Predicting Poor Outcome from Acute Upper Gastrointestinal Hemorrhage

ORIGINAL INVESTIGATION Predicting Poor Outcome From Acute Upper Gastrointestinal Hemorrhage

Thomas F. Imperiale, MD; Jason A. Dominitz, MD, MHS; Dawn T. Provenzale, MD, MS; Lynn P. Boes, RN; Cynthia M. Rose, RN; Jill C. Bowers, RN; Beverly S. Musick, MS; Faouzi Azzouz, MS; Susan M. Perkins, PhD

Background: Uncertainty about the outcome of acute died. Independent predictors of poor outcome 1 were upper gastrointestinal bleeding often results in a longer- APACHE (Acute Physiology and Chronic Health Evalu- than-necessary hospital stay. ation) II score of 11 or greater, esophageal varices, and stigmata of recent hemorrhage. Predictors of poor outcome Methods: We derived and internally validated clinical 2 were these 3 factors plus unstable comorbidity on ad- prediction rules (CPRs) to predict outcome from upper mission. Of patients with no risk factors, only 1 (1.1%) of gastrointestinal bleeding. This multisite, prospective co- 92 experienced poor outcome 1 and only 6 (6.2%) of 97 hort study involved consecutive patients admitted for experienced poor outcome 2. Risks in the validation set acute upper gastrointestinal bleeding. Multivariate lo- were comparable. The CPRs identified 37.8% and 32.2% gistic regression was used to derive CPRs on two thirds of patients in the derivation and validation sets, respec- of the cohort (derivation set) that predicted bleeding- tively, who were eligible for a shorter hospital stay. specific outcomes (rebleeding, need for urgent surgery, or hospital death [poor outcome 1]) and bleeding- Conclusions: Patients admitted with acute upper gastro- specific outcomes plus new or worsening comorbidity intestinal bleeding were unlikely to have a poor outcome (poor outcome 2). Both CPRs were then tested on the if these risk factors were absent. These CPRs might make remaining third of the cohort (validation set). hospital management more efficient by identifying low- risk patients for whom early hospital discharge is possible. Results: A total of 391 individuals (99% men; mean age, 63.4 years) were enrolled, of which 4.6% rebled and 3.1% Arch Intern Med. 2007;167:1291-1296

CUTE UPPER GASTROINTES- ter admission; intermediate-risk patients, tinal (GI) hemorrhage is a who do not need intensive care; and high- life-threatening condition risk patients, who require aggressive care that results in 250 000 to in a closely monitored setting. 300 000 hospitalizations Although several clinical prediction rules and 15 000 to 30 000 deaths per year in the (CPRs) for risk stratification of patients with A1,2 United States. More than $2.5 billion are acute upper GI hemorrhage have been pub- spent annually for inpatient care of this lished,5-11 none have achieved widespread problem.2 Rates of major morbidity and use in clinical practice. The reasons in- mortality are 10% to 12% and 8% to 10%, clude insufficient validation,8-11 limited clini- respectively, and they have remained fairly cal applicability,5,7 complexity of use,6,10-12 constant during the past 40 years. Consid- and inability to identify patients at risk for erable variation in resource use for man- non-GI comorbidity that requires contin- agement has been demonstrated.2-4 ued inpatient care.5-11 We attempted to de- Most patients are at low risk for fur- velop and internally validate 2 CPRs to ther hemorrhage and mortality, but many stratify risk of poor outcome from acute up- remain in the hospital for several days be- per GI hemorrhage—one to predict risk of yond the high-risk period. Less common GI complications and the other to predict are patients readmitted soon after dis- risk of GI and non-GI complications. The charge for rebleeding or a delayed com- goal was to create CPRs that were clini- plication of the index hospitalization. Both cally relevant, valid, and easy to use. scenarios represent inefficiency resulting from uncertainty in predicting patient out- METHODS come. Accurate risk stratification would be useful for identifying low-risk pa- This prospective cohort study involved Veter- Author Affiliations are listed at tients, who might not require hospitaliza- ans Affairs Medical Centers in Durham, India- the end of this article. tion or who could be discharged soon af- napolis, and Seattle. The study protocol was ap-

(REPRINTED) ARCH INTERN MED/ VOL 167, JUNE 25, 2007 WWW.ARCHINTERNMED.COM 1291

Downloaded From: https://jamanetwork.com/ on 10/02/2021 as hematemesis, bloody nasogastric aspirate, or bleeding docu- Table 1. Criteria for Acute Upper mented endoscopically, along with either hypotension or a de- Gastrointestinal Hemorrhage* crease in the hematocrit value of more than 4% in 24 hours.

