C28-20 Empiric Antibiotic Strategies and 30-Day Mortality for Patients With Community-Acquired Pneumonia: A Decision Tree Analysis of 124 U.S. Veterans Affairs Medical Centers

Abstract Rationale Recent ATS guidelines suggested withholding empiric antibiotics for patients with mild pneumonia without comorbidities and a positive viral test, although comparative evidence is lacking. We leveraged increases in withholding antibiotics for patients with possible pneumonia across the VA to compare the estimated effects of four antibiotic treatment strategies on 30-day mortality: (1) antibiotics-for-all, (2) antibiotics-for-none, (3) ATS-guideline-based strategy, and (4) an individualized treatment strategy, using causal inference/decision tree techniques. Methods We identified all emergency department (ED) encounters at 124 VA Medical Centers 6/1/2021-1/1/2025 with an initial diagnosis of possible pneumonia and positive chest imaging, assessing antibiotic prescriptions within 24 hours of ED-arrival. We assumed the decision to prescribe antibiotics was conditionally independent given 89 covariates representing demographics, co-morbid conditions, initial vital signs, lab values (i.e., no unmeasured confounders). We used these factors to build a Bayesian Additive Regression Tree (BART) to obtain predicted probabilities of 30-day mortality under treatment with and without antibiotics and following the ATS guideline (antibiotics-for-all except for mild pneumonia without comorbidities and a positive viral test) (Step 1). To obtain an individual treatment strategy, we identified the optimal treatment from Step 1 to build decision trees, using a Bayesian Classification and Regression Tree (B-CART) algorithm, based on 10 clinically relevant factors including illness severity, comorbidities, and viral positivity. Results Among 90,247 ED encounters with possible pneumonia and a positive chest image, 80% received antibiotics within 24 hours; 30-day population mortality was 7.7%, ranging from 2.2% with mild pneumonia to 30.8% in shock. Crude mortality was 8.0% in patients receiving antibiotics versus 6.1% not receiving antibiotics. The counterfactual mortality was higher overall under an antibiotics-for-none strategy compared to antibiotics-for-all: 8.9% (95% CI:8.2%-9.7%) versus 7.6% (95% CI:7.4%-7.8%). Following the ATS guideline, the counterfactual mortality was similar to estimates under an optimized individual treatment strategy as well as observed mortality, but would increase antibiotics overall (Table). Conclusion Initial antibiotics in patients with possible pneumonia and positive chest imaging have an overall benefit on population 30-day mortality. Withholding antibiotics per ATS 2025 guidelines appears to have minimal mortality risk. However, antibiotics were withheld in a larger proportion of patients in practice compared to those identified by the ATS guideline or optimal decision tree analysis. Given the low 30-day mortality risk, particularly in mild pneumonia, additional outcome measurements to more comprehensively evaluate the safety and impact of different antibiotic strategies are needed. This abstract is funded by: Veterans Affairs