density was analyzed using a sub‐distribution Cox proportional hazard analysis, stratified by hospitals. As patients can have more than one CVC, we adjusted for possible clustering at the patient level using robust covariance estimation. Records with missing values for a variable were excluded from the regression analysis of that variable. Patient and CVC characteristics with a p‐value < 0.2 were included in the multivariable regression model. Following distribution of the dependent variables, the association of the interventions with the CVC insertion score was analyzed using generalized linear mixed modeling with a binomial distribution, and with hand hygiene using generalized linear mixed modeling, with a normal distribution, both allowing for clustering at the hospital level. The association with professional category, type of ICU, shift, weekday, and activity index (number of hand hygiene opportunities per hour [17], averaged per quarter) was evaluated and these were included in the multiple regression model in case of a p‐value < 0.2. All analyses were performed 1) without assuming a time‐dependent trend, and 2) with assuming a baseline hospital‐specific time‐dependent trend and an additional intervention‐specific time‐dependent trend. Time was modeled as quarters during baseline (1 to a maximum of 6), and intervention (1 to a maximum of 8), assuming a linear trend. These models were fixed models, with hospitals included as a covariate to allow interaction terms of hospitals with the time‐dependent trend. The direct association of the two process indicators (CVC insertion score and hand hygiene compliance) with CRBSI incidence was explored using the quarterly averages of the CVC insertion score and hand hygiene compliance (both as percentages) and the quarterly CRBSI numbers, using Poisson regression modeling, allowing for clustering at the hospital level. Quarterly averages of covariates associated with CRBSI in the Cox regression model were evaluated in this analysis and included in the multiple regression model in case of a p‐value < 0.2. We used SAS software, version 9.3 for all statistical analyses. Ethics The medical ethical committees of all participating hospitals approved the study before randomization. In one center individual patient consent was deemed necessary and therefore obtained. Study registration We retrospectively registered the protocol at the ISRCTN registry (ISRCTN24828982).
RkJQdWJsaXNoZXIy MTk4NDMw