Validation Of A Clinical Decision Tree To Predict If A Patient Has A Bacteraemia Due To A Beta-Lactamase Producing Organism

Background: In recent years, several scores and algorithms have been developed in order to guide empirical antibiotic treatment in patients with gram-negative bacilli (GNB) bacteraemia according to the risk of extended-spectrum beta-lactamase (BL) producing. Some of these algorithms do not have easy applicability or present some limitations in their validation. The aim of our study was to validate a recently designed decision tree in our prospective cohort of bacteraemia due to gram-negative bacilli. Methods: We prospectively identified and analyzed all bacteraemia due to gram-negative bacilli in adult patients in our centre between January 2015 and December 2016. Previously developed clinical decision tree was used to classify patients in each of the terminal nodes. Patients were classified as BL group according to whether they were producers of any type of BL. The statistical power of the tree was analyzed by receiver operating characteristics (ROC) curve and by calculation of C-statistics. Results: A total of 448 episodes of bacteraemia were included; 132 (29.5%) were BL group; 68 (15.1%) ESBL producing, 43 (9.6%) due to AmpC and 21 (4.7%) isolates of Pseudomonas aeruginosa. The original clinical decision tree was modified according to the results of our multivariate analysis. The modified tree has a sensitivity of 71%, specificity of 92%, predictive positive value (PPV) of 79% and predictive negative value (NPV) of 88% generating an ROC curve with a C-statistic of 0.76. Conclusions: An easy-to-apply clinical decision tree could be used at the exact moment of diagnosis and adjust the empirical antibiotic treatment in patients with gram-negative bacilli bacteraemia.