Article

External validation of the Glasgow-Blatchford Bleeding Score and the Rockall Score in the US setting

Unlabelled imageAmerican Journal of Emergency Medicine (2012) 30, 673-679

Original Contribution

External validation of the Glasgow-Blatchford Bleeding Score and the Rockall score in the US setting

Subhash Chandra MBBS a, Erik P. Hess MD a,b,?, Dipti Agarwal MBBS a,

David M. Nestler MD a,b, Victor M. Montori MD b,c, Louis M. Wong Kee Song MD d,

George A. Wells PhD f, Ian G. Stiell MD e,f

aDivision of Emergency Medicine Research, Department of Emergency Medicine, Mayo Clinic College of Medicine, Rochester,

MN 55905, USA

bKnowledge and Evaluation Research Unit, Mayo Clinic College of Medicine, Rochester, MN 55905, USA

cDivision of Endocrinology and Nutrition, Department of Medicine, Mayo Clinic College of Medicine, Rochester,

MN 55905, USA

dDivision of Gastroenterology, Department of Medicine, Mayo Clinic College of Medicine, Rochester, MN 55905, USA eDepartment of Emergency Medicine, Ottawa Hospital Research Institute, University of Ottawa, Ottawa, Ontario, Canada K1N 6N5 fDepartment of Epidemiology and Community Medicine, University of Ottawa, Ottawa, Ontario, Canada K1H 8M5

Received 23 December 2010; revised 15 March 2011; accepted 15 March 2011

Abstract

Background: The Glasgow-Blatchford Bleeding Score (GBS) and Rockall Score are clinical decision rules that risk stratify emergency department (ED) patients with Upper gastrointestinal bleeding (UGIB). We evaluated GBS and RS to determine the extent to which either score identifies patients with UGIB who could be safely discharged from the ED.

Methods: We reviewed and extracted data from the electronic medical records of consecutive adult patients who presented with signs or symptoms of UGIB (hematemesis and/or melena) to an academic ED from April 1, 2004, to April 1, 2009. The primary outcome was need for intervention (blood transfusion and/or endoscopic/surgical intervention) or death within 30 days.

Results: We identified 171 patients with the following characteristics: mean age of 69.9 years (SD, 17.0 years ), 52% women, 20% with a history of liver disease, and 22% with history of gastrointestinal bleeding. Ninety (52.6%, 95% confidence interval, 44.9-60.3) patients had the primary outcome. GBS outperformed pre-endoscopy RS [Area Under the Receiver Operating Characteristic Curve = 0.79 vs 0.62; P = .0001; absolute difference, 0.17]. The Prognostic accuracy of GBS and post-endoscopy RS was similarly high (AUC, 0.79 vs 0.72; P = .26; absolute difference, 0.07). The specificity of GBS and RS was suboptimal at all potential decision thresholds.

Conclusions: Although GBS outperformed pre-endoscopy RS, the prognostic accuracy of GBS and post-endoscopy RS was similarly high. The specificity of GBS and RS was insufficient to recommend use of either score in clinical practice.

(C) 2012

* Corresponding author. Division of Emergency Medicine Research, Department of Emergency Medicine, Mayo Clinic College of Medicine, Rochester, MN 55905, USA. Tel.: +1 507 255 7002; fax: +1 507 255 6463.

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

0735-6757/$ – see front matter (C) 2012 doi:10.1016/j.ajem.2011.03.010

Introduction

upper gastrointestinal bleeding is a common reason for hospital admission in the United States, with an annual prevalence of hospitalization of 126 in 100,000 [1]. Accurate risk stratification of patients who present to the emergency department (ED) with UGIB can assist the emergency medicine physician with the decision of whether to discharge the patient from the ED or to admit the patient to the hospital ward or intensive care unit. A clinically sensible and well-validated clinical prediction rule for UGIB has potential to safely reduce the burden of hospitalization and reduce health-care costs.

