A predictive model for failure to control bleeding during acute variceal haemorrhage

Variceal bleeding is a frequent complication of cirrhosis and is associated with a high risk of early rebleeding. In patients with peptic ulcers, continued bleeding or early rebleeding are risk factors for mortality and can be predicted by statistical models; however, no such models exist for acute variceal bleeding.

We prospectively evaluated failure to control bleeding in 695 consecutive patients with cirrhosis, admitted for haematemesis and/or melaena. Criteria were defined for failure to control bleeding, which comprised both continued bleeding or early rebleeding within 5 days of admission. There were 2 sequential groups of patients: (i) those with variceal bleeding initially treated with blood transfusion and vasoactive drugs, and if these failed followed by sclerotherapy (n = 385); (ii) those with variceal bleeding treated with injection sclerotherapy at diagnostic endoscopy (n = 144). The third group was those with bleeding from other sources related to portal hypertension (n = 166).

Failure to control bleeding was noted in 169 (44%) patients in group 1, 55 (38%) in group 2 and 44 (25%) in group 3. Twenty variables that were evaluable within 6 h of admission, pertaining to severity of bleeding, severity of type of liver disease, mode of admission, and time of diagnostic endoscopy, were entered into a multivariate Cox model. Independent predictors of early rebleeding in group 1 were: active bleeding at endoscopy (irrespective of interval from admission) (p<0.0001), encephalopathy (p = 0.007), platelet count (p = 0.002), history of alcoholism (p = 0.002), presentation with haematemesis (p = 0.02), log urea (p = 0.03) and (shorter) interval to admission (p = 0.007). The variables predictive of 30-day mortality were: early bleeding (p<0.0007), bilirubin (p = 0.0006), encephalopathy (p<0.0001), (shorter) interval to admission (p<0.0001), and log urea (p = 0.004); a model based on these variables was also a good predictor of mortality in the other 2 groups. However, the model derived from group 1 for failure to control variceal bleeding was different in group 2, despite similar patient characteristics and a similar failure rate (following a single injection). This could suggest that sclerotherapy may induce bleeding in some patients independently of the baseline risk for failure to control bleeding.

In cirrhotic patients who present with haematemesis or melaena, active variceal bleeding at diagnostic endoscopy is predictive of failure to control bleeding (continued bleeding or early rebleeding within 5 days of admission), and this failure is predictive of 30-day mortality.