admissions and the median Apache II score 21 (IQR 16–27). See Table 1 and 2 and Table B in S1 File. The median age and Apache II scores were comparable for admissions with and without a VAP. The routines and regimens offered to patients differed between hospitals. Apart from one hospital, hospitals preferred either MDI or nebulizers, four of them exclusively. IvSDD was used in three hospitals, intestinal prophylaxis in three, and SOD in five (see Figure B in S1 File). In hospital G, SOD was given for one year, followed by a complete SDD regimen (oropharyngeal and intestinal prophylaxis and four days ivSDD), as part of a study. In hospital E, SOD was initially given occasionally and subsequently extended to all admissions. Univariate results Of the time‐independent covariates, age, length of hospital stay before start of the ventilation, COPD, intubation department, and duration of participation with the surveillance (in years) were significantly associated with VAP (Table 1). All time‐ dependent variables were significantly associated with the risk of acquiring VAP except for type of inhalation therapy (Table 2). Fig. 3 shows the WCE results for the univariate analyses. The overall hazard ratio of specific exposure patterns compared with uniformly non‐exposure can be calculated by multiplying the hazard ratios for all days where, for example, systemic antibiotics (Fig 3C) were administered. Suppose a patient was ventilated for 6 days and treated with ivAB on the first four ventilation days (day ‐5 to ‐2), but not on the last two ventilation days (day ‐ 1 and 0). The hazard ratio for this patient on the present day (day 0) is a multiplication of the hazards on days ‐5, ‐4, ‐3, and ‐2. Note that the hazard ratio before day ‐4 in Fig 3C is assumed to be 1. The WCE model for ivAB suggests a harmful effect of antibiotics taken on the present day (day 0), which is probably a result of reverse causation since system antibiotics were likely taken on the day of a VAP to treat the pneumonia. Further, the narrower confidence intervals shown at the left (e.g. Fig 3A) may be seen as counterintuitive since the number of patients declines with increasing follow‐up duration (Figure A in S1 File), but are an artefact of the analysis since the best‐fitting model for ivAB was one where the hazard ratio was constrained to the null at this point. Multivariate results After backward selection to step‐wise exclude covariates from the model, the final multivariate model included age, COPD, current sedation score, current SOD, inhalation therapy (WCE), ivSDD (WCE), and ivAB (WCE) (Table 3). 4 67 Risk factors for VAP using flexible methods
RkJQdWJsaXNoZXIy MTk4NDMw