Several risk scores that predict the 30-day risk of adverse events in patients with UGIB have been developed [2-9]. Among these, the Rockall Score (RS) has been validated in several independent patient populations and thus constitutes level 2 evidence for clinical prediction rules [10,11], and the Glasgow-Blatchford Bleeding Score (GBS) has been pro- spectively validated and implemented in a recent multicenter study conducted in the UK and constitutes level 1 evidence (see Text box 1) [12]. The RS [5] was prospectively developed in 4 health regions in England and uses both clinical and endoscopic criteria to predict the risk of in- hospital mortality, although the accuracy of the pre- endoscopic components of the score to predict the need for endoscopic intervention has also been assessed in previous studies [9]. The GBS [12] uses variables readily available from the history, physical examination, and basic laboratory tests to predict need for intervention or death after UGIB and does not require endoscopy be performed before application of the score. Limited data exist on external validation of these scores [13], particularly GBS, in ED patients in US health- care settings.

The primary goal of this study was to externally validate GBS and RS to assess their ability to predict need for intervention or death in adults who present to the ED with UGIB.

Methods

Study design and setting

This health records review was conducted in an Urban academic ED in Rochester, Minn. The ED has an annual census of 72,000 patient visits. The electronic medical record (EMR) system at Mayo Clinic Rochester complies with the standards for clinical documentation issued by the American National Standards Institute Clinical Doc- ument Architecture [14]. Medical records of all patient visits (in-hospital, ED, and outpatient) at Mayo Clinic Rochester have been electronically available since Febru- ary 2002. The Mayo Clinic Institutional Review Board approved the study.

Text box 1. Clinical and laboratory variables included in GBS and pre- and post-endoscopy RS

Value

GBS

Blood urea nitrogen

>=6.5 to b8.0

>=8.0 to b10.0

>=10.0 to b25.0

>=25

Hemoglobin Men

>=120 to b130

>=100 to 120

b100

Women

>=100 to b120 b100

Systolic blood pressure (mm Hg) 100-109

90-99

b90

Pulse >=100 beats per minute Presentation with melena Presentation with syncope

Hepatic disease ?

cardiac failure +

Score

2

3

4

6

1

3

6

1

6

1

2

3

1

1

2

2

2

Pre-endoscopy RS

Age (y)

b60 0

60-79 1

>=80 2

Shock

No shock: systolic blood pressure >=100 mm Hg, 0

pulse b100 beats per minute

Tachycardia: systolic blood pressure >=100 mm Hg, 1

pulse >=100 beats per minute

Hypotension: systolic blood pressure b100 mm Hg 2

Comorbidity

No major comorbidity 0

Congestive heart failure, ischemic heart disease, 2

any major comorbidity

Renal failure, Liver failure, disseminated malignancy 3

Post-endoscopy RS

Same as pre-endoscopy RS plus Endoscopic diagnosis

Mallory-Weiss tear or no lesion observed 0

Pelvic ulcer disease, erosive esophagitis 1

Malignancy of upper GI tract 2

Stigmata of recent hemorrhage

Clean-based ulcer, flat pigmented spot 0

Blood in upper GI tract, clot, visible vessel, bleeding 2

* Known history or clinical and laboratory evidence of chronic or acute liver disease.

+ Known history or clinical and echocardiographic evidence of cardiac failure.

Patient variables and data collection”>Selection of participants

We designed a search strategy using a combination of International Classification of Disease, Ninth Revision codes and an electronic search of free text terms contained in the EMR. We searched the chief complaint, ED diagnosis, and primary and secondary hospital discharge diagnosis fields for the following terms: UGIB, GI bleeding, hematem- esis, melena, coffee-ground emesis, hematochezia, or bright red blood per rectum. We obtained a consecutive sample of adult patient records from April 2004 to July 2009 for review. As we assessed 30-day outcomes, presentation to the ED greater than 30 days after the index presentation was considered a new visit. We excluded any return ED visits occurring within 30 days of the index visit or patients who lived outside of Olmsted County, where Mayo Clinic Rochester resides. As all outpatient and inpatient records of patients who live in Olmsted County (population-based data) are available through infrastructure established by the Rochester Epidemiology Project, excluding patients outside of Olmsted county ensured high fidelity follow-up [15]. Records of patients who had not given previous authorization to have their Medical records reviewed for research purposes were also excluded in compliance with Minnesota Statute.

