Predictive model of mortality in patients with spontaneous bacterial peritonitis

Background Hospital mortality in patients with spontaneous bacterial peritonitis (SBP) is high despite albumin treatment, particularly in those with worse liver and/or renal function. Aim To determine the independent predictive factors of in-hospital mortality and to create and validate a predictive model of mortality in patients with SBP. Methods We analysed all cirrhotic patients with high-risk SBP (serum urea >= 11 mmol/L and/or serum bilirubin > 68 mu mol/L) between 2001 and 2011. We developed a predictive model of in-hospital mortality and validated this in a different cohort. Results We included 118 high-risk SBP episodes treated with antibiotics and albumin. In-hospital mortality was 33/118 (28%). The independent predictive factors of in-hospital mortality at SBP diagnosis were serum urea, blood leucocyte count, Child-Pugh score and mean arterial pressure. A predictive model including these four variables showed a discrimination accuracy (AUC) of 0.850, 95% CI 0.777-0.922. A cut-off point of 0.245 showed a sensitivity of 0.85 and specificity of 0.75. The in-hospital mortality was 28/49 (57.1%) in patients with a model value >= 0.245, and 5/69 (7.2%) in patients with a model value < 0.245 (P < 0.001). The validation series included 161 patients with an in-hospital mortality of 40/161 (24.8%), 30/77 (39.0%) in patients with a model value >= 0.245, and 10/84 (11.9%) in those with a model value < 0.245 (P < 0.001). Conclusions We developed and validated a predictive model of mortality that includes serum urea, blood leucocyte count, Child-Pugh score and mean arterial pressure in high-risk patients with spontaneous bacterial peritonitis. These findings may help to identify patients who would benefit from additional therapeutic strategies.