Hematemesis, defined as either DATA MANAGEMENT AND ANALYSIS Ն1 Episode of vomiting either fresh blood or a coffee-grounds–like material, witnessed by medical or nursing personnel OR A history of Ն1 episode of fresh blood or coffee-grounds emesis, Study nurses completed standardized data collection forms for along with either admission data, daily clinical information, endoscopic find- Hypotension (systolic BP Ͻ100 mm Hg), orthostasis (Ն20 ings, and postdischarge follow-up. Descriptive analyses were mm Hg decline in systolic BP from lying supine to standing or conducted on the entire data set. Before univariate analysis, the a decline of Ն10 mm Hg from lying supine to sitting, or an data set was randomly divided into a derivation set containing increase in heart rate of Ն20/min from lying supine to two thirds of the observations and a validation set containing standing, or an increase in heart rate of Ն10/min from lying the remaining third. supine to sitting) OR From the derivation set, we built a logistic regression model New anemia (hemoglobin or hematocrit level lower than the lower using methods suggested by Hosmer and Lemeshow.13 Uni- limit of normal and not previously documented in the medical record) or observed melena variate analysis was used to identify candidate variables for multivariate analysis for both composite outcomes using a thresh- Reported or observed melena with a diagnostic EGD Յ Reported hematemesis with a diagnostic EGD old P .20. Candidate variables were then entered into forward Observed or reported hematochezia with nasogastric aspirate of fresh and backward selection logistic regression to identify vari- or dark blood ables with significant main effects (␣=.05). Likelihood ratio and Score and Wald statistics were examined for each variable in Abbreviations: BP, blood pressure; EGD, esophagogastroduodenoscopy. the model, and each variable coefficient was compared with the *Patients were enrolled in the study if they met 1 or more of these criteria. coefficient from the univariate model. All 3 statistics were consistent in value and significance. Model goodness of fit was ex- proved by institutional review boards at each site. All the patients amined by computing the deviance, the Hosmer-Lemeshow statistic, and by examining outliers and influential groups through admitted to the hospital with acute upper GI hemorrhage were ⌬␤ 13,14 considered for enrollment if they met at least 1 criterion in computation of residuals and s. Model discrimination between patients with vs without poor outcomes was evaluated Table 1. Patients were excluded for any of the following rea- using the C statistic, which is the area under the receiver op- sons: bleeding while hospitalized for another reason; inpa- 15 tient transfer from another hospital; presence of a gastros- erating characteristic curve. Only the subgroup of 360 pa- tomy tube, esophageal stent, or other nonbiliary upper GI device; tients who underwent esophagogastroduodenoscopy were in- terminal illness affecting physician management of the bleed- cluded in the multivariate analyses. ing episode; and use of warfarin or full-dose heparin. To facilitate clinical use of the CPRs we decided a priori to During hospitalization, patients were followed up by re- assign 1 point for each independent predictor variable. Risks search nurses (L.P.B., C.M.R., and J.C.B.) who abstracted ad- of poor outcomes 1 and 2 were then measured based on the mission and daily information, including demographic fea- number of predictor variables, with discrimination measured tures; comorbidity; medications; results of laboratory, using the C statistic. Both models were tested on the remain- radiographic, and endoscopic tests; and treatments. Comor- ing third of the cohort for validation (reproducibility). C sta- bidity was considered unstable if it was identified as a prob- tistics were compared between the derivation and validation lem by the admitting physician and required orders directed subgroups. The nonparametric median test was used to com- toward diagnosis (eg, cardiac enzymes for chest pain) or treat- pare the length of stay among the different risk groups. A sta- ment (eg, nebulized bronchodilators for wheezing). Daily tistical software program (SAS version 9.1; SAS Institute Inc, progress notes were reviewed to determine whether complica- Cary, NC) was used for all analyses. tions occurred during hospitalization, including rebleeding, the need for advanced techniques to control bleeding (repeated en- RESULTS doscopy, angiography with embolization, transjugular intra- hepatic portosystemic shunt insertion, and surgery), the development of new or worsened comorbidity (identified as a DESCRIPTIVE FINDINGS separate problem requiring treatment), or death. Patients were contacted by telephone at least 30 days after hospital dis- A total of 1034 patients were evaluated during the 35- charge to determine whether they had rebled or were hospi- month enrollment period; 643 patients were excluded be- talized anywhere within 30 days of discharge. cause they did not meet the case definition of acute upper GI bleeding or they met the exclusion criteria: 231 DEFINITIONS OF OUTCOMES from Durham, 212 from Indianapolis, and 200 from Se- attle. Demographic features of the excluded patients and We defined 2 outcomes a priori. Poor outcome 1 is a GI hem- reasons for exclusion were comparable across study sites. orrhage–specific composite variable and includes rebleeding, the Three hundred ninety-one patients were enrolled: 134 need for urgent surgery or an advanced technique to control hem- from Durham, 110 from Indianapolis, and 147 from Se- orrhage (eg, radiographic embolization and transjugular intra- attle. Mean±SD patient age was 63.4±13.5 years, and 136 hepatic portosystemic shunt), and all-cause hospital mortality. patients (34.8%) were older than 70 years. Seventy-one Poor outcome 2 is a more-inclusive composite outcome that in- percent of the patients were white, 26% were black, and cludes poor outcome 1 plus new or worsening comorbidity, which was identified from a clinical diagnosis that developed subse- 99% were men. Clinical characteristics of the cohort are quent to admission (eg, pneumonia or stroke) or a preexisting given in Table 2. Hypotension (systolic blood pressure illness that required more than the patient’s usual medications Ͻ100 mm Hg) was present initially in 94 patients (24.0%). (eg, use of intravenous diuretics for congestive heart failure). For Esophagogastroduodenoscopy was performed in 360 pa- both outcomes, only major rebleeding was considered, defined tients (92.1%).

