Risk factors for nosocomial pneumonia have been evaluated in hospitalized elderly persons [10], heterogeneous groups of hospitalized patients breathing spontaneously or requiring assisted ventilation [11], and similar mixed populations admitted to the intensive care unit [2, 12-15]. Investigations in ventilated critically ill patients only are needed to identify independent predictors of ventilator-associated pneumonia. Most studies of risk factors for ventilator-associated pneumonia have been conducted at single institutions, and different predictors have been examined. In cohort studies using multivariable analyses [16-21], only reintubation [17, 19, 21] and antibiotic use [18, 20] were identified as risk factors in more than one report. However, a direct relation between antibiotics and pneumonia was found in one study [18], whereas an inverse relation was found in another [20].

We examined factors associated with ventilator-associated pneumonia. We explored baseline and time-dependent characteristics, including measures of illness severity, factors relating to mechanical ventilation, variables in the gastropulmonary route of infection, and drug exposure.

Methods

Top Methods Results Discussion Author & Article Info References

Patients

We used a multicenter national database of 1200 patients who received mechanical ventilation for 48 hours or more and were enrolled from October 1992 to May 1996 in a double-blind, concealed, randomized trial of sucralfate compared with ranitidine to determine rates of ventilator-associated pneumonia and gastrointestinal bleeding [22]. Patients were followed until death, discharge, or clinical suspicion of ventilator-associated pneumonia as determined by the attending intensivist. Bronchoalveolar lavage or protected specimen brush was indicated for patients with a clinical suspicion of ventilator-associated pneumonia. A diagnosis of ventilator-associated pneumonia was considered only in patients who were receiving mechanical ventilation or who stopped receiving ventilation for less than 48 hours. In the absence of a reference standard for diagnosing pneumonia in critically ill patients [23, 24], each patient's clinical, laboratory, and radiographic data were independently reviewed by one of four pairs of adjudicators blinded to treatment group. The readers in each adjudication pair decided on the presence or absence of ventilator-associated pneumonia, resolving any disagreement through discussion. Consensus was achieved on all cases.

The adjudicated outcome was used for the primary analysis [177 cases]. However, each case of pneumonia was classified according to four additional definitions: 1) the bedside clinician's diagnosis [233 cases]; 2) the modified Centers for Disease Control and Prevention criteria [25], which require a new radiographic infiltrate persistent for 48 hours or more plus a body temperature greater than 38.5 °C or less than 35.0 °C, a leukocyte count of more than 10 cells x 10⁹/L or less than 3 cells x 10⁹/L, purulent sputum or change in character of sputum, or isolation of pathogenic bacteria from an endotracheal aspirate [211 cases]; 3) a Clinical Pulmonary Infection score of 7 or more [26] [194 cases]; or 4) a positive culture from bronchoalveolar lavage (>10⁴ colony-forming units/mL) or protected specimen brush (<10³ colony-forming units/mL) (97 cases).

Statistical Analysis

We expressed continuous variables as the mean (±SD) or as the median and interquartile range if their distribution was skewed. We used the Student t-test to compare continuous variables and the chi-square test to compare proportions. All statistical tests were two-tailed.

We performed a forward stepwise Cox proportional-hazards regression analysis to evaluate potential risk factors for ventilator-associated pneumonia. The Cox model assesses the effect of each risk factor on the hazard rate of ventilator-associated pneumonia over time, adjusted for other factors in the model and allowing for censoring because of death or discharge. The hazard function in the Cox model can be used to estimate the event rate per day over the duration of stay in the intensive care unit.

Independent variables recorded at baseline included age; sex; primary diagnosis; medical or surgical status; hospital admission in the fall or winter compared with spring or summer; location before admission to the intensive care unit; treatment center; Acute Physiology and Chronic Health Evaluation II score; Acute Physiology score; Multiple Organ Dysfunction score [27]; Glasgow Coma Scale score; and chronic comorbid conditions, such as alcoholism, a smoking history of 10 or more pack-years, asthma, bronchiectasis, pulmonary fibrosis, cancer, and HIV infection; organ transplantation; and recent corticosteroid therapy or chemotherapy. Additional independent variables evaluated daily throughout the intensive care unit stay were classified as illness severity measures (daily Multiple Organ Dysfunction score and change in Multiple Organ Dysfunction score from baseline), ventilation variables (mechanical ventilation, change or reinsertion of endotracheal tube, discontinuation and reinstitution of mechanical ventilation, and tracheostomy), nutritional variables (nasoenteral nutrition, gastrostomy feeding, jejunostomy feeding, enteral nutrition through any route, central parenteral nutrition, peripheral parenteral nutrition, and insulin requirement of more than 15 U/d), witnessed aspiration, and drug exposure (stress-ulcer prophylaxis with ranitidine or sucralfate, paralytic agents, and antibiotics).

