In Spring of 2008, a sample of hospitals belonging to the Centers for Disease Control and Prevention's (CDC) National Healthcare Safety Network (NHSN) (formerly National Nosocomial Infection Surveillance, or NNIS) was recruited. To participate in the study: (i) a hospital must have conducted NHSN device-associated surveillance of HAI in an adult medical, medical/surgical or surgical ICU in 2007 according to the module protocol; [9
] and (ii) the ICU must have had a minimum of 500 device (central intravascular line or ventilator) days. There were 441 hospitals eligible to participate. While some states had mandated membership in NHSN, lists of NHSN hospitals were not public information. To protect the confidentiality of the hospitals, the CDC developed a list of eligible hospitals and e-mailed them directly to invite them to participate by accessing a web-based survey. In the communications, we asked that only one person complete the survey for their institution. The survey was designed to be answered by either a nurse or physician director or manager of the hospitals' department of infection control.
Hospital demographics of respondents were examined to check for duplicates. In the rare instances when duplicate responses were found for a single institution (n = 31), the surveys were examined for completeness of data and role of the respondent. Those surveys completed by directors of departments and/or those in which responses were most complete were used. All procedures were reviewed and approved by institutional review boards at Columbia University, CDC and RAND Corporation.
The development and content of the survey are described in detail elsewhere [10
]. Briefly, the survey was developed building upon the questionnaire used in the Study on the Efficacy of Nosocomial Infection Control (SENIC), in which staffing of infection control programs and intensity of surveillance, prevention and control activities were first measured in US hospitals during the 1970s [11
]. Respondents were asked about ICU-specific policies and practices as well as the HAI rates in eligible ICUs (i.e. an adult medical, medical/surgical or surgical ICU in which the infection prevention and control department reported to NHSN device-associated HAI surveillance in 2007).
As part of NHSN, all facilities follow specific surveillance protocols and define the presence or absence of HAI using standard CDC definitions [9
]. These protocols, which include accurate case finding and HAI definitions that have both laboratory and clinical criteria, [12
] were developed by CDC epidemiologists and have become the recognized standard that infection control professionals around the world use [14
]. Previous researchers examined the sensitivity and specificity of HAI defined by these protocols compared with infections found by trained data collectors and confirmed by epidemiologists; they found the sensitivity and specificity for VAP was 68 and 98%, respectively [16
As part of the survey, respondents were asked about four elements of the ventilator bundle: raising the head of the bed, sedation vacation, peptic ulcer and DVT prophylaxis. For each of the elements, respondents were asked the following for each of their medical, medical/surgical or surgical ICUs: (i) whether the ICU had a written policy for that bundle element, (ii) whether compliance with the policy was monitored and, (iii) the proportion of time the policy was correctly implemented. The latter was assessed using the following scale: all of the time (95–100%), usually (75–94%), sometimes (25–74%), rarely/never (<25%) and do not know. Individual ventilator bundle components were characterized based on being present and reported compliance with the policy. If the rate of compliance with policy was missing versus reported as a ‘do not know’, we assumed compliance to be low and set the value to rarely/never.
We also collected data on a number of other setting characteristics including the organizational structure and resources in the infection control department and other information about the hospital that has been associated with HAI rates. The infection control department characteristics assessed included staffing [i.e. presence of a full or part-time hospital epidemiologist, number of full-time equivalent positions for infection preventionists (IPs) per hospital bed and proportion of IP hours that were provided by a certified IP], the number of years that a hospital has been a member of NHSN or the NNIS system (the precursor to NHSN), and the use of an electronic surveillance system for tracking of HAIs. We further identified whether or not the hospital was located in a state with mandatory HAI reporting requirements. Other setting characteristics included hospital teaching status, number of licensed beds, hand hygiene compliance and ICU types (i.e. adult medical, surgical or medical surgical).
Statistics were computed to describe the sample. To assess generalizability of our sample, we compared our hospitals to all NHSN hospitals in terms of VAP rates and demographic characteristics. Then, in multivariable analyses, we examined associations between VAP rates and simply having a written policy, monitoring policy implementation and different levels of compliance. Once the needed level of compliance was established and set at 95% of the time or greater vs. other, we conducted a set of four similar multivariable analyses to elucidate the relationship between the bundle elements and VAP rates. In Model 1, we examined the independent contribution of the four bundle elements on VAP rates. In Model 2, we looked at the impact of complying with either one of the two VAP-specific bundle elements on reducing VAP rates versus compliance with neither of these elements. Model 3 assessed whether compliance with both of the two VAP-specific bundle elements was necessary to see a significant reduction in VAP rates. Next, we examined whether compliance with the two VAP-specific bundle elements was associated with lower CLABSI rates (Model 4). Finally, we estimated the change in VAP rates that would be predicted if an ICU went from no ventilator bundle implementation to full bundle compliance.
All models were multivariable ordinary least squares (OLS) regressions with Huber-White standard errors to account for intra-hospital correlations across ICU. In all analyses, indicators for the setting and infection control department characteristics previously described were included.