(REPRINTED) ARCH INTERN MED/ VOL 167, JUNE 25, 2007 WWW.ARCHINTERNMED.COM 1292

Downloaded From: https://jamanetwork.com/ on 10/02/2021 Table 2. Baseline Clinical, Endoscopic, and Laboratory Data for the Study Cohort

Patients, No. (%) Endoscopic Patients, No. (%) Patients, No. (%) Clinical Variable (n = 391) Variable (n = 360) Laboratory Variable (n = 391) SBP Ͻ100 mm Hg, heart rate Ͼ120/min, 177 (45.3) Therapeutic endoscopy 90 (25.0) Hemoglobin Ͻ10 g/dL 176 (45.0) or orthostasis Comorbid conditions, No. Endoscopy within Platelet count Ͻ150 000 76 (19.4) 0 108 (27.6) 12 h 96 (26.7) Albumin Ͻ3 g/dL 103 (26.3) 1 118 (30.2) 24 h 195 (54.2) Creatinine Ն1.5 mg/dL 126 (32.2) 2 95 (24.3) 48 h 292 (81.1) INR Ͼ1.3 67 (17.1) Ͼ2 70 (17.9) Esophageal varices 67 (18.6) Unstable comorbidity 61 (15.6) Dieulafoy lesion* or 47 (13.1) visible vessel Cirrhosis 70 (17.9) Peptic ulcer disease 178 (49.4) APACHE II score Ն11 144 (36.8) Stigmata of recent 108 (30.0) hemorrhage† Aspirin use‡ 233 (59.6) Ethanol use‡ 135 (34.5)

Abbreviations: APACHE, Acute Physiology and Chronic Health Evaluation; INR, international normalized ratio; SBP, systolic blood pressure. SI conversion factor: To convert creatinine to micromoles per liter, multiply by 88.4. *A Dieulafoy lesion is also known as a “caliber-persistent artery,” an artery of persistent large caliber in the submucosa with or without a small, overlying mucosal defect. †Active bleeding, visible vessel (either isolated or in an ulcer bed), fresh blood, or blood clot present endoscopically. ‡As reported in the medical record.

Clinical and demographic features were comparable Table 3. Selected Demographic and Clinical Characteristics among the 3 sites (Table 3). At all the sites, esophago- of the Patients at the 3 Study Sites* gastroduodenoscopies were performed by staff gastro- enterologists and fellows concurrently. Acid suppres- Site sion was used initially in more than 90% of patients, with no difference in use among sites based on whether stig- Durham, Indianapolis, Seattle, mata of recent hemorrhage (SRH) were present. Among NC Ind Wash patients with SRH, the proportions treated endoscopi- Characteristic (n = 110) (n = 134) (n = 147) cally were comparable across sites: 52% for Durham, 62% Age, mean ± SD, y 65.6 ± 14 62.4 ± 13 62.5 ± 13 for Indianapolis, and 55% for Seattle. The frequencies of Male sex 109 (99.1) 133 (99.3) 145 (98.6) White race 57 (51.8) 99 (73.9) 122 (83.0) endoscopic treatment types (ie, injection, electrocoagu- Initial orthostasis, 44 (40.0) 72 (53.7) 61 (41.5) lation, and ligation) were also comparable. hypotension, or pulse Thirty-one patients (7.9%) experienced major rebleed- rate Ͼ120/min ing (n=18; 4.6%), required surgery for bleeding (n=6), Unstable comorbidity at 18 (16.4) 25 (18.7) 18 (12.2) or died during the hospital stay (n=12) (poor outcome hospital admission Initial hemoglobin level 47 (42.7) 65 (48.5) 64 (43.5) 1), whereas 85 patients (21.7%) had major rebleeding, Ͻ10 g/dL required surgery, died during hospitalization, or devel- Endoscopy 102 (92.7) 122 (91.0) 136 (92.5) oped new or unstable comorbidity (poor outcome 2). Hos- Peptic ulcer 51 (46.4) 54 (40.3) 73 (49.7) pital mortality was 3.1% (n=12); death within 30 days Therapeutic endoscopy 16 (14.5) 39 (29.1) 35 (23.8) of hospital discharge occurred in 11 patients (2.8%). Hospital stay, median, d 5 4 3

