Measuring and Predicting Appropriate Antibiotic Use to Combat Resistant Bacteria

点击翻译按钮获取中文摘要

Project Summary: Measuring and Predicting Appropriate Antibiotic Use to Combat Resistant Bacteria Antimicrobial resistant infections already cause over 2.8 million illnesses and 24,000 deaths per year in the US alone. The Centers for Disease Control and Prevention (CDC) identify antibiotic prescribing stewardship as the most important action to slow resistant infections. Our objective is to produce the methods for clinical decision support systems to reduce both over and under use of broad-spectrum antibiotics. We will test novel methods to measure and predict better antibiotic choices on urinary tract infections (UTIs), the most common human bacterial infection that accounts for 25- 50% of antibiotic prescriptions with resistance already exceeding 20% for common antibiotics. The key challenge is that prescriptions for antibiotics are almost always guesses before definitive test results are available. This actionable, arbitrary, and ascertainable process where an important decision (antibiotic prescribing) depends on humans predicting a verifiable result (diagnostic culture results) is ideally suited for innovative machine learning that can produce Personalized Antibiograms that predict antibiotic susceptibility for individuals based on patterns learned from large collections of prior examples. Major scientific barriers to progress in combating antibiotic resistant bacteria include the limited personalization of conventional tools for prescribing guidance, overly optimistic retrospective evaluations of predictive models, and the lack of measures for effective diagnostic antibiotic prescribing decisions. With the combined expertise of our multi-site team (Stanford, UT Southwestern, Harvard), we will overcome these barriers and achieve the objectives of this proposal through the following aims: (1a) Multi-site data harmonization and sharing of electronic health records for suspected UTIs (1b) Develop and validate Personalized Antibiogram prediction models for microbial culture results (2) Prospective validation of antibiogram models with real-time electronic health record integration (3) Develop and validate automated methods for electronic phenotyping UTIs (4) Develop and validate a measure of antibiotic appropriateness and desirability