The Multiple Organ Dysfunction Score combines measures of physiologic dysfunction in six domains: the cardiovascular system (heart rate x right atrial pressure/mean arterial pressure), pulmonary system (PaO₂:FIO₂ ratio), the renal system (serum creatinine concentration), the hepatic system (serum bilirubin level), the hematologic system (platelet count), and the central nervous system (Glasgow Coma Scale score) [25]. These continuous variables are constructed in 5 categories so that 0 represents normal function and 4 reflects marked physiologic derangement and is associated with a mortality rate greater than 50%. Composite scores provide a quantitative measure of global physiologic dysfunction at a discrete point in the intensive care unit stay. For individual domains, either the raw data or their score provides a measure of dysfunction in the system of interest.

Variables that were associated with ventilator-associated pneumonia in the univariable analysis and had a P value less than 0.10 were entered into a multivariable regression analysis. Factors were considered significant if the P value was less than 0.05 in the multivariable analysis. We calculated risk ratios and 95% CIs for all significant predictors of ventilator-associated pneumonia in the multivariable analysis using all five definitions.

We tested for all pairwise interactions between significant risk factors in the multivariable model. We also tested for interactions between predictors in the multivariable model and factors that were significant in the univariable analysis. Finally, we investigated the possibility that the effects of risk factors may vary over the duration of stay in the intensive care unit; this was done by using a nonproportional Cox model testing the interaction of each variable in the final model with time. For example, we hypothesized that the influence of antibiotics on ventilator-associated pneumonia may vary over time. Time was considered both a continuous variable and a discretized variable by using 3-day and 5-day intervals. We did not consider antibiotics administered 24 hours before the diagnosis of ventilator-associated pneumonia, so that drugs prescribed in response to early ventilator-associated pneumonia would be excluded [28].

Role of Study Sponsor

Our funding sources had no role in the acquisition, analysis, or interpretation of data or in the submission of this report.

Results

Top Methods Results Discussion Author & Article Info References

Of the 1200 patients enrolled in this study, 186 were excluded (85 were discharged, 71 died, and 30 had pneumonia in the first 48 hours), leaving 1014 patients ventilated for 48 hours or more who were free of pneumonia at admission to the intensive care unit. Of these patients, 177 (17.5%) developed ventilator-associated pneumonia 9.0 ± 5.9 days after admission (median, 7 days [interquartile range, 5 to 10 days]). The characteristics of patients with and without ventilator-associated pneumonia are shown in Table 1. The total duration of mechanical ventilation among patients with ventilator-associated pneumonia was 19.3 ± 16.0 days (median, 15 days [interquartile range, 9 to 23 days]) compared with 10.2 ± 10.5 days (median, 7 days [interquartile range, 4 to 11 days]) in other patients (P < 0.001). Among the 177 patients with ventilator-associated pneumonia, 131 (74.0%) had bronchoscopic testing with bronchoalveolar lavage or protected specimen brush. Patients in the following admitting diagnostic categories had ventilator-associated pneumonia: cardiovascular disease (41 of 277 patients [14.8%]), respiratory disease (24 of 157 patients [15.3%]), thoracic or abdominal surgery (18 of 130 patients [13.9%]), gastrointestinal disease (4 of 31 patients [12.9%]), central nervous system disease (39 of 164 patients [23.8%]), trauma (31 of 119 patients [26.1%]), burns (7 of 10 patients [70%]), sepsis (5 of 63 patients [7.9%]), transplantation (0 of 16 patients [0%]), and metabolic or other disease (8 of 47 patients [17.0%]) (P < 0.001 for difference across groups).

The cumulative risk for developing ventilator-associated pneumonia over successive days in the intensive care unit is shown in Figure 1. The overall incidence of ventilator-associated pneumonia was 14.8 cases per 1000 ventilator-days. The daily risk for developing ventilator-associated pneumonia increased until day 5, then decreased over the duration of stay in the intensive care unit (Figure 2). The risk per day was approximately 3.3% at day 5, 2.3% at day 10, and 1.3% at day 15.