UNIVARIATE AND MULTIVARIATE ANALYSES *Data are given as number (percentage) except where indicated otherwise.

Several variables were associated with each composite out- Based on the same derivation set of 244 patients, of come (Table 4). Because only 19 patients (7.8%) from which 52 (21.3%) experienced poor outcome 2, 4 vari- the derivation set of 244 experienced poor outcome 1, the ables were identified: presence of unstable comorbidity number of candidate variables was limited to 3 to avoid on admission, APACHE II score of 11 or greater, SRH, deriving an overfitted model. The final model included SRH, and esophageal varices (Table 4). This model demon- esophageal varices, and APACHE (Acute Physiology and strated good discrimination between patients with and Chronic Health Evaluation) II score of 11 or greater without poor outcome 2 (C statistic=0.78) and good cali- (Table 5). The C statistic was 0.81, indicating good model bration (P=.69). Forward and backward elimination re- discrimination, and the goodness-of-fit test indicated good gression again resulted in the same independent vari- model calibration (P=.80). Forward and backward elimi- ables. In validation, in which the risk of poor outcome 2 nation regression produced the same 3 variables. Valida- was 22%, the C statistic was 0.76, which is no different tion on the remaining third of the cohort (n=116), in which statistically from that of the derivation group. the risk of poor outcome 1 was 7%, produced a C statistic Using the independent variables, we created 2 CPRs of 0.83, which was not statistically significantly different with which to predict poor outcome. Results for both from that for the derivation set. CPRs are given in Table 5. For poor outcome 1, 1 point

