Upper gastrointestinal (GI) bleeding occurs proximal to the Treitz ligament and may be caused by gastric or duodenal ulcers or gastric cancer (1). The mortality rate of patients with upper GI bleeding ranges from 3.5 to 7.4% (2,3). Upper GI bleeding is diagnosed by endoscopy (4). Upper GI bleeding is treated with endoscopy, such as clipping and bipolar electrocoagulation (4) However, for patients for whom endoscopic treatment fails, interventional radiology is applied (5,6). The mortality rate rises to 40% when patients become hemodynamically unstable (7); therefore, it is important to predict upper GI bleeding prior to this.

The Glasgow Batchford Scores, Modified Early Waning Score and Pre-endoscopic Rockall scores are useful for stratification of patients with regard to unstable state, requirement of transfusion and hospitalization (8,9). These scoring systems are useful for triage and management of upper GI bleeding (10). Analyses of on data from emergency departments or intensive care units identified malignancy, hypotension on admission, low Glasgow coma scale and kidney dysfunction as predictors of poor outcome (11). It would be beneficial to diagnose upper GI bleeding in patients prior to presentation at the emergency department. Moreover, certain patients with upper GI bleeding do not exhibit any symptoms (12). Therefore, it is desirable to diagnose upper GI bleeding prior to the advancement of affected patients to the unstable state.

As blood test variables are easy to obtain and reliable, the present study investigated the predictive value of their changes with regard to upper GI bleeding. The change ratios of blood test variables at the time-point of endoscopy and 3 months previously were evaluated.

To identify blood test parameters whose changes were significantly associated with upper GI bleeding, change ratios were compared between patients without or with upper GI bleeding using one-way analysis of variance (Table III). The change ratios of WBC (P=0.05) and ALT (P=0.037) in patients with upper GI bleeding were higher were higher than in patients without bleeding. Furthermore, the change ratios of Hb (P<0.001), TP (P<0.001) and Alb (P=0.01) in patients with upper GI bleeding were lower that in those without bleeding. These results suggested that the change ratios of WBC, Hb, TP, Alb and ALT were significantly associated with upper GI bleeding.

Logistic regression analysis was performed to identify blood test variables with change ratios that were strongly associated with upper GI bleeding (Table IV). The change ratio of Hb was determined to be significantly associated with upper GI bleeding (P=0.017); whereas no significant correlation was identified for the change ratios of the remaining blood test variables with P-values close to 1. These results suggested that the change ratio of Hb was significantly correlated with upper GI bleeding.

ROC analysis was performed to determine the threshold values of change ratios for the prediction of upper GI bleeding (Fig. 1). A threshold value was determined as a value that was in contact with the line with an increment of 45° (Fig. 1, solid straight line) and had the greatest distance from the reference line (Fig. 1, broken line). Variables obtained via ROC analysis are presented in Table V. The threshold value of the change ratio of Hb was 0.813, which was converted into the threshold of the change rate using the following formula: (1.000–0.813) × 100%=18.7%. Therefore, a 18.7% reduction of Hb over three months was the threshold value for the prediction of upper GI bleeding.

To date, only a limited number of studies have assessed the association of change ratios of blood test variables with upper GI bleeding. It has been previously demonstrated that the change ratio of the Model for End-Stage Liver Disease score is useful for the prediction of bleeding from esophageal varices (14); however, this model is applicable to patients with liver cirrhosis only and was therefore not used in the present study.

Decreased Hb has been previously reported to be associated with upper GI bleeding (15). However, a limitation with regard to evaluation of Hb is that results differ between first-time and repeated sampling (16). Hb below the normal range indicates upper GI bleeding. The present study further assessed the predictive value of the change ratio of blood parameters with regard to upper GI bleeding. In a previous study by our group, a 21.3% reduction of Hb within 3 months was determined to be the threshold value for the prediction of upper GI bleeding (17). In line with this, an 18.7% reduction was determined to be the threshold value for the prediction of upper GI bleeding in the present study. Although the patient cohorts were totally different between the two studies, the threshold was ~20% in each, suggesting that a change ratio of Hb of ~20% or above is of predictive value regarding upper GI bleeding.

A limitation of the present study was the difficulty in differentiating between upper and lower GI bleeding, which may have affected the assessment of the association of the change ratio of Hb with upper GI bleeding (18). To differentiate between upper and lower GI bleeding, an elevated BUN value may be useful (19), as higher BUN indicates severe upper GI bleeding (20). In the present study, BUN was not significantly different between patients with or without upper GI bleeding. This result may suggest that upper GI bleeding was not very severe in the patients.

In conclusion, an 18.7% reduction of Hb over three months was shown to be associated with upper GI bleeding in the present study. These findings suggested that patients with this change ratio of Hb are at risk of upper GI bleeding. It is recommended that patients are subjected to endoscopic examination if presenting with symptoms of upper GI bleeding and Hb levels lowered by ≥20%.