Sepsis phenotypes at risk for infections caused by multidrug resistant Gram-negative bacilli: elucidating the impact of sepsis definition and patient case mix on prediction performance

点击翻译按钮获取中文摘要

SUPPLEMENT ABSTRACT Sepsis is a devastating syndrome that represents a leading cause of death, morbidity, and healthcare costs. Its impact is amplified by rising rates of antimicrobial resistance. Improving sepsis outcomes primarily results from prescribing timely antibiotics based on the estimated risk of multidrug resistance (MDR). Artificial intelligence (AI) and machine learning (ML) are data- driven approaches looking for patterns in massive datasets. While the AI/ ML algorithms rapidly advanced and built successful imaging processing applications, the promise of AI/ML in sepsis and antimicrobial resistance research remains largely unfulfilled. The main reasons stem from deficient, inaccessible and poorly labeled clinical data allowing for only a small portion of the electronic health records (EHR) data to be used. More so, clinical narratives such as notes and imaging reports which contain unstructured data elements in free text format are almost never used. Our parent K08 award aims to identify sepsis phenotypes at risk for MDR GNB that will enable better antibiotic prescribing practices and standardize comparisons across hospitals. We propose to accomplish our goal by leveraging big data and using innovative methods such as ML methods. This supplement will strengthen our project by analyzing in detail the barriers to efficiently using EHR data including unstructured data elements and providing data engineering solutions. The objective is to provide the framework for ML use in sepsis research. Demonstrating reproducibility and rigor of our ML methods and making the algorithms and datasets accessible per FAIR and TRUST principles will be responsive to NIGMS and broader NIH priorities. Our aims reflect these priorities: 1) Analyze barriers to use of EHR structured data and provide data engineering solutions for data enrichment, 2) Extract and assess the importance of unstructured data in developing ML sepsis models, and 3) Compare the ML sepsis models using unstructured and structured data VS structured data only and ensure algorithm fairness by testing it across subgroups of interest based on gender and race. We will incorporate clinical data from the 15 hospitals in our healthcare system serving an ethnically and socioeconomically diverse patient population in rural, suburban and urban hospitals. Dr. Vazquez Guillamet has training in Infectious Diseases and Critical Care Medicine and experience in sepsis research. This supplement complements and broadens the initial K08 award. It serves as the natural next step in deepening her expertise in innovative methods. This supplement will provide the opportunity for meaningful collaborations with data scientists with ample expertise in unstructured data methods and data engineers specialized in ML methods. It will help Dr. Vazquez Guillamet to promote clinically applicable algorithms for challenging problems such as sepsis treatment. For this supplement, Dr. Vazquez Guillamet will continue the collaboration with her multidisciplinary team of mentors and add data engineering support. An accredited course in unsupervised machine learning will be added to her career development plan. She will continue her path to becoming an analytics translator at the intersection of clinical medicine and clinical applied informatics. The fertile research environment at Washington University in St. Louis with focus on data availability, the experienced mentorship team now incorporating data engineering expertise and a well-crafted career development plan will enable Dr. Vazquez Guillamet to achieve her long-term goal of becoming an independently funded clinician-investigator utilizing big data to develop applications for risk prediction, surveillance, and outcome comparisons in sepsis and antimicrobial resistance.