Superiority of urgent vs early endoscopic hemostasis in patients with upper gastrointestinal bleeding with high-risk stigmata

Abstract Background Guidelines recommend that all patients with upper gastrointestinal bleeding (UGIB) undergo endoscopy within 24 h. It is unclear whether a subgroup may benefit from an urgent intervention. We aimed to evaluate the influence of endoscopic hemostasis and urgent endoscopy on mortality in UGIB patients with high-risk stigmata (HRS). Methods Consecutive patients with suspected UGIB were enrolled in three Japanese hospitals with a policy to perform endoscopy within 24 h. The primary outcome was 30-day mortality. Endoscopic hemostasis and endoscopy timing (urgent, ≤6 h; early, >6 h) were evaluated in a regression model adjusting for age, systolic pressure, heart rate, hemoglobin, creatinine, and variceal bleeding in multivariate analysis. A propensity score of 1:1 matched sensitivity analysis was also performed. Results HRS were present in 886 of 1966 patients, and 35 of 886 (3.95%) patients perished. Median urgent-endoscopy time (n = 769) was 3.0 h (interquartile range [IQR], 2.0–4.0 h) and early endoscopy (n = 117) was 12.0 h (IQR, 8.5–19.0 h). Successful endoscopic hemostasis and urgent endoscopy were significantly associated with reduced mortality in multivariable analysis (odds ratio [OR], 0.22; 95% confidence interval [CI], 0.09–0.52; P = 0.0006, and OR, 0.37; 95% CI, 0.16–0.87; P = 0.023, respectively). In a propensity-score-matched analysis of 115 pairs, adjusted comparisons showed significantly lower mortality of urgent vs early endoscopy (2.61% vs 7.83%, P < 0.001). Conclusions A subgroup of UGIB patients, namely those harboring HRS, may benefit from endoscopic hemostasis and urgent endoscopy rather than early endoscopy in reducing mortality. Implementing triage scores that predict the presence of such lesions is important.


Introduction
Gastrointestinal bleeding is a major cause of hospital admissions, incurring significant economic burden to healthcare systems, accounting for >500,000 hospital admissions, 2.2 million days of hospitalization, and $5 billion in direct costs according to a 2018 report in the USA [1]. When upper gastrointestinal bleeding (UGIB) is suspected, endoscopy is undertaken to treat high-risk stigmata (HRS) [2][3][4][5][6][7][8][9]. The ability to predict HRS before endoscopy is a meaningful and important parameter in clinical decisionmaking because patients without HRS could be managed as outpatients and patients harboring HRS need an expedited endoscopy [10][11][12]. Thus we developed and validated the Horibe gAstRointestinal BleedING prEdiction scoRe (HARBINGER), which consists of only three variables to predict HRS and manage suspected UGIB [7,13].
Although endoscopic hemostasis of HRS significantly reduced recurrent bleeding and the need for urgent intervention and surgery, it is unclear whether performing endoscopic hemostasis leads to reduced mortality [14]. Furthermore, although guidelines recommend that all patients with UGIB undergo endoscopy within 24 h, the optimal timing of endoscopy for patients with HRS remains unclear [15].
In this study of patients suspected of having UGIB, we aimed to evaluate the influence of endoscopic hemostasis on overall mortality and identify the optimal timing of endoscopy in patients with HRS.

Study design and patients
All consecutive adult patients with a suspected UGIB were enrolled at three acute care hospitals in Japan (Tama Medical Center between 2008 and 2015, and Keio University Hospital and Saitama National Hospital between 2012 and 2015). Although all clinically suspected cases of UGIB were included (i.e. presence of hematemesis, coffee-ground emesis, nasogastric lavage with blood or coffee-ground material, melena, tarry stools or black stools as determined by a rectal examination or medical history), we excluded patients with post-procedural UGIB (e.g. post endoscopic submucosal dissection). The institutional review board of each hospital approved the study (20130069). Informed consent or an opt-out clause was obtained from each patient.

