A Predictive Model for Avoiding Follow-Up Paracentesis in Spontaneous Bacterial Peritonitis

BACKGROUND: Recent studies suggest that follow-up paracentesis (FuP) in cases of spontaneous bacterial peritonitis (SBP) should only be performed if there is a clinical and/or analytic worsening. We aimed to evaluate which patients with SBP would benefit from the FuP. MÉTODOS: This retrospective study included consecutive patients diagnosed with SBP between January 2011 and June 2018 in our tertiary center. Clinical and analytical data were obtained at baseline and on the third day of antibiotic therapy. An adequate response on the third day of treatment was defined by a decrease of ≥25% in the neutrophil count of the ascitic fluid (AF). RESULTS: Ninety-six episodes of PBE in 75 patients (79% male sex, mean age 61 ± 11 years old) were included. At admission, a higher serum neutrophil count (p = 0.043), a lower level of serum total proteins (p = 0.040), a positive culture in AF (p < 0.001) and a previous diagnosis of diabetes mellitus (p = 0.035) were related to inadequate response (IR). At day 3, acute kidney injury (p = 0.023), C-reactive protein >100 mg/L (p < 0.001), the presence of fever (p = 0.047) and abdominal pain (p < 0.001) were also associated with IR. In multivariate analysis, the presence of respiratory insufficiency (OR = 16.403; 95% CI: 2.315–116.222; p = 0.005) and abdominal pain (OR = 10.381; 95% CI: 1.807–59.626; p = 0.009) at admission, serum white blood cell count >9 × 10⁹ (OR = 5.832; 95% CI: 1.275–26.669; p = 0.023), and CRP >100 mg/L (OR = 5.043; 95% CI: 1.267–20.076; p = 0.022) at day 3 of antibiotic therapy were predictors of IR. The predictive model presented good accuracy [AUROC of 0.893 (p < 0.001)] − a cutoff of 0.090 had a sensitivity, specificity, positive predictive value, and negative predictive value for IR of 97, 46, 83, and 77%, respectively. CONCLUSIONS: The performance of FuP on day 3 after the beginning of empiric therapy should be individualized, according to clinical and analytic variables of this predictive model.