Patients’ baseline characteristics and outcomes
A total of 2259 cirrhotic patients who admitted for acute GIB were selected, among whom 20 cases were excluded for incomplete data. Overall, 2239 records were included in this analysis. The mean age was 53.0±13.1, 1568 (70%) were male. Except for 273 (12.2%) cases without identified etiology, the other etiologies for cirrhosis were: 1385 (61.9%) HBV infections; 264 (11.8%) alcoholic liver diseases; 159 (7.1%) autoimmune liver diseases; 82 (3.7%) HCV infections; 26 (1.2%) secondary cholestatic liver diseases; 15 (0.7%) vascular disorders, 35 (1.6%) others. According to the Child-Pugh classification, there were 636 (28.4%) patients classified as class A, 1096 (49%) as class B , and 507 (22.6%) as class C. PVT and HCC were present in 407 (18.2%) and 414 (18.5%) patients, respectively. In terms of the rebleeding sources, there were 1960 (87.5%) cases with upper GI tract bleeding and 71 (3.2%) with lower GI tract bleeding (jejunoileal or colonic lesion). There were 208 (9.3%) patients who did not receive endoscopy interventions because of massive hemorrhage with rapid deterioration conditions, or patients avoid endoscopic interventions. Variceal lesion causes bleeding in 1901 cases (84.9%), among which 44 cases were concomitant with healing ulcer lesion. Non-variceal lesion only causes 124 cases (5.5%) of bleeding. In variceal bleeding, balloon tamponade was initially applied in 273 cases to control active bleeding, 33 emergent transjugular intrahepatic portosystemic shunt (TIPS) were performed as rescue therapy. During hospitalization, 691 (30.9%) patients received endoscopic treatment, among whom 16 cases underwent additional radiological interventional treatment due to the failure of control bleeding, while 535 (23.9%) patients only underwent radiological interventional treatment.
Overall, there were 245 (10.9%) patients for whom in-patient rebleeding occurred, with an average of 12.3±6.6 hospitalization days. As regards the in-hospital mortality, a total of 135 (6%) patients died during hospitalization and 110 of them experienced in-patient rebleeding. The reasons of deaths were as followed: hemorrhagic shock (n=95), hepatorenal syndrome (n=18), hepatic encephalopathy (n=7), liver failure (n=6), infection (n=4), cerebrovascular event (n=3), metastasis (n=1), acute myocardial infarction (n=1).
The association between albumin/RBC infusion for in-patient rebleeding
Univariate variables association with rebleeding were displayed in Table 1. Patients suffered rebleeding were more likely to have HCC (17.6% vs 26.1%, p=0.002), higher Child-Pugh classification (Child A 30.6% vs 10.6%，Child B 48.8% vs 50.2%, Child C 20.6% vs 39.2%, p<0.001), higher bilirubin level (32.5±39.7 vs 47.8±78.8, p<0.001), lower albumin level (30.8±6.0 vs 28.2±6.6, p<0.001), and deteriorated prothrombin time at baseline (16.3±8.8 vs 17.7±8.6, p=0.026). The risk of rebleeding also varies depending on the bleeding sources, with more rebleeding occurs in cases with unidentified lesions (8.6% vs 14.7%, p=0.003) and less in non-variceal lesion (5.9% vs 2.0%, p=0.018). Rebleeding was more likely to occur in patients rescued by balloon tamponade at admission (7.3% vs 52.2%, p<0.001), while subsequent endoscopic treatment effectively prevents rebleeding (32.9% vs 13.9%, p<0.001) (Table 1).
