INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Nosocomial pneumonia is the second most common nosocomial infection (1) and is the nosocomial infection with the highest case/fatality ratio (2). Unfortunately, past studies of nosocomial pneumonia, particularly ventilator-associated pneumonia (VAP), have been plagued by uncertainty regarding the accuracy of diagnosis based on clinical and roentgenographic criteria and culture of endotracheal secretions. When compared with diagnosis by autopsy, use of clinical and roentgenographic criteria for the diagnosis of VAP resulted in misclassification of from 29 to 62% of cases (3). Further, endotracheal intubation leads to contamination of tracheal secretions with bacteria that colonize the upper respiratory tract, compromising the value of endotracheal secretions for determining the cause of pneumonia (6). Although the causative agent of VAP is usually present in cultures of endotracheal secretions, bacteria isolated from endotracheal secretions are frequently not found in the lower respiratory tract (5, 6, 8, 9).

In 1979, Wimberly and colleagues (10) introduced quantitative culture of lower respiratory tract secretions obtained by fiberoptic bronchoscopy using a protected specimen brush (PSB) for the diagnosis of pneumonia. Subsequently, studies have been published describing the use of quantitative cultures obtained from a variety of techniques that both do [PSB (9, 11), protected BAL (12), nonbronchoscopic BAL (13)] and do not [endotracheal aspirates (14, 15), BAL (16), NBBAL (17, 18)] protect against contamination by proximal airway secretions. These quantitative culture techniques have demonstrated variable operating characteristics, with sensitivity and specificity ranging from 65 to 100%. Variability was in large part due to different standards for comparison and antibiotic management. Although the superiority of any one technique continues to be debated (19), quantitative cultures from any of these techniques consistently show less misclassification of VAP than does diagnosis based solely on clinical and roentgenographic criteria (3, 8, 12). Further, comparison of nonquantitative cultures of endotracheal secretions with quantitative cultures of specimens obtained by protected bronchoscopic techniques in mechanically ventilated patients show discrepancies between the microorganisms recovered from these two sources (6, 14, 15).

Most previous studies of risk factors for VAP have been limited by either the potential diagnostic misclassification noted above or the lack of multivariable analysis. Of eight prospective studies using multivariable analysis, six used nonspecific clinical and roentgenographic criteria to identify most or all cases of pneumonia (22). The only studies that employed both an accurate diagnostic technique and multivariable analysis were limited to pneumonia caused by Staphylococcus aureus (28) or patients with structural coma (29). Other studies of VAP that used more accurate diagnostic techniques for classification did not use multivariable analytic techniques to study risk factors (30, 31).

Use of different diagnostic criteria for pneumonia may potentially lead to discovery of alternative risk factors, with important implications for subsequent prevention strategies (29, 32). Because management decisions are based on quantitative cultures of lower respiratory tract specimens in our institutions and in many others, we proposed to study risk factors for pneumonia using these criteria. The objectives of our study were to: (1) determine rates of VAP in Medical Intensive Care Unit (MICU) patients; (2) accurately identify the causative microorganism(s); (3) identify the risk factors for VAP, as defined by quantitative cultures of specimens obtained by protected bronchoscopic techniques.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Study Population

The study was carried out between July 1, 1990 and June 5, 1991 in patients hospitalized in the MICU at the Regional Medical Center at Memphis and at the Memphis Veterans Affairs Medical Center (VAMC). The study was approved by the institutional review boards (IRB) of both the University of Tennessee, Memphis and the VAMC. The IRB of the University of Tennessee, Memphis required no consent for participation in the study, and the IRB of the VAMC, Memphis required oral consent only. Separate informed consent was obtained for all bronchoscopies. In addition, written informed consent was obtained for the use of the balloon-tipped catheter for PBAL (12) since this device was still on experimental protocol at the time of this study.

The study was a prospective, observational cohort study. All patients admitted to these MICUs were studied except those who met the following exclusion criteria: (1) patients unlikely to require intensive care for at least 3 d; (2) patients with pneumonia at the time of MICU admission; (3) patients from whom consent could not be obtained for bronchoscopy; (4) patients studied on a previous MICU admission. Only the first episode of pneumonia was studied. Data were recorded for patients only while they were in the MICU, but patients were followed for 48 h after discharge from the MICU for the diagnosis of pneumonia.