View larger version (11K): [in this window] [in a new window]   Figure 1. Proportion of patients free of pneumonia during their stay in the intensive care unit.

View larger version (11K): [in this window] [in a new window]   Figure 2. Hazard rate for ventilator-associated pneumonia during the stay in the intensive care unit. The hazard function presents the conditional probability of ventilator-associated pneumonia in the next day, given that a patient is event free. Estimation of the hazard function shows the event rate per day over the duration of ventilation.

Factors associated with ventilator-associated pneumonia in the univariable analysis were age, male sex, primary admitting diagnosis, intensive care unit admission from the emergency department, low Glasgow Coma Scale score, increased Multiple Organ Dysfunction score compared with that at admission, mechanical ventilation in the previous 24 hours, nasoenteral nutrition, enteral nutrition by any route, witnessed aspiration, administration of paralytic agents, and absence of exposure to antibiotics (Table 2). Administration of stress-ulcer prophylactic agents was not a predictor of ventilator-associated pneumonia in the univariable analysis; this is in keeping with our randomized trial results showing that patients receiving ranitidine had a rate of ventilator-associated pneumonia similar to that seen in patients receiving sucralfate (relative risk, 1.18 [95% CI, 0.92 to 1.51]) but had a lower rate of gastrointestinal bleeding (relative risk, 0.44 [CI, 0.21 to 0.92]) [22].

Independent predictors included a primary admitting diagnosis of burns (risk ratio, 5.09 [CI, 1.52 to 17.03]), trauma (risk ratio, 5.0 [CI, 1.91 to 13.11]), central nervous system disease (risk ratio, 3.4 [CI, 1.31 to 8.81]), respiratory disease (risk ratio, 2.79 [CI, 1.04 to 7.51]), cardiac disease (risk ratio, 2.72 [CI, 1.05 to 7.01]), mechanical ventilation in the previous 24 hours (risk ratio, 2.28 [CI, 1.11 to 4.68], witnessed aspiration (risk ratio, 3.25 [CI, 1.62 to 6.50]), and paralytic agents (risk ratio, 1.57 [CI, 1.03 to 2.39]). Exposure to antibiotics conferred protection (risk ratio, 0.37 [CI, 0.27 to 0.51]) (Table 2). Results are adjusted for center, which was significant in the final model (P = 0.007). Patients with and those without ventilator-associated pneumonia are described according to the time-dependent independent predictors in Table 1. Risk factors were the same when ventilator-associated pneumonia was diagnosed according to the bedside clinician's assessment, the Centers for Disease Control and Prevention definition, a Clinical Pulmonary Infection score of 7 or more, and positive culture from bronchoalveolar lavage (>10⁴ colony-forming units/mL) or protected specimen brush (>10³ colony-forming units/mL).

No significant interactions were seen between the independent predictors in the multivariable analysis or the independent predictors and the significant predictors found in the univariable analysis. Evaluation of risk factors in the nonproportional hazards model demonstrated that only the effect of antibiotics changed statistically over time. The risk ratio of ventilator-associated pneumonia associated with antibiotics was 0.30 (CI, 0.17 to 0.52) at day 5, 0.43 (CI, 0.21 to 0.87) at day 10, 0.62 (CI, 0.24 to 1.62) at day 15, and 0.89 (CI, 0.25 to 3.13) at day 20. The CIs around these risk ratios widen as the duration of stay in the intensive care unit increases because of the small number of patients and the low proportion of ventilator-associated pneumonia outcomes observed late after intensive care unit admission. Of all cases of ventilator-associated pneumonia, 40.1% developed before day 5, 41.2% developed during days 6 to 10, 11.3% developed during days 11 to 15, 2.8% developed during days 16 to 20, and 4.5% developed after day 21.

Antibiotics administered included cephalosporins (62.9%), penicillin derivatives (31.4%), aminoglycosides (23.6%), metronidazole (20.1%), macrolides (13.4%), quinolones (10.7%), sulfa derivatives (8.9%), miscellaneous drugs (16.5%), antifungal agents (7.6%), and antiviral agents (3.3%) (the latter two agents were not included in the antibiotic analysis). Patients with ventilator-associated pneumonia received fewer antibiotics and were exposed for a shorter time than patients without ventilator-associated pneumonia. When analyzed according to the cumulative number of antibiotics administered, the risk ratio was 0.94 (CI, 0.92 to 0.97) for each antibiotic prescribed.

