Epidemiologic Methods: Resistant Nosocomial Infections

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): The growth of antimicrobial-resistant nosocomial infections (ARNI) necessitates the identification and widespread implementation of effective interventions to reduce their incidence, such as changes in prescribing and improvements in infection control. Candidate interventions are often identified by observational studies of modifiable risk factors for ARNI; candidates are then evaluated in clinical trials. Existing methods for both observational studies and clinical trials assume that patient outcomes are independent of each other. This is not true for ARNI, because pathogens are transmissible, so infection of one host may make others more likely to be infected; similarly, use of antibiotics by others in the hospital can increase an individual's risk of ARNI, even if s/he has not received the drug. We have shown that nonindependence is common in ARNI data, obscures the mechanistic effects of antimicrobial use on the incidence of ARNI, and can lead to false results (negative or positive) when interventions are assessed; thus, there is an emerging consensus on the inadequacy of many existing studies and the need for better methods. We will develop and test methods for observational studies and clinical trials that account for nonindependence of patients. For observational studies, we will use data from the University of Utah (UU) to assess simultaneously the effects of individual antibiotic use and total hospital-wide use on risk of ARNI. For clinical trials, we will develop three methods for evaluating interventions while accounting for nonindependence. We will test these methods on real data from UU and the CDC/Emory ICARE project, and on simulated data, for their fit to data, ability to detect effective interventions, and ability to avoid false positive detection of intervention effects that are not real. Methods will include an auto regressive negative binomial model, which is easily implemented in standard software, and more sophisticated approaches, such as hidden Markov models. We will identify methods that perform well on data and will reliably determine the effectiveness of interventions. Dissemination of the results of these studies via peer-reviewed publications, free distribution of software, didactic seminars and future work with specific collaborators will aid in the reliable identification of candidate interventions and trustworthy ways to assess whether these interventions work. This will in turn lead to better practices to reduce the incidence of ARNI.