To examine the determinants of mortality, a prospective cohort study was performed recruiting all patients with microbiologically confirmed VAP during a three-year period (1 April 2004 to 31 March 2007) in our 54-bed medical and surgical ICU of the 1060-bed Ghent University Hospital. To address the determinants for MDR bacterial aetiology of VAP, we performed a subsequent case-control study of patients with MDR VAP. Patients with VAP caused by non-MDR pathogens acted as controls. All patients aged 16 years and older and ventilated for at least 48 hours were assessed daily for evidence of VAP. Patients who were chronically mechanically ventilated were excluded. Only microbiologically confirmed episodes of VAP were considered for analysis. The study was approved by the Ethics Committee of Ghent University Hospital. Written informed consent to obtain patients' data was given by the patient or the patient's representative if the patient was unable to give consent.

Routine microbiological work-up of airway samples consisted of rapid Gram-staining and semi-quantitative culturing of tracheal aspirate. On clinical request (in cases of discrepancy between likely clinical diagnosis of VAP and semi-quantitative microbiological results), quantitative culturing was performed on tracheal aspirate or on broncho-alveolar lavage fluid obtained by means of fibre-optic bronchoscopy. Plate quantitation for semiquantitative cultures and use of selective media to identify MDR pathogens was performed as described previously 2223. Semiquantitative scoring was derived from streaking and diluting the specimen in three segments, scored as few (+-) for less than 10 colonies, light (+), moderate(++) and heavy (+++) growth when moderate to heavy growth was observed in first, second and third streaks respectively. Antibiotic susceptibility was determined according to methods recommended by the Clinical and Laboratory Standards Institute 24.

At our hospital, initial antibiotic therapy in ICU-acquired infection is guided by surveillance cultures, as described previously 222325. At clinical diagnosis of VAP, the following antibiotics were prescribed if surveillance cultures did not grow Pseudomonas aeruginosa or MDR organisms: a second-generation cephalosporin or amoxicillin-clavulanic acid in pneumonia diagnosed within one week or less after ICU admission of a patient without prior antibiotic exposure; or an antipseudomonal β-lactam in patients with prior antibiotic exposure or an ICU stay of more than one week. If additional risk factors for P. aeruginosa were present (e.g. bronchiectasis, corticosteroid therapy) or if P. aeruginosa was isolated from surveillance cultures, an antipseudomonal β-lactam treatment was complemented with an aminoglycoside or fluoroquinolone. In patients with surveillance cultures growing MDR organisms, initial antibiotic therapy, consisting of an antipseudomonal β-lactam antibiotic or carbapenem was complemented by a glycopeptide, fluoroquinolone or aminoglycoside as appropriate; alternatively, targeted therapy directed at the MDR pathogen was provided.

Data collected at ICU admission included demographics, admission diagnosis, presence of comorbidity, severity of illness on admission as assessed by Acute Physiology and Chronic Health Evaluation (APACHE) II score, presence of coma (defined by Glasgow Coma Scale (< 6) and development of circulatory shock on ICU admission (defined as requirement of vasopressor therapy after restoring intravascular volume within 48 hours of ICU admission). Presence of comorbidity was quantified using the Charlson index of comorbidity 26, as described previously 2427.

Data collected at diagnosis of VAP were prior duration of mechanical ventilation (days), prior antibiotic therapy within the same hospitalisation period, number of infectious episodes and number of different antibiotic classes prescribed. We recorded a diagnosis of underlying Acute Respiratory Distress Syndrome (ARDS) and underlying acute kidney injury requiring renal replacement therapy, if present at least two days (day -2) before VAP, and the presence or absence of shock on ICU admission. Clinical pulmonary infection score (CPIS) was calculated at suspicion of VAP to corroborate clinical diagnosis 28. Microbial aetiology was recorded if available (see definitions below). Sequential Organ Failure Assessment (SOFA) score was calculated on the day of diagnosis of VAP, as well as at day -2 and two days after (day +2) diagnosis of VAP. The SOFA score is a scoring system quantifying the extent of a critically ill patient's organ dysfunction or failure, and is composed of six subscores, one each for the respiratory, cardiovascular, hepatic, coagulation, renal and neurological systems 29.

Antibiotic prescription on diagnosis of VAP was noted: antibiotic prescription on the same calendar day and the calendar day after clinical diagnosis of VAP was considered as antibiotic therapy within 24 hours and 48 hours of VAP, respectively. Primary outcome parameter was 30-day mortality after diagnosis of VAP.

VAP was considered clinically likely if a new or progressive and persistent infiltrate was present on chest X-ray together with at least two signs of systemic inflammation, such as fever with a temperature higher than 38°C or hypothermia with a temperature lower than 36°C, leucocytosis (>11,000 white blood cell count (WBC)/mm³) or leucopenia (<4000 WBC/mm³), rising C-reactive protein (> 2 mg/dL within 48 hours), and with at least one sign of local inflammation such as purulent sputum and a decrease of partial pressure of oxygen in arterial blood (PaO₂₎/fraction of inspired oxygen (FiO₂) of at least 10%. Moreover, a CPIS of at least six was required to maintain diagnosis of clinically likely VAP 28.