(REPRINTED) ARCH INTERN MED/ VOL 167, JUNE 25, 2007 WWW.ARCHINTERNMED.COM 1293

Downloaded From: https://jamanetwork.com/ on 10/02/2021 groups for each score level, we combined groups to in- Table 4. Odds Ratios and 95% Confidence Intervals crease the precision of the risk estimates. For the cohort Ͻ for Variables Significant in Univariate Analysis (P .20) of 360, risks of poor outcome 1 were 1.5%, 4.7%, and 24.7%, in the low-, intermediate-, and high-risk groups, Variable Poor Outcome 1 Poor Outcome 2 respectively. For poor outcome 2, comparable risks were Stigmata of recent hemorrhage 4.6 (1.7-12.0) 1.82 (0.96-3.5) 5.7%, 17.5%, and 49.0%, respectively. Risk gradients for APACHE II score Ն11 3.17 (1.3-7.9) 4.77 (2.5-9.0) all 3 risk categories resulted in clinically meaningful and Albumin level Ͻ2.5 mg/dL 3.58 (1.3-10.0) 2.53 (1.1-5.6) Esophageal varices 3.17 (1.2-8.6) 2.16 (1.0-4.5) statistically significant separation of risk for both CPRs INR Ͼ1.3 2.98.(1.1-7.9) 3.68 (1.8-7.6) (Table 6). C statistics for the derivation group, valida- Chronic liver disease 2.72 (1.1-6.6) 2.21 (1.2-4.2) tion group, and entire cohort were 0.81, 0.83, and 0.81, Ն2 Comorbid conditions 2.19 (0.9-5.3) 1.49 (0.82-2.7) respectively, for poor outcome 1 and 0.78, 0.76, and 0.78, Endoscopy within 24 h of 2.45 (0.85-7.0) NA respectively, for poor outcome 2. hospital admission The CPRs were then dichotomized into a score of 0 BUN level Ͼ50 mg/dL 2.11 (0.84-5.3) NA Table 7 Creatinine level Ն1.5 mg/dL 1.97 (0.81-4.8) 2.71 (1.5-5.0) vs 1 or more points ( ). The dichotomized CPR Vascular lesion seen on 2.37 (0.92-6.1) NA detected 25 of 27 patients with poor outcome 1 (sensi- endoscopy tivity, 92.6%) and identified 136 patients (37.8%) with Platelet count Ͻ150 000/mL NA 1.83 (0.94-3.6) a score of 0 and a good outcome who were eligible for a Hematocrit level Ͻ30% NA 1.79 (0.96-3.3) shorter hospital stay. For poor outcome 2, the dichoto- Aspirin use NA 1.65 (0.91-3.0) mized CPR detected 71 of 78 patients (sensitivity, 91.0%) Electrocardiogram with acute NA 1.69 (0.93-3.1) changes and identified 116 patients (32.2%) who were eligible for Initial systolic blood pressure NA 1.83 (1.0-3.3) a shorter hospital stay. Positive and negative likelihood Ͻ100 mm Hg, heart rate ratios are 1.6 and 0.17, respectively, for outcome 1 and Ͼ120/min, or orthostasis 1.5 and 0.22 for outcome 2, suggesting greater value in excluding poor outcomes among persons with none of Abbreviations: APACHE, Acute Physiology and Chronic Health Evaluation; the factors, reflective of the CPRs’ high sensitivity. BUN, blood urea nitrogen; INR, international normalized ratio; NA, not applicable. Length of hospital stay was closely related to risk score. SI conversion factor: To convert creatinine to micromoles per liter, For poor outcomes 1 and 2, median lengths of stay for multiply by 88.4. patients with risk scores of 0, 1, and 2 or more were 3, 4, and 5 days, respectively (PϽ.001). Of the 136 patients who did not experience poor outcome 1 and who had Table 5. Odds Ratios and 95% Confidence Intervals no risk factors for it (ie, a risk score of 0), 11 (8.1%) were for Variables Significant in Multiple Logistic Regression readmitted within 30 days of discharge. These 11 indi- and C Statistics for the Derivation Set of 244 Patients viduals, readmitted a mean of 13.7 days after discharge, had readmission diagnoses consistent with their index Variable Poor Outcome 1 Poor Outcome 2 admission diagnoses: 5 had recurrent upper GI bleed- Stigmata of recent hemorrhage 5.28 (1.89-14.8) 2.24 (1.06-4.61) ing, 1 had elective resection of a gastric adenocarci- Ն APACHE II score 11 3.11 (1.13-8.57) 3.87 (1.92-7.81) noma, 1 had upper GI bleeding after starting anticoagu- Esophageal varices 3.85 (1.31-11.3) 2.77 (1.22-6.32) Unstable comorbidity at NA 5.25 (2.34-11.79) lation, and 1 each had anemia, syncope, bowel obstruction, hospital admission and a perforated gastric ulcer. These same 11 patients C statistic (95% confidence 0.81 (0.71-0.90) 0.78 (0.71-0.85) (9.5%) compose the group readmitted within 30 days interval) among the 116 patients who did not experience poor outcome 2 and who had no risk factors for it. Abbreviations: APACHE, Acute Physiology and Chronic Health Evaluation; The APACHE II score is often not available, so we used NA, not applicable. the derivation set post hoc to rerun the regression analyses without APACHE II scores to determine whether other was assigned to each predictor such that scores ranged models would discriminate comparably with the original from 0 to 3. Because the magnitude of risk of poor out- models. The model for poor outcome 1 had a C statistic come 1 was comparable for patients with scores of 2 and of 0.79 and contained the variables SRH (odds ratio [OR], 3, we combined these 2 subgroups, which resulted in 3 5.16; 95% confidence interval [CI], 1.86-14.34), esopha- risk categories: low risk (a score of 0), intermediate risk geal varices (OR, 4.39; 95% CI, 1.47-13.10), and 2 or more (a score of 1), and high risk (a score of 2 or 3), with risks comorbid conditions (OR, 3.15; 95% CI, 1.09-8.52). When of 1.1%, 5.0%, and 25.5%, respectively, in the deriva- this model was used as a risk index with 1 point for each tion group. We used the same procedures for poor out- variable, the risk of poor outcome 1 in the low-risk group come 2, assigning 1 point for each of 4 variables (the afore- (score of 0) was 1.2%, in the intermediate-risk group (score mentioned 3 plus unstable comorbidity on admission). of 1) was 5.3%, and in the high-risk group (score Ն2) was In this case, the high-risk category consisted of patients 24%. The model for poor outcome 2 had a C statistic of with scores of 2 or greater. Risks of poor outcome 2 for 0.76 and contained the following 4 variables: unstable co- low-, intermediate-, and high-risk categories were 4.8%, morbidity on admission (OR, 5.90; 95% CI, 2.55-13.68), 16.7%, and 46.5%, respectively. serum creatinine level of 1.5 mg/dL or greater (Ն133 Risk scores for the 2 CPRs in the validation group are µmol/L) (OR, 3.05; 95% CI, 1.27-7.41), esophageal vari- given in Table 6. Because the risks of both poor out- ces (OR, 3.41; 95% CI, 1.47-7.91), and SRH (OR, 2.85; comes were comparable in the derivation and validation 95% CI, 1.36-5.97). For the corresponding risk index, the

