A Hierarchical Transmission Model Evaluating the Effectiveness of Hospital Infection Control Strategies
Background: Although studies have been done to evaluate the efficacy of hospital infection control strategies, results have been inconsistent and interpretation is complicated by variation in design and implementation. Challenges of modeling data from such studies can be met through the development of flexible frameworks that incorporate implementation specific components.
Methods: We developed a hierarchical transmission model based on a Bayesian statistical framework for modeling nosocomial transmission. Markov chain Monte Carlo was used for estimation which allowed for the modeling of missing data and tractability of estimates. Additionally imperfect testing, importation probability and transmission rate were incorporated in the model for parameter estimation. Transmission rates had a common bivariate log-normal distribution for pre- and post-intervention means in the control ICUs, but the post-intervention mean for ICUs in the intervention arm was modified by an additive intervention effect parameter. We retrospectively analyzed data collected from an intervention study including 18 ICUs, 10 of which were assigned to the intervention arm. The study period was from April 2005 to August 2006, during which time 20,945 patients were admitted to an ICU. Surveillance cultures for methicillin-resistant Staphylococcus aureus were collected on patients weekly, at admission and at discharge.
Results: The model estimated significant variation in importation across the 18 ICUs (Figure 1) and estimates of importation were consistently higher than those of admission prevalence (Figure 2). The intervention effect parameter estimate was 0.02 with 95% CI [-0.45, 0.49] on the log-scale, suggesting no effect, while estimates of the mean transmission rate varied little between the pre- and post-intervention periods (Figure 3).
Conclusion: Modeling importation and transmission did not alter the conclusions of the original study, which showed no intervention effect. Transmission models provide an efficient, flexible framework for parameter estimation, and can be used to analyze infection control interventions. Future work includes incorporating patient-level and ICU-level characteristics, and development of an R package for general use.
A. Redd, None
M. Leecaster, None
T. Greene, None
Y. Zhang, None
W. C. Huskins, None
M. Samore, None