Discussion

Top Methods Results Discussion Author & Article Info References

We examined the incidence of and risk factors for ventilator-associated pneumonia in a cohort of 1014 patients. The risk for developing ventilator-associated pneumonia increased cumulatively, with an overall rate of 14.8 cases per 1000 ventilator-days. However, we found ventilator-associated pneumonia rates of approximately 3% per day in the first week of mechanical ventilation, 2% per day in the second week, and 1% per day in the third week and beyond. This decreasing hazard reflects the high risk for early ventilator-associated pneumonia in ventilated patients and suggests that long-term survivors are patients at lower intrinsic risk for ventilator-associated pneumonia. Other studies have emphasized a high risk for ventilator-associated pneumonia in the first week of mechanical ventilation [20, 29]. Early-onset pneumonia may account for as many as 50% of cases of ventilator-associated pneumonia, and most etiologic organisms represent common respiratory tract pathogens or normal oropharyngeal flora [2, 3, 15].

Like other investigators [17], we found aspiration to be an independent risk factor for ventilator-associated pneumonia. Pharyngeal aspiration in patients with depressed consciousness is common [30]. Impaired airway reflexes [2, 15] and neuromuscular disease [15] in a heterogeneous group of spontaneously breathing and mechanically ventilated patients were documented as risk factors for nosocomial pneumonia. However, the identification of muscle relaxants as an independent risk factor for ventilator-associated pneumonia has not been demonstrated in studies of mechanically ventilated patients. Although the association was seen in a univariable analysis of 277 ventilated patients [18], two smaller studies found no association [16, 20]. In an observational study of 66 patients with head trauma, 40% of those who received thiopental and 18% of those who did not receive thiopental developed ventilator-associated pneumonia [31]. Paralytic agents and sedatives may predispose to ventilator-associated pneumonia by decreasing the cough reflex, reducing endotracheal secretion clearance, or impairing gut motility.

The relation between administration of antibiotics and occurrence of ventilator-associated pneumonia is complex. Antibiotics were associated with an increased risk for ventilator-associated pneumonia in a cohort study of 320 patients [18]; however, other investigators found that antibiotics administered during the first 8 days were associated with a reduced risk for early-onset ventilator-associated pneumonia in patients with ineffective subglottic secretion drainage [20]. Using different metrics for antibiotic exposure, we found antibiotic administration to be associated with lower rates of ventilator-associated pneumonia (Table 1). The inverse relation between antibiotic use and ventilator-associated pneumonia is consistent with results of randomized trials of selective digestive decontamination, recently summarized in an updated systematic review from the Cochrane Collaboration [32]. Selective digestive decontamination confers a large, clinically important, and statistically significant reduction in lower respiratory tract infections (common odds ratio, 0.35 [CI, 0.29 to 0.41]) [33]. With respect to mortality, the combination of topical and systemic antibiotic prophylaxis was associated with lower mortality (common odds ratio, 0.80 [CI, 0.69 to 0.93]), whereas topical antibiotics only were not (common odds ratio, 1.01 [CI, 0.84 to 1.22]). That systemic antibiotics may protect against ventilator-associated pneumonia is also consistent with results from a recent controlled trial by Sirvent and colleagues [34], who showed that a short course of cephalosporin prophylaxis was associated with a lower rate of ventilator-associated pneumonia in patients with structural coma. We demonstrated that the apparent protective effect of antibiotics disappears after 2 to 3 weeks. Sparse data preclude extrapolation beyond this point; however, an increased risk for ventilator-associated pneumonia cannot be excluded, and prolonged antibiotic administration in intensive care unit patients is expected to favor selection and subsequent colonization with resistant pathogens [35].

We found that nasoenteral nutrition and enteral nutrition by any route were associated with ventilator-associated pneumonia in univariable analysis but not multivariable analysis. Enteral nutrition delivered by nasogastric tube is associated with gastric and endotracheal colonization, which may predispose to ventilator-associated pneumonia [36-38]. In addition, nasal instrumentation favors the development of sinusitis [39]. Trends toward higher rates of sinusitis and pneumonia were found in randomized trials of oral compared with nasal endotracheal intubation [40-43]. Although the incidence of pneumonia in proximally compared with distally delivered feeds was similar in patients receiving gastric and jejunal nutrition (2 of 19 patients compared with 0 of 19 patients) [44], the association between nutrition and ventilator-associated pneumonia warrants further evaluation in large trials examining the sites of intubation and nutritional delivery.