(REPRINTED) ARCH INTERN MED/ VOL 167, JUNE 25, 2007 WWW.ARCHINTERNMED.COM 1294

Downloaded From: https://jamanetwork.com/ on 10/02/2021 Table 6. Risk of Poor Outcomes 1 and 2 According to the Number of Factors Present

Patients With Risk, No. With Outcome/No. in Risk Group (%)

Derivation Set (n = 244) Validation Set (n = 116) Factor, No. Risk Group Outcome 1 Outcome 2 Outcome 1 Outcome 2 0 Low 1/92 (1.1) 4/83 (4.8) 1/46 (2.2) 3/40 (7.5) 1 Intermediate 5/101 (5.0) 15/90 (16.7) 2/48 (4.2) 10/53 (18.9) Ն2 High 13/51 (25.5) 33/71 (46.5) 5/22 (22.7) 13/23 (56.5) C statistic 0.81 0.78 0.83 0.76

low-risk group (score of 0) had a risk of 5%, the intermediate-risk group (score of 1) had a risk of 20%, and the Table 7. Effect of Dichotomizing the CPRs high-risk group (score Ն2) had a risk of 47%. In valida- on Test Characteristics and Identification of Patients Eligible for a Shorter Hospital Stay* tion, the C statistics for both models declined by approxi- mately 0.10. Poor Outcome Poor Outcome No. of Factors Present Absent Total COMMENT Poor outcome 1† Ն1 25 197 222 Acute upper GI hemorrhage is a life-threatening prob- 0 2 136 138 lem, one for which hospital-based care is often inefficient Subtotal 27 333 360 Poor outcome 2‡ because of a lack of accurate prognostic information. For Ն1 71 166 237 this reason, we derived and internally validated CPRs that 0 7 116 123 stratify risk of poor outcome, either directly related to bleed- Subtotal 78 282 360 ing or for any outcome that would prolong the hospital stay. These CPRs showed good discrimination, detecting Abbreviation: CPR, clinical prediction rule. nearly all patients with a poor outcome and identifying a *Data are presented as number of patients. †Sensitivity: 25/27 (93%); specificity: 136/333 (40.8%); eligible for subgroup for which a shorter hospital stay was possible. shorter hospital stay: 136/360 (37.8%). Although both CPRs performed reasonably well, nei- ‡Sensitivity: 71/78 (91%); specificity: 116/282 (41.1%); eligible for ther detected all patients with “poor outcomes.” When shorter hospital stay: 116/360 (32.2%). scores were dichotomized, the GI bleed–specific CPR missed 2 patients who experienced poor outcome 1 (of Several systems for risk stratification of acute upper a total of 138), both of whom had rebleeding. Both pa- GI hemorrhage have been published. They vary in how tients underwent endoscopy on the day of admission, re- and when they apply to acute upper GI hemorrhage. For vealing a clean-based duodenal ulcer in 1 patient and gas- example, 2 can be used to aid the decision about the need tric and esophageal erosions in the other. In retrospect, for hospitalization,7,10 whereas others apply to level of in- these erosions were likely due to nasogastric tube trauma. patient care,8,10,11,16 timing of endoscopy,5 and length of Both patients rebled within 24 hours of endoscopy. Re- hospital stay.6,12 All these studies5,7-11,16,17 contain vari- peated endoscopy revealed the same clean-based ulcer ables that we tested as candidate predictor variables. in the first patient, whereas the second patient had a Dieu- Probably the most widely known CPR for acute upper lafoy lesion, which was treated endoscopically. Because GI bleeding was created by Rockall and colleagues.6,17 This rebleeding occurred soon after the index endoscopy in CPR uses age, shock, comorbidity, endoscopic diagnosis, both patients (and within 24 hours of admission), it is and SRH to predict the risks of rebleeding and death. It unlikely that these patients would have been “missed” was developed on patients admitted with bleeding and those as rebleeders unless the CPR was used to discharge pa- who bled while hospitalized for other reasons, 2 sub- tients immediately after endoscopy, a use for which it is groups with distinctly different prognoses. It was subse- not intended but one for which it could be tested. quently validated retrospectively on 2531 patients admit- The second, more inclusive CPR missed 7 of 123 pa- ted for acute upper GI hemorrhage,17 of which 744 in the tients. Two of the 7 patients experienced rebleeding and low-risk group (score Ͻ2) had a risk of rebleeding of 4.3% have already been described. The remaining 5 patients and mortality of 0.1%. Although the Rockall CPR effec- developed nonbleeding complications: 1 developed uri- tively stratifies the risks of rebleeding and mortality, the nary catheter–induced Escherichia coli sepsis; 3 devel- scoring is difficult to remember, and it does not consider oped pneumonia manifested by fever, leukocytosis, and the non-GI morbid events that often affect hospital stay. a new infiltrate on chest radiography; and 1 developed The study populations of Rockall et al6,17 and the pres- alcohol withdrawal syndrome. ent study differ in calendar time by several years and in All variables in the 2 CPRs are readily measured and criteria for upper GI hemorrhage. Furthermore, the Rock- available except the APACHE II score. Although these post all rule and the present CPRs measure performance for dif- hoc models without APACHE II scores did not perform ferent outcomes: the Rockall CPR measures discrimina- as well as the originals, they offer potential alternatives and tion for rebleeding and death separately, whereas the are worthy of further development and testing. present CPRs measure discrimination for composite out-