Case Definitions

Two groups of clinicians, the primary team caring for the patient as well as the research/epidemiology team, prospectively and independently evaluated all patients for the presence of pneumonia 7 d/wk. Bronchoscopy was requested on all patients, including those not intubated, who had onset of fever > 38.3° C, leukocytosis (WBC > 15,000/ mm³) or leukopenia (WBC < 4,000/mm³), and a new pulmonary infiltrate on chest radiograph not easily attributed to another cause. If the primary service did not suspect pneumonia but the research team noted that the patient met the above criteria, the primary service was contacted and bronchoscopy requested. Specimens of lower respiratory secretions were collected by PSB and/or PBAL. Only two next of kin refused permission for bronchoscopy, both because of the terminal nature of the patient's illness. Both were followed prospectively and subsequently met nonbronchoscopic criteria for pneumonia.

A case of pneumonia was defined by the presence of any of the following criteria: (1) growth of  10³ colony-forming units (cfu)/ml of a microorganism from a PSB specimen; (2) growth of  10⁴ cfu/ml of a microorganism from a PBAL specimen; (3) histopathologic demonstration of consolidation with intense polymorphonuclear leukocyte accumulation in bronchioles and adjacent alveoli of several adjacent low-power microscopic fields in a specimen obtained at postmortem examination (33); (4) rapid progression of a radiographic lung infiltrate to cavitation in the absence of carcinoma of the lung (33).

Definitions used for infections at other body sites were modified from those published by the Centers for Disease Control and Prevention (34).

Data Collection at Study Entry

Data collection at study entry included date and time of hospital admission, previous location, date and time of MICU entry, indication for MICU admission, date of birth, sex, race, height, and weight. Other data recorded on admission to the study included all diagnoses (using ICD-9 codes), smoking status (current, former, never), cumulative pack-years of smoking, ethanol abuse (current, former, never), cardiopulmonary resuscitation in the 24 h prior to study entry, results of chest radiographs, serum albumin, and severity of illness indices, including the Acute Physiology and Chronic Health Evaluation II (APACHE II) Score (35), the Therapeutic Intervention Scoring System (TISS) Score (36), and the Glascow Coma Score (37).

Daily Observations

Data collected and entered daily included endotracheal intubation (for each episode: orotracheal intubation [nasotracheal intubation was rare], surgical tracheostomy, or percutaneous tracheostomy and tube diameter, dates and times of insertion and removal), enteric intubation (for each episode: entry site [nose, mouth, or percutaneous], tip location [stomach, duodenum], tube gauge, dates and times of insertion and removal), insertion of thoracostomy tubes (for each episode: indication for placement [pneumothorax, hemothorax, empyema], location [right or left], dates and times of insertion and removal), and mechanical ventilation (dates and times begun and discontinued). Types of medications recorded included antimicrobial agents, antacids, histamine type-2 receptor antagonists (H2-blockers), sucralfate, corticosteroids, sedative hypnotics, muscle relaxants, and anticholinergics. Data on medications were collected daily including dose, route of administration, and start and stop dates. Clinical data recorded daily included rectal or core temperatures, status of gastrointestinal motility (normal or impaired [impaired motility defined as continuous absence of bowel sounds, flatus, and defecation for  24 h]), Glascow Coma Score, lowest systolic blood pressure in the preceding 24 h, whether pressor agents were required to support the blood pressure in the previous 24 h, and the level of positive end-expiratory pressure (PEEP) at 8:00 A.M. each day. Other data collected daily included infections at body sites other than the lungs, diagnoses acquired after study entry (date and time of onset and ICD-9 code), serum albumin, and whether patients were receiving enteral nutrition (type, maximum daily volume, start and stop dates) or parenteral nutrition (start and stop dates). At study exit, date and mode of exit (discharge from MICU, death, or diagnosis of pneumonia) were recorded.