Patient variables and data collection

A protocol (available upon request) was written with input from the coauthors before conducting the study. We designed and piloted standardized data abstraction forms before beginning the review. Two physician reviewers (S.C. and D.A.A.) were trained by means of two 1-hour training sessions. The data abstraction form included all the variables in GBS and RS, as well as patient demographics, laboratory data, endoscopic findings and diagnoses, interventions performed including the type and date of intervention(s) (blood transfusion, endoscopic, angiographic, and/or surgical intervention required for hemostasis), hospital length of stay, and recurrence of UGIB or death within 30 days of the ED visit. We also recorded the ED length of stay and patient disposition from the ED (intensive care unit, hospital ward, observation unit, or home). The quality and accuracy of data collection were assessed by 2 independent reviewers, and the interrater reliability of historical variables potentially requiring judgment (history of congestive heart failure, history of liver disease, and presentation with syncope) was assessed in a random sample of 20% of included patients. The remaining variables were laboratory values or test results (did not require interpretation), and thus interrater reliability was not assessed. Any questions that arose during the chart review process were discussed at weekly meetings.

Score calculation

The GBS was calculated from 8 clinical or laboratory variables (initial pulse, hemoglobin, blood urea nitrogen,

systolic blood pressure, and presentation with melena, syncope, hepatic disease, or heart failure) obtained during ED assess- ment, as defined by Blatchford et al [2]. Pre-endoscopy RS was calculated from 3 clinical variables available at the initial ED assessment (age at presentation, shock, and comorbidities), and post-endoscopy RS was calculated from the admission RS plus endoscopic findings, as published by Rockall et al [5].

Outcome measures

The primary Composite outcome of this study was the need for intervention or death within 30 days of the ED visit. Need for intervention included blood transfusion or hemostasis achieved by endoscopic, angiographic, or surgical intervention. A patient who required any of these interventions or died within 30 days of ED presentation was classified as positive for the primary outcome. Additional outcomes recorded included bleeding recurrence or Hospital readmission for UGIB within 30 days. To decrease the risk of information bias, primary outcomes were recorded on a separate sheet from the rest of the data extraction form by reviewers who were blinded to the predictor variables. All potentially positive and 10% of randomly selected Negative outcomes were reviewed independently by 2 investigators who were blinded to the predictor variables. Disagreements were resolved by a third investigator.

Data analysis

SAS software version 9.1 TS Level 1M3 (SAS Institute, Cary, NC) was used for descriptive data analysis. All variables are presented as either means with SD or medians with interquartile range, as appropriate for the distribution of the data. Categorical variables are presented as percent frequency of occurrence. Estimates of sensitivity, specificity, and positive and negative likelihood ratios with 95% confidence intervals (CIs) were calculated at each potential decision threshold for GBS and the pre- and post-endoscopy RS using StatsDirect version 2.6.1 (StatsDirect, Cheshire, UK). The prognostic accuracy of the risk scores was compared by generating receiver operating characteristic curves and comparing the area under the curve (AUC) using the method described by Hanley and McNeil [16] for comparing ROC curves derived from the same cases (MedCalc Software version 10.4.0.0; MedCalc, Mariakerke, Belgium). When assessing potential decision thresholds in each respective risk score, our goal was to achieve near 100% sensitivity for need for intervention or death at the highest possible specificity.

Results

Between March 1, 2004, and July 1, 2009, we screened medical records from 986 ED patient visits. Of 986 records screened for inclusion, 614 were from patients who lived

outside of Olmsted County and were thus excluded. There were 171 patient visits included in the final analysis (Fig. 1). Complete follow-up data were available for all 171 visits. The ? values for the variables “presentation with syncope,” “congestive heart failure,” and “liver disease” were 0.65, 0.74, and 0.87, respectively, indicating substantial to near- perfect agreement in data abstraction. A third investigator resolved a disagreement in outcome assessment in 3 cases.

