Background: Antibiotics are known risk factors for the development of clinical infections caused by multidrug resistant organisms. However, studies use a wide-range of variables to determine associations. We aim to ascertain the impact of these different exposure variables, including novel variables, on their association with clinical outcomes.
Methods: We used data from a retrospective cohort study in which antibiotics were considered exposure variables and colonization with carbapenem-resistant Acinetobacter baumannii (CRAB) was deemed as the outcome variable. Daily antibiotic variables were constructed for each day of hospitalization containing the number of grams (and DDDs) received for each antibiotic. For the purposes of this study we only analyzed carbapenems and vancomycin. Hazard ratios were calculated using time-dependent and time-independent variables as per Table 1. Novel ways to classify the exposure variables included: 1.Exposure windows: only a period of time (e.g. 3 days) was considered as the exposure. These windows could be entered in the model as binary (any antibiotic received during the window) or continuous variables (sum of antibiotics received during the window). 2. Daily exposures: only the exposure within a single day (either binary or continuous values) was entered in the model in order to calculate the immediate hazards. Both of these variables were allowed to move over time within the model. Analyses were done using macros in SAS 9.3 (Cary, NC).
Results: The effect of carbapenems was evident using binary, time dependent, cumulative exposures (in DDDs), 3-day and 14-day exposure windows. No effect was seen using cumulative exposure in grams, sum of total exposure, and 21-day windows. Vancomycin was only associated with the outcome when time dependent variables were used.
Conclusion: The way we enter antibiotic exposures within our statistical models markedly affect the significance of our associations. Further studies should be done in order to determine which variables should be used to measure the impact of antibiotics on clinical outcomes.
L. S. Munoz-Price,
S. Tarima, None
See more of: Poster Abstract Session