Background: The COVID-19 pandemic exposed critical failures in the management of pneumonia. Pneumonia is the leading cause of death from infectious diseases, resulting in over 20,000 hospitalizations and thousands of deaths across the VA system each year. For the past thirty years, the mainstays of treatment have been antibiotics and supportive care, with little recognition of viral pathogens and the host immune response. Our reliance on antibiotics has led not only to overuse and resistance, but also to a stagnation in diagnostic and therapeutic research that left us ill-equipped for the viral pandemic. Significance: The devastation of COVID19 has made it clear that our old models of disease are inadequate for the optimal management of respiratory infection. Existing evidence surrounding empiric treatment in pneumonia is poor, fraught with previous research that has been challenged by heterogeneity and a failure to characterize patients with enough detail to identify beneficial treatment approaches. It is unlikely that more of the same approach will advance care. This proposal contributes to a direction of clinical approach toward a more complex causal model of infection that requires complex solutions. Innovation and Impact: We will use state-of-the-art exploratory mixed methods that integrate EHR data with survey and qualitative data to examine practice change. National analyses will allow for more inclusive and feasible implementation solutions in diverse VA settings. This proposal breaks scientific ground in VA informatics by leveraging variation with state-of-the-art causal inference methods. If we take the opportunity to study new treatment approaches based on more complex clinical assessments, we will take an important step toward developing better treatments in pneumonia and being better prepared for future pandemics. Specific Aims: Aim 1. Describe emerging changes in the empiric use of antibiotic and steroids for pneumonia using national practice data and qualitative interviews. Aim 2. Identify local conditions related to emergent change in the use of empiric antibiotics and steroids using an exploratory mixed-methods design. Aim 3. Identify and evaluate optimized, interpretable, tailored decision trees for empiric antibiotic and steroid treatments in Veterans with pneumonia. Methodology: Our mixed methods approach includes secondary data analyses of patient-, provider-, and setting-level EHR data including treatment decisions and patient outcomes, combined with natural language processing. We will apply mixed effects models to model the changes in selected treatments and outcomes (hospitalization, deaths, secondary infection) between the pre-pandemic and later (July 2021-present) periods, and to characterize heterogeneity in the trajectories of these variables across VA sites. To that quantitative analysis, we will add qualitative data examining changes in VA providers’ cognitive processes of diagnosis and management of pneumonia, including beliefs and norms surrounding treatment. We will conduct configurational analyses and validate our analytic results with our expert advisory group for face validity, feasibility and usefulness. We will then identify a optimized treatment regimes, in the form of interpretable decision trees that minimize 30-day mortality, for empiric antibiotic and steroid use in Veterans with pneumonia using machine-learning-based, causal inference algorithms, coupled with clinical expertise. Next Steps/Implementation: Results will inform recommendations for the management of Veterans with pneumonia that can be integrated with other evidence streams and disseminated by the national program directors in the Advisory Group. We will produce recommendations for implementation strategies of interventions in pneumonia care for Veterans that will be developed and tested in future work. We will also produce recommendations for future research, including (1) pragmatic clinical trials; (2) creation of VHA- approved living guidance for pneumonia care; and (3) decision support and other implementation strategies.
External Links for this Project
Grant Number: I01HX003500-01A2
None at this time.
Infectious Diseases, Lung Disorders, Aging, Older Veterans' Health and Care
TRL - Applied/Translational, Data Science
Best Practices, Care Management Tools, Outcomes - Patient
None at this time.