External validation of the predictive ability of Charson, SOFA, Pitt, INCREMENT-ESBL and bloodstream infection mortality Risk for 30-day-mortality in bacteraemia using the PROBAC cohort data

Abstract

Introduction: The development of predictive mortality scores for bacteraemia is fundamental for identifying patients in whom increasing our management efforts. However, it is necessary to assess the validity of the results obtained when they are applied to new cohorts.

Methods: We evaluated the ability of different scales (Charlson, also age-adjusted Charlson and updated Charlson, SOFA, Pitt, INCREMENT-ESBL and BSIMRS) to predict 30-day mortality in bacteraemia through the AUROC and calibration plots. The scales were applied to specific patient from PROBAC cohort (prospective, multicentre with bacteraemia of any aetiology) according to the population in which the scale was originally developed. We also applied the recently developed PROBAC score (this time applied to the entire PROBAC cohort, rather than only to patients who did not die within 48 h of blood culture collection as in the original development of the scale).

Results: After applying Charlson, age-adjusted Charlson, updated Charlson, SOFA, Pitt and PROBAC to the entire PROBAC cohort, we obtained AUROC values: 0.60 (95% CI: 0.58-0.62); 0.62 (95% CI: 0.60-0.64); 0.60 (95% CI: 0.58-0.62); 0.69 (95% CI: 0.66-0.71); 0.71 (95% CI: 0.69-0.82) and 0.80 (95% CI: 0.79-0.81), respectively. INCREMENT-ESBL was applied only to gram negative bacteraemia yielding 0.81 (95% CI: 0.79-0.82) and BSIMRS to gram negative bacteraemia who received adequate empirical antibiotic yielding 0.72 (95% CI: 0.70-0.75).

Conclusions: Scores that have been developed in bacteraemia cohorts and have been used for the prediction of short-term mortality were found to be better at predicting mortality in our analysis.

Keywords: Bloodstream infection; mortality; prognostic factors, adult, Spain, logistic models.