Increasing antimicrobial resistance (AMR) is one of the most urgent public health threats. In 2019, the Center for Diseases Control and Prevention (CDC) estimated that infections with AMR affect at least 2.8 million people and are associated with at least 35,900 excessive deaths annually in the US. This threat is particularly problematic among Gram-negative rod (GNR) pathogens in which high-rates of resistance to last-line antimicrobials have emerged globally, while our efforts to develop new antimicrobials have stumbled. To decelerate the emergence of AMR among GNR pathogens, it is essential to guide clinicians away from choosing unnecessarily broad-spectrum antimicrobials. An antibiogram, a facility-level summary of antimicrobial susceptibility data, is a common local reference tool which clinicians use when choosing empiric therapy. However, antibiograms have major limitations. First, little is known about how clinicians are currently using them when making empiric therapy decision. Second, antibiogram data is aggregated at the facility-level, and data may be skewed based on the type of practice or geographic area. Lastly, but most importantly, an antibiogram does not consider any patient-level factors. Therefore, there are strong, and pressing needs to understand 1) how an antibiogram is used by the clinician, 2) how much an antibiogram reflects the risk of AMR for individual patients and 3) how we can overcome limitations of antibiogram to optimize empiric therapy and reduce AMR. The overall goal is to create a novel, real-time personalized antibiogram (?Smart Antibiogram?) to overcome current limitations of antibiogram and to optimize clinician choice of empiric therapy for GNRs by providing a ?predicted risk of AMR? based on a machine learning model incorporating patient- and facility-level data. This goal will be accomplished through (a) Master of Science in Health Informatics coursework, (b) a Mentorship Advisory Committee, (c) carefully selected conferences and workshops, and (d) a mentored research study.
Our specific aims are to (1) Characterize the current use of antibiograms in clinical practice and measure the acceptable risk of resistance when clinicians make empiric therapy decisions for Gram-negative bloodstream infections and urinary tract infections within diverse clinical settings; (2) Assess the accuracy of currently-used antibiograms to predict the risk of resistance for individual patients in a large retrospective microbiology cohort for GNR infections; (3) Develop a machine learning model to predict the individualized risk of AMR for patients infected with GNR pathogens and validate prospectively and externally. This will lead to the future development of a personalized decision support tool (?Smart Antibiogram?). The expected outcomes of this AHRQ K08 Award will be the comprehensive understanding of the effectiveness and limitations of antibiogram, and the informatics toolkits to develop Smart Antibiogram. At the end of this K08 Award, the candidate will be well-prepared to become an independent investigator with expertise in AMR and health informatics, with specific strength in AMR prediction model decision-support system.
Antimicrobial resistance (AMR) among Gram-negative rod organisms is one of the most urgent public health threats in the United States and globally. By using series of survey of clinicians, retrospective cohort analyses, and machine learning prediction analytics, this study elucidates the current perceptions for AMR among providers and creates a machine learning model to accurately predict AMR risk at patient-level. The findings from this study will help create a decision-support system to guide clinicians for appropriate empiric therapy to treat infections due to Gram-negative rod organisms and prepare the applicant for a career as an independent infectious diseases health service investigator.
