The optimal timing for urgent endoscopy in nonvariceal upper gastrointestinal bleeding

We found the article of Lim and colleagues of great interest as they try to identify patients who will benefit from endoscopy at the earliest possible opportunity for nonvariceal upper gastrointestinal bleeding (NVUGIB), the “holy-grail” in endoscopy [ 1 ]. The authors demonstrated statistically significant results by re-defining the original Glasgow-Blatchford score (GBS) for patients at low or high risk of needing urgent endoscopy to manage their upper gastrointestinal bleeding [ 2 ]. However, there are aspects of their study that warrant further discussion. First, we were surprised by the authors’ choice of the Student’s t test for evaluating continuous variables. In their study, the distribution of the continuous variables was skewed, thus invalidating this test. The skewed nature of the data was highlighted by the fact that the mean ± 2 SDs gave negative lower values for presentation to endoscopy time (hours), length of admission, platelets, and creatinine. The means were less than twice the SD [ 3 ]. Therefore, the P values for these variables are potentially invalid. Secondly, re-defining low and high risk for in-hospital mortality for NVUGIB by a GBS cut-off of 12 was not supported by their results and may lead to confusion for nongastroenterologist physicians, as the original GBS paper recommended out-patient management for low-risk groups [ 2 ]. There was 4 % mortality in this newly defined low-risk group, which should not be regarded as insignificant. Furthermore, the measured sensitivity for the original GBS low-risk group was 99 % vs. only 40 % in the newly defined group with GBS > 12. The original area under the receiver operating characteristic curve (AUROC) was 0.92 (95 % confidence interval [CI] 0.88 – 0.95) for clinical intervention vs. 0.81 (95 %CI 0.76 – 0.86) for in-hospital mortality, which is no better than Rockall score (0.75)[ 2 ]. The x-axis of the AUROC graph should be labeled “1-specificity,” a small but relevant point with regards to the overall presentation and interpretation of their results. Thirdly, we believe that the most significant contribution to in-hospital mortality was patient co-morbidity, as shown in the logistic regression analysis with odds ratio of 10; however, this did not reach statistical significance because the study was underpowered as demonstrated by the wide 95 % CI (0.54 – 186). This result was consistent with the original Rockall risk assessment paper [ 4 ]. Fourthly, we were surprised to read that local regulation prevented doctors from performing urgent endoscopy for high-risk groups who cannot give consent. The hippocratic oath recommends doctors to always act in the best interest of their patients [ 5 ]. Furthermore, next of kin gives assent not consent. We believe the exact timing for urgent endoscopy for suspected NVUGIB should be determined by the on-call endoscopist, and clinicians should not be distracted by any formal risk scores or neglect prompt resuscitation, reversal of anticoagulation, and frequent monitoring and acting on deterioration of vital signs.