Demographic, clinical characteristics, and 30-day out- comes are shown in Table 1. Melena was the most common (70.2%) presenting complaint. A total of 145 (84.8%) patients were admitted to the hospital after their initial ED assessment. Upper GI endoscopy was performed in 136 patients (79.5%). There were 90 (52.6%, 95% CI, 44.9-60.3) patients positive for the primary outcome. UGIB recurred in 19 (11.1%) patients, and 12 (7.0%) patients were rehospi- talized for UGIB within 30 days of the ED visit. Of the 20 (11.7%) patients discharged home from the ED, 2 required endoscopic intervention within 30 days (thermal cautery); there were no deaths. Eight (4.7%) patients died within 30 days; 6 deaths occurred in-hospital.

There were 136 patients who underwent Upper endoscopy within 30 days. Of these, 46 (33.8%) had stigmata of recent bleeding. The most common endoscopic findings were hiatal hernia with Cameron erosions/ulcerations, gastritis, and Duodenal ulcer.

The overall accuracy of GBS for need for intervention or death within 30 days was 53.8%. The accuracy of pre-

endoscopy RS was 52.6%, and the accuracy of post- endoscopy RS was 61.0%. At a score of 0, GBS had a sensitivity of 99% (95% CI, 93-100) and specificity of 6%

(95% CI, 2-14) (Table 2). Sensitivity and specificity were also suboptimal at higher decision thresholds (N1, N2, and N3). None of the patients had a pre-endoscopy RS of 0. At a potential decision threshold N1, the specificity of pre- endoscopy RS was 4% (95% CI, 1-11). At higher decision thresholds, the sensitivity of pre-endoscopy RS was suboptimal. For post-endoscopy RS, none of the 136 patients were classified as low risk at the cutoff of N0. At higher decision thresholds, the specificity and/or sensitivity were suboptimal.

On ROC curves analysis, GBS outperformed pre- endoscopy RS in terms of overall prognostic accuracy (AUC = 0.79 vs 0.62; P = .0001; absolute difference, 0.17) (Fig. 2). The prognostic accuracy of GBS and post- endoscopy RS was similarly high (AUC, 0.79 vs 0.72; P =

.26; absolute difference, 0.07).

Discussion

Summary of major findings

We externally validated 3 clinical prediction models for ED patients with UGIB using GBS and pre- and post-

Excluded patient records = 201 Hematochezia or lower GI bleed = 133 Gross blood in ostomy bag = 11 Shortness of breath = 10

Weakness = 6

Anemia = 4

Other = 37

Not from Olmsted County = 614

Eligible ED visits (n = 171)

Death within 30 days

Number of ED visits for GI bleeding (n = 372)

Total number of patient records screened (n = 986)

Yes = 8

No = 163

Fig. 1 Flow diagram of study Selection process. ?Excluded records classified as “shortness of breath,” “weakness,” “anemia,” and “other”

were ED visits unrelated to GI bleeding identified by free text search of the EMR.

Fig. 2 Receiver operating characteristic curves of GBS, post- endoscopy RS, and pre-endoscopy RS. Area under the curve: GBS,

Table 1 Demographic, clinical characteristics, and outcomes of 171 ED patients with UGIB

Characteristic

Mean age, SD (y) Female (%) Presenting symptoms Blood in vomitus

Coffee-ground emesis Melena

Presentation with syncope Medical history Congestive heart failure

Known coronary artery disease Chronic renal insufficiency Liver disease

Previous gastrointestinal bleeding Previous gastrointestinal surgery Disposition

Admitted to the hospital

Admitted to the intensive care unit Admitted to the observation unit Dismissed from ED

30-d outcomes

Need for intervention or death Blood transfusion Interventions to stop bleeding Endoscopic

Surgical

Bleeding recurrence

Hospital readmission for upper GI bleeding Death

No. (%) patients (N = 171)

