Upper Gastrointestinal Hemorrhage: Development of the Severity Score

Background Emergency endoscopy for every patient with upper gastrointestinal hemorrhage is not possible in many medical centers. Simple guidelines to select patients for emergency endoscopy are lacking. The aim of the present report is to develop a simple scoring system to classify upper gastrointestinal hemorrhage (UGIH) severity based on patient clinical profiles at the emergency departments. Methods Retrospective data of patients with UGIH in a university affiliated hospital were analyzed. Patients were criterion-classified into 3 severity levels: mild, moderate and severe. Clinical and laboratory information were compared among the 3 groups. Significant parameters were selected as indicators of severity. Coefficients of significant multivariable parameters were transformed into item scores, which added up as individual severity scores. The scores were used to classify patients into 3 urgency levels: non-urgent, urgent and emergent groups. Score-classification and criterion-classification were compared. Results Significant parameters in the model were age ≥ 60 years, pulse rate ≥ 100/min, systolic blood pressure < 100 mmHg, hemoglobin < 10 g/dL, blood urea nitrogen ≥ 35 mg/dL, presence of cirrhosis and hepatic failure. The score ranged from 0 to 27, and classifying patients into 3 urgency groups: non-urgent (score < 4, n = 215, 21.2%), urgent (score 4 - 16, n = 677, 66.9%) and emergent (score > 16, n = 121, 11.9%). The score correctly classified 81.4% of the patients into their original (criterion-classified) severity groups. Under-estimation (7.5%) and over-estimation (11.1%) were clinically acceptable. Conclusions Our UGIH severity scoring system classified patients into 3 urgency groups: non-urgent, urgent and emergent, with clinically acceptable small number of under- and over-estimations. Its discriminative ability and precision should be validated before adopting into clinical practice.


Introduction
Upper gastrointestinal hemorrhage (UGIH) is a common challenge encountered in emergency medicine departments. Hospital admissions were approximately 300,000 cases per year in the United States [1]. Case fatalities were also as high as 7-10%, with a yearly expenses of 2.5 billion $US [2].
Endoscopy plays a key role in classifying patients. It is generally suggested that endoscopy should be scheduled within 24 hours after hospital admission [3][4][5][6]. The American Society for Gastrointestinal Endoscopy suggested a somewhat earlier timing, within 12 hours [7], but actual early endoscopic examinations were usually scheduled between 2 to 24 hours [8][9][10][11][12][13]. Almost 80% of UGIH is self-limited [14]. Therefore in most cases endoscopy could be postponed to the following day. This implies that there are only a fraction of patients who actually required an emergency endoscopy. However, this is possible only in hospitals which are 24-hour well-equipped [15]. Clinicians all agreed that endoscopy for patients with mild and moderate UGIH could practically be delayed, and that emergency endoscopy is necessary only for some patients with severe bleeding or those who are in a state of shock. Existing screening procedures for patients with UGIH were mostly based on a scoring system that stratifies patients into those with high or low risk, focusing on ultimate clinical outcomes such as re-bleeding and/or death [16][17][18]. The purpose of such screening was mainly to help clinicians discharge patients with low risk early and safely, to be treated as out-patients, and to selectively admit patients with high risk to an intensive care unit for close monitoring [16][17][18].
Our study focused on developing a simple scoring system to predict UGIH severity, by using patient clinical profiles on arrival at the emergency departments, as had been reported earlier [19]. The scores may be used to identify and discriminate UGIH patients with different severity levels without depending entirely on endoscopic examinations and findings.

Patients and methods
The study was conducted in Kamphaeng Phet Hospital, a university-affiliated tertiary hospital in the northern region of Thailand. We retrieved medical files of patients who present-*P-value from chi-squared for linear trends, or two-way ANOVA by rank. SD: standard deviation; SBP: systolic blood pressure; BUN: blood urea nitrogen.

Indicator parameters
The patient profiles of interest were: 1) Demographic profiles: gender and age; 2) Mode of presentation: hematemesis, coffee ground vomiting, hematochezia, melena, and syncope; 3) Hemodynamic profiles: pulse rate and systolic blood pressure; 4) Biochemical profiles: hemoglobin and blood urea nitrogen, and 5) Co-morbidities: presence of cirrhosis, hepatic failure, cardiac failure and renal failure.

Definitions of UGIH severity: an outcome of interest
We used the following criteria to define UGIH severity: 1) Severe: patients who required surgical interventions to stop bleeding, patients in a state of grade 3 and 4 shock [20], and patients who did not survive; 2) Moderate: patients who required endoscopy to stop bleeding (endotherapy), patients with re-bleeding, patients in a state of grade 1 and 2 shock, and patients who required blood transfusion; 3) Mild: patients with no signs of shock, patients who required endos-copy without hemostasis, and patients who did not required any blood transfusions.

Data analysis
The patient profiles were compared across the three severity groups by chi-squared tests for linear trends, or two-way ANOVA by rank. Significant indicators for UGIH severity were presented by odds ratios from a multivariable ordinal continuation ratio logistic regression, which is most suitable for ordinal-natured outcomes such as in this study. The significant coefficients were transformed into item scores and added up to a single total score for each patient. The discriminative performances of the scores were displayed graphically. The patients were classified by the scores into 3 urgency groups corresponding to their original severity: non-urgent, urgent and emergent. The score-classification of urgency was compared to the criterion-classification of severity.
The study was approved by The Kamphaeng Phet Hospital Ethical Committee for Clinical Research. No patients inform consents were required for this retrospective data collection. Traceable patient information was omitted in all processes of data analysis and presentations. The authors declared no out-source grants received and no conflicts of interests.

Results
We retrieved 1,043 medical files corresponding with the above definitions. Among these, 255 were criterion-classified as mild, 664 as moderate, and 124 as severe.