Outcome and clinical details
The primary outcome was the all-cause mortality within 30 days after a suspected UGIB. We divided patients into two groups: those found to have HRS on endoscopy and those without HRS. One of the following conditions was defined as HRS: (i) when the source of bleeding was peptic ulcer disease, spurting, gushing, oozing bleeding, or non-bleeding visible vessel (Ia, Ib, or IIa in the Forrest classification) was seen [4]; (ii) when the source of bleeding was variceal hemorrhage, active bleeding, or evidence of recent bleeding (e.g. red or white plug) was found; (iii) when other sources of bleeding, not related to peptic ulcer disease or variceal hemorrhage, were identified, we considered HRS to be spurting or gushing bleeding while oozing lesions, which resolved spontaneously, were not deemed as such. The evaluation was done after removing an adherent clot if one was encountered.
The determination of HRS was adjudicated by showing the endoscopic images of all encountered lesions to experienced endoscopists who were blind to the clinical information of patients. The cause of death, when it occurred, was captured in the medical record.
To achieve endoscopic hemostasis, one of or a combination of the following techniques was implemented: argon plasma coagulation, electrocautery, clips, band ligation, epinephrine injections, or sclerosants injection. An urgent endoscopy was defined as one performed within 6 h after the presentation of suspected UGIB, while early endoscopy was done >6 hours of presentation [16,17]. We also calculated the HARBINGER and Glasgow Blatchford Score (GBS) at the time of presentation [7,13,18].

Statistical analysis
Categorical variables are expressed in number and percentage, and were compared using the chi-square test or Fisher's exact test. Quantitative numerical variables are presented by mean and standard deviation (SD) or median and interquartile range (IQR), and were compared using Student's t-test or the Mann-Whitney U test. The best cut-off value of endoscopy timing for mortality was calculated by a receiver-operating characteristic (ROC) curve with the greatest sensitivity À(1À specificity). The odds ratio (OR) and 95% confidence interval (CI) of endoscopic hemostasis and endoscopic timing were calculated using univariable and multivariable logisticregression analysis.
Three multivariable logistic-regression models with different confounding factors were used to demonstrate robustness because the number to be adjusted for as a confounding factor was limited due to the small number of mortality events (n ¼ 50). The continuous variables in confounding factors were classified into binary categorical variables according to the existing cut-off in the literature or ROC. It means that we used the cut-off of AIMS65 for age (!65 years), the cut-offs of GBS for systolic blood pressure (SBP) (<90 mmHg) and heart rate (HR) (!100 beats/min), and the cut-off of the randomized-controlled trial (RCT) in the New England Journal of Medicine for hemoglobin level (<7.0 g/dL), but the cut-offs for creatinine (!1.0 mg/dL) were not available in the past literature, so we calculated the optimal cut-off from our current data [18][19][20]. The confounding factors in model 1 were chosen as those that were significantly  The confounding factors in model 3 were other treatments including chosen proton-pump-inhibitors use, blood transfusion, and variceal bleeding. Since pursuing endoscopy 6 or >6 h was not randomly assigned, propensity-score matching was applied to minimize possible confounding and selection biases between the urgent ( 6 h) and early (>6 h) endoscopy groups in patients with HRS [21]. Propensity scores, which have the advantage that there is no limit to the number of confounding factors that need to be accounted for, were calculated with multivariable logistic regression using all the characteristics that could be obtained before performing the endoscopy. The propensity-score model included age, gender, any hepatic disease, congestive heart failure, SBP, HR, index presentation hemoglobin level, blood urea nitrogen, and creatinine. We formed matched pairs between urgent-and early-endoscopy groups using a one-to-one nearest neighbor approach with a caliper width of 0.2 without replacement. McNemar test was performed to evaluate the efficacy of urgent endoscopy [22].  As sensitivity analyses, time cut-offs ( 6 vs 6-12 vs >12 h) [16,23,24], subgroup analysis (variceal-bleeding group and nonvariceal-bleeding group with HRS) and threshold analyses (HARBINGER !2, GBS !12 and SBP <90 mmHg [hemodynamic instability] groups) for UGIB-specific mortality were evaluated [7,13,25]. A two-sided p-value <0.05 was considered statistically significant. All analyses were done using JMP(r) statistical software version 14.1.0 (SAS Institute, Inc., Cary, NC, USA).