Surveillance Cultures and Microbiologic Techniques

Specimens for culture were taken daily from both nares and the oropharynx using a sterile cotton-tipped swab. An attempt was made daily to obtain sputum or endotracheal aspirates for culture. Gastric aspirates were obtained daily from patients with nasogastric tubes.

Swabs from nares and oropharynx were transported to the laboratory in sterile tubes. Sputum specimens were collected in a sterile cup. Tracheal specimens were collected in sterile suction traps, which were then closed and transported to the laboratory. Specimens from nares and oropharynx, sputum, and tracheal secretions were cultured on blood, chocolate, and MacConkey's agar. Gastric aspirates were cultured quantitatively at serial 100-fold dilutions on blood, chocolate, and MacConkey's agar. Colonies were counted after overnight incubation, and the number of cfu/ml was calculated using plate counts and dilution factors.

PSB specimens were cultured according to standard methods (38). Undiluted thioglycolate broth eluate from the PSB and undiluted fluid obtained by PBAL were plated on blood, chocolate, and MacConkey's agar for aerobic cultures, and undiluted broth from the PSB was inoculated onto Schaedler's plain, Schaedler's KV, and Schaedler's CNA agar and into chopped meat broth for anaerobic cultures. Thioglycolate broth eluate from the PSB and PBAL fluid were cultured quantitatively in dilutions of 10¹, 10³, and 10⁵ on chocolate agar. Microorganisms were identified by standard techniques. Bronchoscopy specimens were also cultured on buffered charcoal yeast extract agar for Legionella species. Fluid from PBAL was examined for Legionella using a direct immunofluorescent stain.

Statistical Analysis

Data were entered onto computer forms designed for use with Relational Database Management Software (Oracle Corp., Redwood Shores, CA) on a cluster of three Digital VAX 6330 computers (Digital Equipment Corp., Maynard, MA). Data were entered on the computer forms daily. Prior to data entry, all computer forms were edited by one of us (D.L.G.). After data entry, a random 10% double entry of the data revealed an error rate of < 1%.

Data were analyzed using SAS software (Statistical Analysis System; SAS Institute, Inc., Cary, NC). Univariate analysis was carried out using the chi-square test and Fisher's exact test for categorical data and the t test for independent samples for continuous data.

Multivariable analysis was performed using logistic regression. Selection of variables for a logistic regression model began by screening groups of variables such as demographic variables, severity of illness variables, diagnoses, pulmonary diagnoses, variables related to respiratory failure, types of intubation, variables related to nutrition, neurologic disorders, medications, and colonization by various species of microorganisms. Microorganisms were grouped for analysis. Gram-negative organisms were grouped according to the following classes: (1) enterics: all Enterobacteriaceae; (2) respiratory gram-negatives: Hemophilus sp.; (3) Pseudomonas aeruginosa; (4) other nonfermenters: Acinetobacter, Burkholderia, and Stenotrophomonas species. Continuous variables were analyzed for linear trend. The p value for entry of variables into each of the group models was set at 0.3. The variables retained in each of these models were used for further analysis by logistic regression. At each stage of development, the participating investigators met to determine whether variables in the models were biologically plausible.

The potential reduced sensitivity of protected bronchoscopic techniques for the diagnosis of VAP in the presence of antibiotic therapy (9, 39, 40) was addressed in the data analysis. Patients whose protected bronchoscopic procedures yielded one or more microorganisms in a concentration  than the diagnostic threshold were assumed to have VAP, whether or not they were receiving antibiotics at the time of bronchoscopy. Likewise, patients who were not receiving antimicrobial therapy at the time of bronchoscopy and who had a negative culture result were assumed not to have pneumonia. Fifteen patients had negative cultures of protected bronchoscopic specimens while receiving antimicrobial therapy. By study protocol, no patient was receiving antibiotics for suspected pneumonia at the time of bronchoscopy. To control for potential reduced sensitivity of bronchoscopy cultures in patients receiving antibiotics, the univariate and multivariable analyses were first carried out on the total population and then repeated with the latter 15 patients excluded.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Incidence Rates

Six hundred ninety-seven patients were admitted to the MICUs during the time of the study. Three hundred eleven patients were excluded because of MICU stay of  48 h and 28 were excluded for pneumonia at the time of admission. The 339 excluded patients were significantly younger, had significantly fewer diagnoses, had significantly lower APACHE II and TISS Scores, and were significantly less likely to have had mechanical ventilation compared with the 358 patients who remained in the analysis. Two hundred twenty-three (62.3%) of the remaining 358 patients required mechanical ventilation.

