. World Medical Association Decla-ration of Helsinki: ethical principles for medical researchinvolving human subjects. JAMA 2013;310:2191–2194.
18. British Society of Gastroenterology Endoscopy Committee.Non-variceal upper gastrointestinal haemorrhage: guidelines.Gut 2002;51:iv1–iv6.
19. Laursen SB, Jørgensen HS, Schaffalitzky de Muckadell OB.Management of bleeding gastroduodenal ulcers. Dan Med J2012;59:C4473.
20. National Institute for Health and Clinical Excellence. Man-agement of acute upper gastrointestinal bleeding. Clinicalguideline 141. 2012. Available at: http://guidance.nice.org.uk/CG141. Accessed November 1, 2013.
21. Masaoka T, Suzuki H, Hori S, et al. Blatchford scoring system isa useful scoring system for detecting patients with uppergastrointestinal bleeding who do not need endoscopic inter-vention. J Gastroenterol Hepatol 2007;22:1404–1408.
22. Wells PS, Anderson DR, Bormanis J, et al. Value of assessmentof pretest probability of deep-vein thrombosis in clinical man-agement. Lancet 1997;350:1795–1798.
23. Sreedharan A, Martin J, Leontiadis GI, et al. Proton pump in-hibitor treatment initiated prior to endoscopic diagnosis in uppergastrointestinal bleeding. Cochrane Database Syst Rev 2010;7:CD005415.
Reprint requestsAddress requests for reprints to: Stig Borbjerg Laursen, MD, PhD, Departmentof Medical Gastroenterology, Odense University Hospital, Søndre Boulevard29, 5000 Odense C, Denmark. e-mail: [email protected]; fax: (45) 66111328.
AcknowledgmentsThe authors thank the members of the Upper Gastrointestinal HemorrhageInternational Consortium: Chelsea Baines, George Benson, Aidan Cahill, EmilyFawcett, Terry Fesaitu, Jose A García, Daniel R Gaya, Nicola Hare, HyderHussaini, Hin Leung, Cara McLaughlin, Peter McLeod, William R Murray,Heather Norton, Nathan O’Donnell, Jeong-Yoon Park, Adibah Salleh, UsamaWarshow, and Wei Zhang for help with data collection and support in runningthe study.
Conflicts of interestThe authors disclose no conflicts.
NPV, negative predictive value; PPV, positive predictive value.aNumber (%) of patients classified as low-risk patients.bNumber (%) of patients needing transfusion among patients classified as low risk.cNumber (%) of patients needing endoscopic therapy, surgery, or arterial embolization among patients classified as low risk.dNumber (%) of patients dying during admission among patients classified as low risk.
Supplementary Table 1. Predictive Ability of GBS at a Cut-off Value of Zero
NPV, negative predictive value; PPV, positive predictive value.aNumber (%) of patients classified as low-risk patients.bNumber (%) of patients needing transfusion among patients classified as low risk.cNumber (%) of patients needing endoscopic therapy, surgery, or arterial embolization among patients classified as low risk.dNumber (%) of patients dying during admission among patients classified as low risk.
121.e1 Laursen et al Clinical Gastroenterology and Hepatology Vol. 13, No. 1
Supplementary Table 3. Predictive Ability of GBS at a Cut-off Value of 2 or Less
NPV, negative predictive value; PPV, positive predictive value.aNumber (%) of patients classified as low-risk patients.bNumber (%) of patients needing transfusion among patients classified as low risk.cNumber (%) of patients needing endoscopic therapy, surgery, or arterial embolization among patients classified as low risk.dNumber (%) of patients dying during admission among patients classified as low risk.
Supplementary Table 4. Predictive Ability of amGBS1
NPV, negative predictive value; PPV, positive predictive value.aNumber (%) of patients classified as low-risk patients.bNumber (%) of patients needing transfusion among patients classified as low risk.cNumber (%) of patients needing endoscopic therapy, surgery, or arterial embolization among patients classified as low risk.dNumber (%) of patients dying during admission among patients classified as low risk.
Supplementary Table 5. Predictive Ability of amGBS2
NPV, negative predictive value; PPV, positive predictive value.aNumber (%) of patients classified as low-risk patients.bNumber (%) of patients needing transfusion among patients classified as low risk.cNumber (%) of patients needing endoscopic therapy, surgery, or arterial embolization among patients classified as low risk.dNumber (%) of patients dying during admission among patients classified as low risk.
January 2015 Optimal Use of the Glasgow Blatchford Score 121.e2