Clinically likely VAP was considered as microbiologically confirmed if: a pathogen showed ++ or +++ semiquantitative culture or more than 10⁵ colony forming units (CFU)/mL quantitative growth on a good quality endotracheal aspirate; growth of ++ semiquantitative culture of more than 10⁴ CFU/mL on broncho-alveolar lavage fluid; growth of at least + together with positive Gram-staining if antibiotic therapy had been initiated or changed within 48 hours before sampling; or when a pathogen was isolated both from endotracheal aspirate and blood cultures. Based on a previous in-house analysis where semiquantitative and quantitative cultures on endotracheal aspirate correlated well (data not shown), and supported by other reports, we rely both on semiquantitative and quantitative cultures for microbiological confirmation of VAP 3031. If more than one pathogen grew above the semiquantitative or quantitative threshold, VAP was considered polymicrobial and if at least one MDR pathogen grew above these thresholds, VAP was considered as MDR.

The following pathogens were considered as MDR: methicillin-resistant Staphylococcus aureus (MRSA), extended-spectrum β-lactamase producing Gram-negative Enterobacteriaceae (ESBL), Pseudomonas aeruginosa and other non-fermenting organisms (Acinetobacter baumannii, Stenotrophomonas maltophilia) resistant for three or more of the following antibiotic classes: antipseudomonal cephalosporins or penicillins, carbapenems, fluoroquinolones and aminoglycosides (MDR NF). Antimicrobial therapy within 24 hours and 48 hours of diagnosis of VAP was considered appropriate if it included at least one antimicrobial drug with in vitro activity against the aetiologic agent identified. ARDS was defined according to the criteria of the American-European consensus conference 32, and shock was defined as the requirement of vasopressor therapy (noradrenaline or adrenaline) to restore adequate arterial pressure and organ perfusion despite appropriate intravenous fluid substitution.

Continuous variables are described as mean (± standard deviation), median (25th to 75th percentile) and categorical variables are described as n (%). For comparative tests on continuous variables, the Mann-Whitney U test and student's t-test were used as appropriate, depending on variable distribution. For categorical variables, the Pearson chi-square test or the Fisher's exact test were used as appropriate. The response variable used in the mortality analyses was vital status (alive or dead) 30 days after diagnosis of VAP. In patients with multiple episodes of VAP, only the first microbiologically confirmed VAP was retained for further analysis. Logistic regression analysis was used to assess the multivariate relation between multiple patient characteristics and the probability of involvement of MDR as compared with non-MDR pathogens in VAP. To adjust for the association of MDR versus non-MDR microbial aetiology on 30-day mortality after diagnosis of VAP on potential confounders and to check whether MDR is a independent predictor, we performed a competing risk analysis using the Fine and Gray model 21, with 30-day mortality after diagnosis of VAP as the endpoint of interest, and discharge alive from the hospital within 30 days after diagnosis of VAP as the competing risk.

Recently some authors discussed the application of these recently developed models in the specific ICU-setting where censoring due to discharge alive from the ICU violates the assumption of non-informative censoring 3334. Consequently, standard survival methods which rely on non-informative censoring appear not to be appropriate here 3536. As the primary aim was to determine whether MDR constituted an independent risk factor for mortality in the presence of other covariates, the enter method was primarily used, complemented with stepwise forward and backward analysis to test stability of the models. Overall, predictors showing a p < 0.1 association with in-hospital mortality in univariate analysis as well as those variables that seemed clinically important were incorporated in the regression analyses. Correlation matrixes for all predictors included in the regression analyses were constructed to avoid inclusion of significantly associated sets of predictors and to limit the risk of colinearity.

When appropriate, odds ratios (OR) and 95% confidence intervals (CI) were reported, and the Hosmer-Lemeshow goodness-of-fit test and the area under the curve of the resulting receiver operator curve were provided. Results from the competing risk analysis were reported as sub-hazard ratios, which are the ratios of hazards associated with the cumulative incidence function. The various models were tested for the presence of clinically significant interaction. Statistical analyses were executed with SPSS 11.0 (SPSS Inc. Chicago, IL) and the R 2.6.2 software package 37. The competing risk analysis was performed using the crr routine available in the cmprsk package developed by Gray 38. All tests used were two-tailed and statistical significance was defined as p < 0.05.

Characteristics of patients with VAP caused by MDR versus non-MDR pathogens are provided in table 1, as well as the predominant pathogen identified. The univariate odds ratio for isolating a MDR pathogen in patients previously exposed to an increasing number of antimicrobial classes is shown in figure 2. Rates of appropriate antibiotic therapy achieved within 24 hours and 48 hours after diagnosis of VAP were lower in patients with MDR as compared with other pathogens, and 30-day, ICU- and in-hospital mortality were significantly higher in patients with VAP caused by MDR than in patients with non-MDR VAP.

Results of the multivariable analysis of predictors of MDR as compared with non-MDR microbial aetiology of VAP are shown in table 2. We included exposure to one, two and more than two antibiotic classes (with no prior antibiotics as reference category), together with duration of mechanical ventilation (days) prior to VAP as predictors, as these are known risk factors for MDR 20, as well as those predictors that showed a significant association (p < 0.1) in univariate analysis. Exposure to more than two antibiotic classes during hospitalisation before VAP diagnosis was significantly associated with MDR aetiology in enter and backward stepwise regression analysis.

Characteristics of nonsurvivors and survivors are shown in univariate analysis in table 3. Results of the Fine and Gray regression model are shown in table 4. MDR bacterial aetiology was not independently associated with mortality. Renal replacement therapy before diagnosis of VAP (standardised hazard ratio (SHR) = 2.69, 95% CI = 1.47 to 4.94, p = 0.001), the Charlson index of comorbidity (SHR = 1.21, 95% CI = 1.03 to 1.41, p = 0.02) and shock on ICU admission (SHR = 1.86, 95% CI = 1.03 to 3.35, p = 0.04) were significant predictors of 30-day mortality after VAP diagnosis.