Results
The study flow is shown in Figure 1. Of 2,063 consecutive patients with suspected UGIB in three institutions, 97 patients were excluded and reasons for exclusion are provided in Figure 1. The remaining 1,966 patients were included in the analysis; the mean age in the cohort was 68.6 years (SD, 15.5 years) and 1,307 (66.5%) were males. The etiologies of UGIB are shown in Table 3. The median duration from the first clinical examination to the initiation of endoscopy was 3.8 h (IQR, 2.0-6.0 h). Fifty patients perished (2.5%) within 30 days from presentation, of which 19 deaths (1.0%) were related to UGIB. Table 1 shows the differences in characteristics between patients who survived and those who perished. Table 2 shows differences in characteristics, treatments, and outcomes of patients who underwent urgent vs early endoscopy.

HRS group
Among 1,966 patients, 886 had HRS, of whom 35 (3.95%) patients perished (19 events were related to bleeding and 16 to other causes). In this group of patients with HRS, the relationship between endoscopic timing and mortality was assessed, and the best cut-off value to achieve the lowest mortality was 5.  (Table 4). After adjusting for confounding factors in three models of multivariable analysis, both endoscopic hemostasis and urgent endoscopy were significantly associated with lower overall mortality (Table 4).  After propensity-score matching, 115 matched pairs were generated from 769 patients in the urgent-endoscopy group and 117 patients in the early-endoscopy group in a 1:1 manner. There were no significant differences in characteristics between the two matched groups ( Table 5). The mortality in the urgentendoscopy group was significantly lower than that in the earlyendoscopy group (2.61% vs 7.83%, P < 0.001).
Of 730 patients with GBS !12, HRS was found in 427 (58.5%). After adjusting for confounders (model 1) in multivariable logistic analysis, urgent endoscopy was not significantly associated with mortality and UGIB-specific mortality.
Of 389 patients with SBP <90 mmHg (hemodynamic instability), HRS was found in 279 (71.7%). After adjusting for confounders (model 1) in multivariable logistic analysis, urgent endoscopy was not significantly associated with mortality and UGIB-specific mortality.