(REPRINTED) ARCH INTERN MED/ VOL 167, JUNE 25, 2007 WWW.ARCHINTERNMED.COM 1295

Downloaded From: https://jamanetwork.com/ on 10/02/2021 comes that include rebleeding, need for surgery, and death. Medical Center and Duke University School of Medicine, Nonetheless, we examined performance of the Rockall Durham, NC (Dr Provenzale and Ms Rose). rule on the present study population. In the validation Correspondence: Thomas F. Imperiale, MD, The Regen- study by Rockall et al,17 the C statistic was 0.72 (95% CI, strief Institute Inc, 1050 Wishard Blvd (RG-6), India- 0.69-0.74) for rebleeding and 0.81 (95% CI, 0.78-0.84) napolis, IN 46202. for death. On this Veterans Affairs cohort, the Rockall Author Contributions: Dr Imperiale had full access to all rule has a C statistic of 0.52 (95% CI, 0.44-0.60) for re- of the data in the study and takes responsibility for the in- bleeding (P=.03 for the difference in C statistics) and 0.64 tegrity of the data and the accuracy of the data analysis. Study (95% CI, 0.48-0.80) for death (P=.03 for the difference concept and design: Imperiale, Dominitz, and Boes. Acquisi- in C statistics). The C statistic for poor outcome 1 was tionofdata: Imperiale,Dominitz,Provenzale,Boes,Rose,and 0.81 (95% CI, 0.73-0.90), which is clinically and statis- Bowers. Analysis and interpretation of data: Imperiale, tically greater than Rockall et al’s values of 0.52 for re- Dominitz, Musick, Azzouz, and Perkins. Drafting of the bleeding and 0.64 for death. manuscript: Imperiale and Bowers. Critical revision of the Despite the development of several CPRs for out- manuscriptforimportantintellectualcontent:Imperiale,Dom- come from acute upper GI hemorrhage, none has achieved initz,Provenzale,Rose,Musick,andPerkins.Statisticalanaly- widespread use. This lack of integration into clinical prac- sis: Azzouz and Perkins. Obtained funding: Imperiale. Admin- tice may be due to a lack of appropriately rigorous vali- istrative, technical, and material support: Dominitz, Boes, and dation,10,18 medicolegal concerns, incongruence with cur- Bowers. Study supervision: Imperiale and Provenzale. rent management,5 imprecision for clinical outcomes, or Financial Disclosure: None reported. lack of ease of clinical use in day-to-day practice.6,10-12,16 Funding/Support: This study was funded by grant IIR For these reasons, we attempted to develop the CPRs de- 98-213 from the Veterans Affairs Health Services Re- scribed herein. Both CPRs perform reasonably well and search and Development Service. are easy to use. However, they have limitations that re- Previous Presentation: This study was presented at the quire comment. First, the population from which they American College of Gastroenterology meeting; Octo- were created was US veterans, 99% of whom were men. ber 13, 2003; Baltimore, Md. This population has unique characteristics that may not generalize to nonveteran populations. Second, although REFERENCES both CPRs performed reasonably well in internal, split- sample validation, such validation is preliminary. More 1. Gilbert DA. Epidemiology of upper gastrointestinal bleeding. Gastrointest Endosc. 1990;36(suppl):S8-S13. rigorous validation is required before these CPRs can be 2. Johanson JF. Curbing the costs of GI bleeding. Am J Gastroenterol. 1998;93:1384- 18 used in clinical practice. Third, the performance of these 1385. CPRs was not perfect. Neither identified all persons who 3. Quirk DM, Barry MJ, Aserkogg B. Physician specialty and variations in the cost of treating patients with acute upper gastrointestinal bleeding. Gastroenterology. had adverse outcomes either directly or indirectly re- 1997;113:1443-1448. lated to GI bleeding. Such imperfection of prediction rules 4. Cooper GS, Chak A, Way LE, Hammar PJ, Harper DL, Rosenthal GE. Endoscopic is a reminder that these tools must be considered as aids practice for upper gastrointestinal hemorrhage: differences between major teach- to clinical judgment, not as substitutes for it. ing and community-based hospitals. Gastrointest Endosc. 