Use of a Precision Antibiotic Therapy (PAT) Prediction Model to Identify Multidrug-Resistant (MDR) Enterobacteriaceae. (4th October 2017)

• Record Type: Journal Article
• Authors: Varga, Alexandra; Cressman, Leigh; Lautenbach, Ebbing; Cluzet, Valerie; Tolomeo, Pam; Bilker, Warren; Hamilton, Keith
• Is Part Of: Open forum infectious diseases. Volume 4(2017)Supplement 1
• Page Start: S328
• Page End: S328
• Publication Date: 2017-10-04
• Subjects: Communicable diseases -- Periodicals; Medical microbiology -- Periodicals; Infection -- Periodicals; 616.9
• DOI: 10.1093/ofid/ofx163.775
• Languages: English
• ISSNs: 2328-8957

Abstract

Abstract: Background: Emergence of multidrug-resistant (MDR) Enterobacteriaceae complicates the selection of empiric antibiotic therapy. Software called Precision Antibiotic Therapy (PAT) (Teqqa, LLC; Jackson, WY) operationalizes a predictive model using patient factors to make real-time, personalized predictions of antibiotic susceptibility for each antibiotic, allowing prescribers to choose empiric therapy for patients at risk for resistant infections. The purpose of this study was to determine the performance of PAT software in identifying MDR Enterobacteriaceae bloodstream infections (BSI) as well as to determine optimal thresholds of predicted antibiotic susceptibility to choose a broader-spectrum antibiotic. Methods: We conducted a retrospective cohort study including 475 unique patients with BSIs caused by Enterobacteriaceae from January 1, 2016 through December 31, 2016. First-line antibiotic therapy for BSI was defined as cefepime, piperacillin-tazobactam, levofloxacin, or aztreonam. Susceptibilities predicted by PAT were compared with known susceptibilities determined by routine laboratory testing. PAT thresholds for broadening antibiotics were assessed when predicted susceptibilities were 80%, 85%, 90%, and 95% using receiver-operating characteristic (ROC) curves. Performance characteristics were calculated for each threshold. Brier score calculations were then used to compare the accuracy of PAT predictions using the optimized predicted susceptibility threshold, to that of aggregate institutional susceptibility data. Results: ROC curve analysis demonstrated an area under the curve of 0.82 for the 95% threshold. The sensitivity for the PAT prediction utilizing the 95% threshold was 91.7% with a specificity of 74.3%. The Brier score for the 2016 antibiogram to determine antibiotic therapy was 0.085, whereas the Brier score using PAT software was 0.071, representing a 16% improvement in accuracy. Conclusion: PAT software demonstrated excellent capability to discriminate between Enterobacteriaceae BSIs resistant and susceptible to first-line therapy. A predicted susceptibility threshold of 95% should be used to indicate a need for escalation of empiric antibiotic therapy using PAT. Disclosures: All authors: No reported disclosures.