Discussion
In this large multicenter cohort of Japanese patients with suspected UGIB, we found that performing endoscopic hemostasis and urgent endoscopy ( 6 h from presentation) were factors associated with lower overall mortality compared with early endoscopy (>6 h), but only in those patients later found to harbor HRS. In contrast, those patients without HRS did not benefit from this urgent intervention. In addition, the mortality rate was significantly lower with urgent endoscopy either in multivariate logistic-regression analysis or in the propensity-scorematched analysis. Although the P-value of the propensitymatching score is 0.047, the results are likely robust and unlikely attributed to chance because all analyses showed a significant difference in the influence of urgent endoscopy significantly reducing mortality. In a subgroup analysis, urgent endoscopy was significantly associated with lower mortality rates in patients with non-variceal bleeding. Although urgent endoscopy has been recommended for patients with variceal bleeding with HRS, this result may support that the patients with non-variceal bleeding may also benefit from urgent endoscopy when HRS are suspected to be present [5].
HRS are defined by well-established endoscopic findings indicating the need for endoscopic hemostasis by international consensus statements [2][3][4][5][6][7][8][9]. To the best of our knowledge, although improved outcomes, other than death, were reported in patients with HRS, this is the first study to show an association between endoscopic hemostasis and decreased all-cause mortality [14]. When contrasted with previous studies, our findings may be different due to the relatively short interval in To date, there are no RCTs designed to randomly allocate endoscopy timing in UGIB patients based on their likelihood of harboring HRS [26]. Nonetheless, in certain high-risk groups (e.g. GBS >7, patients with frankly bloody nasogastric output), one observational study and one RCT showed that earlier endoscopy ( 6 vs 6-48 h and 12 vs >12 h, respectively) was significantly associated with better outcomes [16,23]. On the other hand, a large cohort study suggested that urgent endoscopy 6 h in hemodynamically unstable patients may worsen   mortality [27] and a recent RCT showed no benefit from urgent endoscopy 6 h in a high-risk population (GBS !12) of patients with UGIB [17]. These discrepancies between studies may be due to the variability in the targeted high-risk groups. For example, the GBS was designed to predict composite outcomes (e.g. need for intervention/transfusion, mortality) and was not intended to predict the presence of HRS. Indeed, of the enrolled 516 patients with GBS !12, 280 (54.3%) required endoscopic treatment [17]. It is reasonable to deduce that urgent endoscopy could not improve the prognosis in about half of patients with GBS !12 who did not require endoscopic hemostasis. In this context, in our current study, urgent endoscopy 6 h was not associated with a statistically significant improvement in allcause mortality or UGIB mortality. We surmise that more cases may be needed to detect a significant difference in mortality for urgent endoscopy when choosing GBS !12 as a high-risk group of interest. The HARBINGER, which has been recently validated, is the first simple score incepted to predict the presence of HRS [7,13]. The HARBINGER (AUC ¼ 0.78) had significantly high accuracy of predicting HRS compared with the GBS (AUC ¼ 0.68) [13]. In our current study, patients with HARBINGER !2 included 66.3% (735/1108) patients with HRS and those had benefited from urgent endoscopy 6 h in reducing UGIB-specific mortality compared with early endoscopy (>6 h), albeit not from an all-causemortality standpoint. This finding may support that patients with HARBINGER !2 should receive urgent endoscopy [7,13]. However, HRS were found in 58.5% (427/730) of patients with GBS !12 and they had not benefited from urgent endoscopy in reducing UGIB-specific mortality and all-cause mortality. Moreover, the HARBINGER !2 (56.4%, 1108/1966) accounted for a higher proportion among patients with suspected UGIB compared with GBS !12 (37.1%, 730/1966). Thus, the HARBINGER may be a better stratification tool for RCTs than GBS when patients are allocated to undergo urgent endoscopy.
Our study has several shortcomings, inherent in the lack of randomization and the restriction of our demographics to Japanese patients. To ascertain whether mortality is truly reduced as a result of urgent intervention in patients with HRS, a powered RCT is needed in which patients are randomized according to the HARBINGER. However, ethical considerations may not permit the randomization of such high-risk groups.
In addition, it is almost certain that there are other confounders not controlled for our analyses. For example, we were unable to calculate for the American Society of Anesthesiologists score for each patient as underlying conditions and the size/number of ulcers was not described in our database. We attempted to compensate for such shortcomings by adjusting and controlling for not only the confounding factors identified from the analysis of this study data, but also those identified in the literature, including congestive heart failure and any hepatic disease as important underlying diseases in UGIB and by other offered treatments (e.g. use of protonpump inhibitors, blood transfusions). Further, we were able to describe the etiology of UGIB (Table 4) accurately. All sensitivity analyses were consistent in showing the benefit of urgent endoscopy and endoscopic hemostasis after adjusting for confounding factors including variceal bleeding. In the varicealbleeding group, urgent endoscopy showed a trend toward lower mortality but did not differ significantly, which likely relates to the small sample size.
In conclusion, both endoscopic hemostasis and urgent endoscopy within 6 h from the presentation were significantly associated with lower all-cause mortality in patients who harbored HRS compared with early endoscopy (conducted >6 h) in a large multicenter Japanese cohort. In patients with HARBINGER !2, urgent endoscopy was significantly associated with reduced mortality from UGIB. Future studies should investigate whether UGIB patients' triage, according to the HARBINGER, will accrue a mortality benefit when urgent endoscopic hemostasis is pursued.