We did not systematically collect data on body position. A causal relation between ventilator-associated pneumonia and body position has been suggested by documentation of supine positioning as an independent predictor of ventilator-associated pneumonia in regression analysis [18], observational data on gastric colonization and ventilator-associated pneumonia [45], and results of three randomized trials showing that patients nursed in the supine rather than the semirecumbent position had a significantly higher rate of aspiration demonstrated scintigraphically [33, 46, 47]. However, a systematic review of this issue shows that most studies using regression modeling to examine risk factors for ventilator-associated pneumonia have not found this association with body position [48]. Therefore, this diverse evidence is consistent with, but not necessarily proof of, the gastropulmonary route of infection.

Because our results were derived from a randomized trial that enrolled patients who received ventilation for at least 48 hours, not all patients at risk for ventilator-associated pneumonia were included. Specifically, we excluded patients on admission who were breathing spontaneously or had pneumonia and patients who required ventilation but were likely to die or be extubated and discharged within 48 hours. We previously documented that participating physicians were approximately 90% accurate at predicting a duration of ventilation of more than 48 hours [49]. Our purpose was not to examine risk factors for ventilator-associated pneumonia due to particular microorganisms [47, 50, 51], risk factors for death in patients with ventilator-associated pneumonia [52-54], risk factors for attributable ventilator-associated pneumonia mortality [55], or the relation between mechanical ventilation and intensive care unit-acquired infection [1].

Strengths of our study include use of a national multicenter database and examination of acute and chronic health variables, demographic data, and several time-dependent factors to evaluate their relation to ventilator-associated pneumonia. Both independent and dependent variables were collected prospectively; outcomes were determined without knowledge of the predictor variables, and vice versa [56]. Our sample size was larger than those in previous studies. We used a definition of ventilator-associated pneumonia that was established through adjudication of all suspected cases and that incorporated radiographic, clinical, and bronchoscopic criteria. In the absence of a reference standard for ventilator-associated pneumonia, we validated our findings by using five definitions of ventilator-associated pneumonia. We used the more powerful Cox proportional-hazards analysis rather than logistic regression to retain information on the exact timing of ventilator-associated pneumonia, taking censoring into account. Modeling the effect of each independent risk factor over time shows that the apparent protective effect of antibiotics attenuates as the time in the intensive care unit increases.

Risk factors offer prognostic information about the probability of developing lung infection in individual patients and in populations. They help us understand some of the pathophysiologic mechanisms that predispose to pneumonia, and these insights may lead to development of effective preventive strategies [45]. Finally, risk stratification can highlight which patients may be most likely to benefit from pneumonia prophylaxis. We found several potentially amenable risk factors for ventilator-associated pneumonia. Inferences about fixed and variable risk factors are increased when associations are strong, independently predictive, reproducibly demonstrated, and biologically plausible and when they are validated by using different outcome definitions. When variable risk factors are modified in experiments and found to influence the target outcome, these factors may be viewed as causal. Pneumonia prevention approaches focused on nutrition [9], and the ventilator circuit and secretion management [8] have been studied in 28 randomized, controlled trials to date. Building on the results of risk factor studies of ventilator-associated pneumonia, future prevention trials may consider diverse strategies related to nutrition, sedation, paralytic agents, and systemic antibiotic therapy targeted at high-risk populations with long-term follow-up for emergence of resistant organisms.

Drs. Walter and Guyatt and Ms. Griffith: Department of Clinical Epidemiology, McMaster University Medical Center, 1200 Main Street West, Hamilton, Ontario L8N 3Z5, Canada.

Dr. R. Cook: Department of Statistics and Actuarial Science, University of Waterloo, 200 University Avenue, Waterloo, Ontario N2L 361, Canada.

Dr. Leasa: Department of Medicine, London Health Sciences Centre-University Campus, 339 Windermere Road, London, Ontario N6A 5A5, Canada.

Dr. Brun-Buisson: Department of Medical Intensive Care and Infection Control Unit, Hopital Henri Mondor, 51, avenue du Mal de Lattre de Tassigny, 94010 Creteil Cedex, France.