The scoring system
The above significant coefficients were divided by the smallest coefficient (0.44) and rounded up or down to the nearest 0.5 integers to serve as item scores. Scores were not available for 30 patients with missing information of key variables. The item scores ranged from 0 to 10.5 and the sum (total) scores may range from 0 to 27 (Table 2).

Discriminations
The mean scores for patients in the mild, moderate and severe groups were 3.0 ± 3.5, 8.3 ± 3.3, and 16.1 ± 4.2 respectively (Table 3), and clustered within different distributions (Fig. 1). The derived scores discriminated moderate UGIH from mild UGIH at moderate scores, and discriminated severe UGIH from moderate UGIH at higher scores (Fig. 2).

Calibrations
For each of the 3 levels of severity, the score-predicted percents were calibrated against the criterion-classified percents. The compared percentages agreed correspondingly, yielding P-values (goodness-of-fit statistics) of 0.992, 0.996 and 0.992 for prediction of mild, moderate and severe UGIH respectively, implying no lack-of-fits.

Clinical predictions
The score predicted patients who were at least in a moderate group (moderate or severe vs. mild) with high percentage (area under the ROC curve = 0.8813; 95% CI = 0.8600 -0.9008, data not shown), and predict patients who were in a severe (severe vs. mild or moderate) group with greater proportion (area under the ROC = 0.9274; 95% CI = 0.9092 -0.9422, data not shown).

Discussion
The development of scoring systems for screening patients presenting with UGIH may be classified into 2 groups. The first group used information on clinical and laboratory parameters without endoscopic findings. Examples are the Blatchford Score [21], the Bleed Risk Classification [22,23], and the Clinical Rockall Score. The second group used endoscopic examinations in addition to clinical and labora- tory parameters. Examples are the Complete Rockall Score, the Baylor-College Score [24] and the Cedars-Sinai Score [25]. Advantages or disadvantages of the two scoring systems depend on the patient settings and clinical outcomes to be predicted. The Blatchford Score has high sensitivity in predicting the need for intervention, but its low specificity results in obvious overestimation [26]. The Rockall Score has a good prediction for death, but a poor prediction for rebleeding, or the need for surgical procedures [27].
The existing scoring systems to classify UGIH patients into severity levels, like the Blatchford Score [21] or the Rockall Score, classified patients into only 2 groups, high risk and low risk. Very narrow ranges of the scores seemed to cause some limitations on clinical practice. In the Blatchford Score, patients scoring 0 were classified as the "low risk" [21,28], and those scoring 1 or more as the "high risk". The same rule was also used in the Rockall Score [28,29]. As there are only a small number of patients in the low risk group, the score classified a large number of patients as "high risk". For examples, using the Blatchford Score, there would be 92.1% of patients classified as the "high risk", 81.6% from using the Clinical Rockall Score, and 70.1% by using the Complete Rockall Score [28]. This over classification will results in increasing patient loads and medical resources overuse.
Classifying patients into broader ranges may be more practical for clinicians and surgeons. Our study used 2 cut-off points to classify patients into 3 groups: non-urgent, urgent and emergent. Directive actions are correspondingly suggested as follows: 1) Patients scoring < 4, the "non-urgent" representing "mild" group, had low level of severity. They could be managed conservatively. No blood transfusion may be needed, and elective endoscopy may be appointed in 96 hours to 10 days. These patients may be treated as out-patients. This low risk group correspond to the "low risk" scoring 0 in the Blatchford Score and the Pre-Endoscopic Rockall Score, in which patients were successfully treated as out-patients, the re-bleeding rate was very low and no deaths reported [5,22]; 2) Patients scoring between 4 and 16, the "urgent" representing the "moderate" group, had a higher severity level. These patients should be admitted to hospital with additional interventions. Resuscitation may be needed. Fluid replacement and/or blood transfusion may be given as indicated. Endoscopy should be appointed within 24 -96 hours after admission. Patients in this group comprised approximately 60% of all UGIH patients and corresponded to some patients in the high risk group from the Blatchford Score and the Rockall Score [5,22]; 3) Patients scoring > 16, the "emergent" representing the "severe" group, had the highest severity level. These patients may experience blood pressure drop. Vigorous resuscitation may be needed. They should be admitted to an intensive care unit (ICU) for close monitoring. Endoscopy should be appointed immediately, within 24 hours, or as soon as vital signs are stable. Some patients may require surgical interventions to stop bleeding or for life savings. These patients corre-sponded to the "high risk" group in both the Blatchford Score and the Rockall Score.
Our algorithm classified 21.2% patients as the mild group, similar percentages to the Blatchford Score and the Rockall Score [22], but the number of patients in the high risk group would be diminished from 75-80% by the Blatchford Score [30] and the Rockall Score to only 11.9%. The rest of the patients would be classified as "moderate" or "urgent" group instead (66.9%). This algorithm will cut down the number of patients requiring immediate endoscopy (< 24 hours) to those who were actually severe. The unnecessary cost of care, both medical personnel, medical instruments, and other medical resources for mild and moderate patients could vastly be reduced.
Our scoring system relied on similar clinical and laboratory parameters, without using endoscopic findings, as the Blatchford Score [21] and the Clinical Rockall Score [5]. It should therefore be usable in many settings where endoscopy is unavailable. Primary or secondary care settings may use this scoring system to selectively transfer "severe" patients to the nearest tertiary care settings. Like other scoring systems developed, our score also need external validation to confirm its discriminative ability and precision.

Conclusions
Patient profiles might be combined to develop a simple UGIH severity scoring system, which could classify patients into non-urgent, urgent and emergent groups. Small numbers of under-and over-estimations were clinically acceptable. However, its discriminative ability and precision should be validated with a new group of similar patients.