Ahmed El-Tahtawy
Pfizer Inc., USA
Posters & Accepted Abstracts: J Pharmacogenomics Pharmacoproteomics
Introduction & Aim: Catheter-related bloodstream infections (CRBSIs) remain a common challenge in critically ill patients. Predictors of mortality in this population across different treatments have not been well studied. This study was aimed at developing useful prognostic tools and predictive models for relative risk adjustment for mortality in patients with CRBSI. Methods: We used a recent trial data of 731 patients with CRBSIs randomized to drug (x) and vancomycin (VAN). Our mortality analysis plan involved a sequence of specific step; data mining, non-parametric methods and finally parametric (logistic) modeling. Results: Both CART and logistic regression identified MPMS, age, baseline corticosteroid exposure, region of world of enrolling study site and infection with a Gram negative pathogen as the most important factors associated with mortality. Together, these five predictors contained more than 95% of the prognostic information in the clinical data (baseline, developed). Logistic modeling allowed us to combine and investigate the effect of different prognostic variables on mortality. The validated model accurately estimated likelihood of mortality across different patient population with unique characteristics. Conclusions: Appropriate antibiotic therapy remains a key driver of mortality in CRBSI. Efforts to improve outcomes can be facilitated with using a validated predictive models and the use of prognostic tools, like nomograms, to calculate the probability of mortality for any specific patient. The early prognosis would assist clinicians to identify high risk patients and to select the appropriate therapy.
Email: Ahmed.El-Tahtawy@pfizer.com