Variation in Use Metrics by Type of Antibiotic and Among Veterans Affairs Acute Care Medical Facilities

Variation in Use Metrics by Type of Antibiotic and among Veterans Affairs Acute Care Medical Facilities

Background: Antibiotic stewardship in acute care facilities will help control healthcare associated C. difficile infection (CDI). Knowing current patterns of prescribing informs interventions and models of disease transmission. We performed a retrospective cohort study to describe patterns of antibiotic use for four categories of antibiotics and three metrics of use to assess variation among facilities and over time.    

Methods: The cohort included acute care admissions between 11/1/12 and 10/31/13 with day-level information on laboratory tests, treatments, and transfers. There were 299,452 admissions and 1.8 million patient days. Antibiotic use metrics, 1) any antibiotic during an admission (A), 2) days of therapy per patient day (DOT), and 3) any antibiotic per day per patient day (AD). Antibiotics were categorized as 1) treatment for CDI (e.g. oral Vancomycin) and 2) high (e.g. Fluoroquinolones), 3) medium (e.g. Penicillin), and 4) neutral (e.g. Tetracycline) risk for CDI, were analyzed separately. The binary outcome A was fit with a mixed effects logistic regression. Facility was a random effect and age, Charlson Comorbidity Index (CCI), log length of stay (total and ICU), sine, and cosine (for seasonality) were fixed effects. The count outcomes of DOT and AD were fit with negative binomial mixed effects models. Log total length of stay was included as an offset term, with other terms as above.  

Results: The models resulted in similar associations across metrics but differed by antibiotic category for length of stay and seasonality (only medium and high risk antibiotics had significant seasonality). Variation among facilities (estimated density curves) differed by metric and by category. Notably, the high risk AD are highly skewed, indicative of pervasive prescribing.

Conclusion: Antibiotic prescribing patterns are complex, varying by category and among facilities and through the year differently by category. Knowledge of these patterns will provide baseline information for developing interventions. Functions representing these complex relationships will be used in simulation models to determine individual patient prescribing probabilities, increasing realism and improving our understanding of the relationships with CDI.