An average of 2.4±3.4 units of RBC and 15.3 ± 36.1g albumin were administrated to all included patients. More total-RBC (1.8±2.5 vs 6.7±6.0 units, p<0.001) and total-ALB (13.6±32.6 vs 29.6±54.9g, p<0.001) were prescribed to patients who had rebleeding, compared to those without rebleeding. Nevertheless, the pre-RBC transfusion was higher in rebleeding patients (1.8±2.5 vs 4.2±4.4 units, p<0.001), while the pre-ALB infusion was comparable (13.6±32.6 vs 12.3±34.1g, p=0.630). Moreover, in patients with Child-Pugh A and B class, both the pre-ALB and pre-RBC infusion were higher in those with rebleeding in contrast with Child-Pugh C patients in which pre-ALB infusion was substantially lower in those with rebleeding (15.2 vs 30.6g, p=0.009). (Table 2)
To explore the association between the dose and risk of rebleeding, the amount of pre-ALB was further classified into two subgroups as ≤40g and > 40g. The use of pre-ALB was not linked to rebleeding in univariate analysis. In the multivariate model adjusted for baseline albumin and Child-Pugh class, however, the use of an increased dose of albumin reduces the risk of rebleeding (OR for ≤40g vs 0g, 0.463 [0.319-0.672], p<0.001; OR for > 40g vs 0g, OR 0.374 [0.226-0.619], p<0.001).(Table 3)
Furthermore, in another multivariate model adjusted for all confounders (HCC, bleed from non-varices lesion, unidentified lesion, the initial application of balloon tamponade, level of bilirubin, albumin and prothrombin time, endoscopic treatments), the pre-ALB infusion was still associated with a lower risk of rebleeding, with a negative dose-effect relationship between albumin use and rebleeding risk observed (adjusted OR for ≤40g vs 0g, 0.385 [0.252-0.588], p<0.001; adjusted OR for >40g vs 0g, 0.295 [0.169-0.514], p<0.001) (Table 3). Other independent factors in this model included Child-Pugh classification (adjusted OR for Child B vs A, 1.819 [1.123-2.948], p=0.015; adjusted OR for Child C vs A, 2.834 [1.613-4.980], p<0.001), bleed form non-variceal lesion (adjusted OR 0.362 [0.137-0.954], p=0.040), rescue therapy with balloon tamponade (adjusted OR 14.001 [10.007-19.589], p<0.001) and endoscopic treatment (adjusted OR 0.339 [0.223-0.516], p<0.001) (Figure 1A).
Regarding RBC infusion before rebleeding, the pre-RBC infusion remained associated with more occurrences of rebleeding in univariate (OR for 4-8 vs < 4 units 3.569 [2.591-4.918], p<0.001; OR for >8 vs <4 units 7.837 [4.633-13.257], p<0.001).In the multivariate analysis adjusted for the confounders above, the risk was positively correlated with the infused RBC units (adjusted OR for 4-8 vs <4 units 1.835 [1.249-2.695], p=0.002; adjusted OR for >8 vs <4 units 4.253 [2.242-8.071, p<0.001) (Figure 1B).
The association between albumin/RBC infusion and in-hospital mortality
In the univariate analysis, the in-hospital mortality was more probable in patients with compromised liver function (i.e. higher Child-Pugh classification, lower levels of albumin, higher values of bilirubin, creatine and prothrombin time), concomitant with HCC, bleed from varices or unidentified lesion, initial rescue therapy with tamponade and occurrence of rebleeding. The uses of endoscopic and radiological interventions, however, decreased mortality risk (Table 4). More total-RBC (2.2±3.2 vs 5.6±5.4 units, p<0.001) and total-ALB (15.0±35.2g vs 21.6±47.6g, p=0.004) infusion were observed in patients who died during hospitalization than in those who recovered. The number of death between patients administrated with ≤ 40g compared with those with ＞ 40g was not statistically different (115/1975 vs 20/264, p=0.15). However, after adjusted for baseline albumin level and Child-Pugh class (OR for ≤ 40g vs 0g, 0.557 [0.366-0.908], p=0.017; OR for > 40g vs 0g, 0.525 [0.304-0.906], p=0.021), there was a significant correlation between total-ALB infusion and mortality. In another multivariate model adjusted for all confounders (Child-Pugh classification, presence of HCC, occurrence of rebleeding, bleed from variceal lesion or unidentified lesion, the initial therapy of tamponade, level of bilirubin, creatine, albumin and prothrombin time, endoscopic and radiological interventions), the infusion of more than 40g albumin decreased the risk of in-hospital mortality (adjusted OR for ≤40g vs 0g, 0.730 [0.375-1.423], p=0.356; adjusted OR for >40g vs 0g, 0.389 [0.180-0.838], p=0.016) (Table 5 and Figure 2A).
On the other hand, the infusion of total-RBC increased the mortality risk in the univariate analysis (OR for 4-8 vs ＜4 units, 2.816 [1.807-4.387, p<0.001; OR for >8 vs ＜4 units, 9.097 [5.628-14.706], p<0.001) and a multivariate model adjusted for baseline hemoglobin level and Child-Pugh class (OR for 4-8 vs ＜4 units, 2.626 [1.613-4.276], p<0.001; OR for >8 vs ＜4 units, 9.559 [5.535-16.509], p<0.001). However, this association between RBC use and in-hospital mortality risk was not observed in another multivariate model adjusted for all confounders described above (adjusted OR for 4-8 vs ＜4 units, 0.735 [0.357-1.517], p=406; adjusted OR for >8 vs ＜4 units, 1.156 [0.522-2.562], p=0.721) (Figure 2B).