1998;48:348-353. 5. Bordley DR, Mushlin AI, Dolan JG, Richardson WS. Early clinical signs identify In summary, we created and preliminarily validated low-risk patients with acute upper gastrointestinal hemorrhage. JAMA. 1985; 2 simple and potentially useful CPRs for patients with 253:3282-3285. acute upper GI hemorrhage. Both CPRs identify a low- 6. Rockall TA, Logan RFA, Devlin HB, Northfield TC. Risk assessment after acute upper gastrointestinal hemorrhage. Gut. 1996;38:316-321. risk group in which a shortened hospital stay and per- 7. Lai KC, Hui WM, Wong BCY, Ching CK, Lam SK. A retrospective and prospective haps outpatient management may be considered. Sub- study on the safety of discharging selected patients with duodenal ulcer bleed- sequent research should include further validation of ing on the same day and endoscopy. Gastrointest Endosc. 1997;45:26-30. these CPRs, particularly in nonveteran populations, 8. Corley DA, Stefan AM, Wolf M, Cook EF, Lee TH. Early indicators of prognosis in upper gastrointestinal hemorrhage. Am J Gastroenterol. 1998;93:336-340. and an evaluation of the effect of presenting informa- 9. Hay JA, Lyubashevsky E, Elashoff J, Maldonado L, Weingarten SR, Ellrodt G. tion about risk to managing physicians to determine Upper gastrointestinal hemorrhage clinical guideline: determining the optimal length whether and how such information affects clinical deci- of stay. Am J Med. 1996;100:313-322. 10. Blatchford O, Murray WR, Blatchford M. A risk score to predict need for treat- sion making. ment for upper-gastrointestinal haemorrhage. Lancet. 2000;356:1318-1321. 11. Zimmerman J, Siguencia J, Tsvang E, Beeri R, Arnon R. Predictors of mortality Accepted for Publication: March 2, 2006. in patients admitted to hospital for acute upper gastrointestinal hemorrhage. Scand J Gastroenterol. 1995;30:327-331. Author Affiliations: Department of Medicine, Roude- 12. Hay JA, Maldonaldo L, Weingarten SR, Ellrodt G. Prospective evaluation of a clini- bush Veterans Affairs Medical Center (Drs Imperiale and cal guideline recommending hospital length of stay in upper gastrointestinal Perkins, Mss Bowers and Musick, and Mr Azzouz), The hemorrhage. JAMA. 1997;278:2151-2156. Center of Excellence on Implementing Evidence-Based 13. Hosmer DW Jr, Lemeshow S. Applied Logistic Regression. New York, NY: John Wiley & Sons; 2000. Practice (Dr Imperiale), Indiana University School of 14. Lemeshow S, Hosmer DW Jr. A review of goodness of fit statistics for use in the Medicine (Drs Imperiale and Perkins, Mss Bowers and development of logistic regression models. Am J Epidemiol. 1982;115:92-106. Musick, and Mr Azzouz), and The Regenstrief Institute 15. Braitman LE, Davidoff F. Predicting clinical states in individual patients. Ann In- tern Med. 1996;125:406-412. Inc (Dr Imperiale), Indianapolis; Department of Medi- 16. Schein M, Gecelter G. APACHE II score in massive upper gastrointestinal haem- cine, Veterans Affairs Puget Sound Health Care System, orrhage from peptic ulcer. Br J Surg. 1989;76:733-736. Northwest Health Services Research and Development 17. Rockall TA, Logan RF, Devlin HB, Northfield TC. Selection of patients for early Center of Excellence, and the University of Washington discharge or outpatient care after acute upper gastrointestinal haemorrhage. Lancet. 1996;347:1138-1140. School of Medicine, Seattle (Dr Dominitz and Mr Boes); 18. Justice AC, Covinsky KE, Berlin JA. Assessing the generalizability of prognostic and Department of Medicine, Durham Veterans Affairs information. Ann Intern Med. 1999;130:515-524.

(REPRINTED) ARCH INTERN MED/ VOL 167, JUNE 25, 2007 WWW.ARCHINTERNMED.COM 1296

Downloaded From: https://jamanetwork.com/ on 10/02/2021