69.9 (17.0)

89 (52.0)

31 (18.1)

30 (17.5)

120 (70.2)

21 (12.3)

42 (24.6)

57 (33.3)

54 (31.6)

34 (19.9)

37 (21.6)

49 (28.7)

145 (84.8)

64 (37.4)

6 (3.5)

20 (11.7)

90 (52.6)

61 (35.7)

48 (28.1)

47 (27.5)

1 (0.6)

19 (11.1)

12 (7.0)

8 (4.7)

0.79 (95% CI, 0.72-0.85); post-endoscopy RS, 0.72 (95% CI, 0.64-

0.80); pre-endoscopy RS, 0.62 (0.54-0.69).

sensitivity was near or at 100% at the cutoff of N0 for each of the respective scores, the specificities ranged from 2% to 6%. At higher cutoffs with greater specificity, the sensitivities were suboptimal. Of the 20 (11.7%) patients sent home from the ED in our cohort, only 2 required endoscopic intervention within 30 days (thermal cautery); there were no deaths. These findings suggest that clinical decision making by emergency physicians, in concert with GI specialists when needed, may be more optimal than making the decision to discharge a patient home from the ED on the basis of their GBS or RS alone.

Strengths and limitations

endoscopy RS. To the best of the authors’ knowledge, this is the first study to externally validate GBS to identify patients safe for discharge in a US ED setting. The AUC for GBS was greater than the AUC for pre-endoscopy RS, and the AUC for GBS and post-endoscopy RS was similarly high. The classification performance of GBS and RS at potential decision thresholds, however, was suboptimal. Although the

Table 2 Sensitivity and specificity of GBS, pre-endoscopy RS (at admission), and post-endoscopy RS (full) in predicting need for intervention or death at potential decision thresholds

Cutoff

GBS

Sensitivity (95% CI)

Pre-endoscopy RS (at admission)

Post-endoscopy RS (full)

Specificity (95% CI)

Sensitivity (95% CI)

Specificity (95% CI)

Sensitivity (95% CI)

Specificity (95% CI)

The major strengths of this study are the completeness and accuracy of data collection and 30-day follow-up. The high-fidelity Mayo Clinic Rochester EMR included com- plete data on demographic characteristics, comorbid condi- tions, and endoscopy findings. As Mayo Clinic Rochester is the primary provider of emergency medical services for Olmsted County, restricting the study to patients living

N0

0.99

(0.93-1.00)

0.06

(0.02-0.14)

*

*

*

*

N1

0.97

(0.90-0.99)

0.13

(0.07-0.23)

1.00 (0.95-1.00)

0.04 (0.01-0.11)

1.00 (0.94-1.00)

0.02 (0.00-0.11)

N2

0.96

(0.89-0.99)

0.21

(0.13-0.31)

0.96 (0.88-0.99)

0.07 (0.03-0.16)

1.00 (0.94-1.00)

0.04 (0.01-0.14)

N3

0.92

(0.84-0.97)

0.33

(0.23-0.44)

0.80 (0.70-0.87)

0.39 (0.29-0.50)

0.94 (0.86-0.98)

0.21 (0.11-0.34)

N4

0.91

(0.83-0.96)

0.46

(0.35-0.58)

0.43 (0.32-0.54)

0.72 (0.61-0.81)

0.83 (0.73-0.90)

0.49 (0.35-0.63)

N5

0.87

(0.77-0.92)

0.59

(0.47-0.69)

0.11 (0.06-0.20)

0.96 (0.89-0.99)

0.55 (0.44-0.66)

0.74 (0.59-0.84)

N6

0.81

(0.71-0.88)

0.69

(0.57-0.78)

0.03 (0.01-0.10)

0.99 (0.92-1.00)

0.24 (0.16-0.35)

0.98 (0.88-1.00)

N7

0.78

(0.67-0.85)

0.75

(0.64-0.84)

*

*

0.11 (0.05-0.20)

1.00 (0.91-1.00)

* No patients were classified at this cut point.

within Olmsted County maximized the fidelity of patient follow-up. There are also several limitations as well. The study was conducted in a single center. Upper endoscopy is available on demand in our setting and our findings may not generalize to other settings without ready access to GI specialists who are able to perform upper endoscopy on an emergent basis. The study was retrospective in design. We did, however, take several steps to limit the intrusion of bias into the data abstraction process, including use of a structured protocol written before conducting the study, use of standardized and piloted data abstraction forms, abstractor training and monitoring by means of weekly meetings, blinding data abstractors to knowledge of the outcome and blinding outcome assessors to knowledge of the predictor variables, and interrater reliability assessment of historical variables potentially requiring judgment in data collection. We also had a relatively small sample size. As UGIB is not a high-volume condition (a recent multicenter study conducted in 4 EDs in the UK enrolled only 676 patients [12]), the medical record review design was the most pragmatic and cost-effective study design. Despite including only 171 ED patients with UGIB, there were 90 patients who were positive for the primary outcome within 30 days, providing a sufficient number of positive outcomes for external validation of GBS and RS.

Comparison with other published studies

The GBS has been derived and validated in 7 studies [2,7- 9,13,17,18] and robustly developed and implemented in the UK [12]. Of the validation studies, only 2 have been conducted in the US health-care setting [8,13]. Assessing the prognostic performance of GBS in a US health-care setting is likely important, as the decision to perform endoscopy is often a subjective decision and the threshold to intervene may differ between the US (a private payer) health system and public health-care systems such as the UK. Gralnek and Dulai [8] enrolled 175 patients with UGIB. However, they included only hospitalized patients, increasing the risk of spectrum bias when extrapolating the findings to ED patients with UGIB. In addition, the total number of positive outcomes and the classification performance (sensitivity, specificity, positive LR, and negative LR) of GBS were not reported in this study, providing little information on the prognostic accuracy and potential utility of GBS in ED patients. The study we report here includes patients presenting to the ED with UGIB and also includes estimates of the classification performance of GBS and RS. Farooq et al [13] compared pre-endoscopy RS and GBS with the clinical triage decision (admission to the intensive care unit vs admission to a nonintensive care unit) in 1244 patients identified from a hospital database. Similar to our observa- tions, these authors observed that clinical decision making by emergency physicians performed more optimally compared with triage decisions guided by GBS or RS alone. These authors, however, did not address the safety of the decision to

discharge a patient home from the ED. Finally, Courtney et al

[19] assessed the performance of a clinical protocol developed using clinical and laboratory data previously shown to be of predictive value in the published literature. These investigators retrospectively analyzed the performance of this protocol in 54 consecutive patients presenting to the ED during 1 calendar month. Using this protocol, they estimated that an additional 34% of patients that were admitted to the hospital could have been safely discharged. This protocol has not been validated.

As we observed in our study, in earlier studies, GBS had greater prognostic accuracy than pre-endoscopy admission RS [7,9,12]. In our study, however, the prognostic accuracy of GBS and post-endoscopy RS was similarly high. This may be attributed to our limited sample size; a larger sample size may increase the precision of statistical estimates and provide greater power to detect a difference in AUC.

Clinical implications and future directions

Because of the suboptimal specificity of GBS and RS observed in this study, we cannot recommend use of either score as the sole means to identify patients safe for ED discharge. The clinical judgment of emergency physicians on a case-by-case basis resulted in only 2 patients who were discharged from the ED and required intervention within 30 days; there were no deaths. Better prediction rules are needed to guide clinical practice.

Conclusion

The prognostic accuracy of GBS and post-endoscopy RS were similarly high. Because of the limited specificity of both GBS and RS at all potential decision thresholds, we do not recommend the use of either score as the sole means to identify patients safe for ED discharge.

References

  1. Monjur A. Prevalence of upper gastrointestinal bleeding in a community hospital. Am J Gastroenterol 2003;98:S294-5.
  2. Blatchford O, Murray WR, Blatchford M. A risk score to predict need for treatment for upper-gastrointestinal haemorrhage. Lancet 2000;356:1318-21.
  3. Hay JA, Lyubashevsky E, Elashoff J, Maldonado L, Weingarten SR, Ellrodt AG. Upper gastrointestinal hemorrhage clinical-guideline determining the optimal hospital length of stay. Am J Med 1996; 100:313-22.
  4. Kollef MH, Canfield DA, Zuckerman GR. Triage considerations for patients with acute gastrointestinal hemorrhage admitted to a medical intensive care unit. Crit Care Med 1995;23:1048-54.
  5. Rockall TA, Logan RF, Devlin HB, Northfield TC. Risk assessment after acute upper gastrointestinal haemorrhage. Gut 1996;38:316-21.
  6. Saeed ZA, Winchester CB, Michaletz PA, Woods KL, Graham DY. A scoring system to predict rebleeding after endoscopic therapy of nonvariceal upper gastrointestinal hemorrhage, with a comparison of heat probe and ethanol injection. Am J Gastroenterol 1993;88:1842-9.
  7. Chen IC, Hung MS, Chiu TF, Chen JC, Hsiao CT. Risk scoring systems to predict need for clinical intervention for patients with nonvariceal upper gastrointestinal tract bleeding. Am J Emerg Med 2007;25:774-9.
  8. Gralnek IM, Dulai GS. Incremental value of upper endoscopy for triage of patients with acute non-variceal upper-GI hemorrhage. Gastrointest Endosc 2004;60:9-14.
  9. Pang SH, Ching JYL, Lau JYW, Sung JJY, Graham DY, Chan FKL. Comparing the Blatchford and pre-endoscopic Rockall score in predicting the need for endoscopic therapy in patients with upper GI hemorrhage. Gastrointest Endosc 2010;71:1134-40.
  10. McGinn TG, Guyatt GH, Wyer PC, Naylor CD, Stiell IG, Richardson WS. Users’ guides to the medical literature: XXII: how to use articles about clinical decision rules. Evidence-Based Medicine Working Group. JAMA 2000;284:79-84.
  11. Laupacis A, Sekar N, Stiell IG. clinical prediction rules. A review and suggested modifications of methodological standards. JAMA 1997; 277:488-94.
  12. Stanley AJ, Ashley D, Dalton HR, Mowat C, Gaya DR, Thompson E, et al. outpatient management of patients with low-risk upper- gastrointestinal haemorrhage: multicentre validation and prospective evaluation. Lancet 2009;373:42-7.
  13. Farooq FT, Lee MH, Das A, Dixit R, Wong RCK. Clinical triage decision vs risk scores in predicting the need for endotherapy in upper gastrointestinal bleeding. Am J Emerg Med 2010 Dec. 23 [Epub ahead of print].
  14. Dolin RH, Alschuler L, Boyer S, Beebe C, Behlen FM, Biron PV, et al. HL7 Clinical Document Architecture, Release 2. J Am Med Inform Assoc 2006;13:30-9.
  15. Melton, III LJ. History of the Rochester Epidemiology Project. Mayo Clin Proc 1996;71:266-74.
  16. Hanley JA, McNeil BJ. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983;148:839-43.
  17. Masaoka T, Suzuki H, Hori S, Aikawa N, Hibi T. Blatchford scoring system is a useful scoring system for detecting patients with upper gastrointestinal bleeding who do not need endoscopic intervention. J Gastroenterol Hepatol 2007;22:1404-8.
  18. Stephens JR, Hare NC, Warshow U, Hamad N, Fellows HJ, Pritchard C, et al. Management of minor upper gastrointestinal haemorrhage in the community using the Glasgow Blatchford score. Eur J Gastro- enterol Hepatol 2009;21:1340-6.
  19. Courtney AE, Mitchell RM, Rocke L, Johnston BT. Proposed risk stratification in upper gastrointestinal haemorrhage: is hospitalisation essential? Emerg Med J 2004;21:39-40.

Leave a Reply

Your email address will not be published. Required fields are marked *