Pneumonia was diagnosed in 28 patients, all but one occurring in ventilated patients. Sixty-seven patients met the criteria for bronchoscopy during the study, including four nonventilated patients. Twenty-six of the 67 patients (38.8%) had pneumonia diagnosed by PSB and/or PBAL. Of the other two cases of pneumonia, one was diagnosed by histopathologic examination of lung tissue taken at autopsy, and one was diagnosed by rapid progression of a pulmonary infiltrate to cavitation in the absence of carcinoma of the lung. Eleven (40.7%) of the 27 VAP cases occurred after  5 d of ventilation.

The cumulative incidence of pneumonia was 7.8%, with a 95% confidence interval (CI) of 5.4 to 11.3%. The cumulative incidence of pneumonia for patients who received any mechanical ventilation was 12.1% (95% CI, 8.3 to 17.6%) and for patients who received no mechanical ventilation the rate was 0.74% (95% CI, 0.13 to 4.2%).

The incidence rate of pneumonia was 12.5 cases per 1,000 patient days (95% CI, 8.7 to 18.1). The incidence rate for patients who received any mechanical ventilation was 16.5 cases per 1,000 patient days (95% CI, 11.3 to 23.9), and for patients who received no mechanical ventilation, it was 1.7 cases per 1,000 patient days (95% CI, 0.3 to 9.6). Because patients who were in the MICUs for less than 48 h were excluded from the study, the incidence rate was also calculated after excluding the first 2 d for each patient who remained in the study. Calculated in this manner, the incidence rate was 18.5 cases per 1,000 patient days (95% CI, 12.8 to 26.9). The pneumonia incidence rate per 1,000 d of mechanical ventilation was 20.5 (95% CI, 14.1 to 29.9).

Causative Organisms

Forty microorganisms were isolated in significant quantity by protected bronchoscopic techniques from 26 patients with VAP (Table 1). Ten of 26 cases (38.5%) had a polymicrobial etiology. Staphylococcus aureus, Streptococcus pneumoniae, Hemophilus species, and Pseudomonas aeruginosa made up 65% of all isolates from the cases of pneumonia and were implicated in the etiology of 21 of the 26 cases (80.8%) for which pathogens were identified. Enteric gram-negative bacilli were relatively infrequent as pathogens, accounting for 12.5% of isolates. Only two cases were caused solely by enteric gram-negative bacilli. The distribution of organisms between early-onset ( 5 d) and late-onset VAP followed the usual pattern (41, 42), with S. pneumoniae and Hemophilus sp. predominating in the early-onset and Pseudomonas and methicillin-resistant S. aureus in the late-onset VAPs. All MRSA pneumonias occurred as late-onset VAP (30), whereas two cases of methicillin-sensitive S. aureus also occurred after 5 d of ventilation (6 and 7 d postintubation).

The results of daily surveillance cultures of both nares, oropharynx, trachea, and stomach for patients who developed pneumonia are shown in Table 2. Thirty-one pathogens were isolated from one or more surveillance cultures prior to the diagnosis of VAP. In 29 of 31 instances (93.5%), the isolate was recovered from cultures of tracheal secretions prior to the onset of pneumonia. All other sites were colonized much less frequently, including nares (41.9%), oropharynx (41.9%), and stomach (35.5%). The trachea was the first site colonized for 48.4% of pathogens isolated from surveillance cultures. Tracheal colonization was preceded by gastric colonization in only four instances. If recovered from surveillance cultures prior to the onset of VAP, S. pneumoniae and Hemophilus species always appeared in the tracheal aspirate first. S. aureus often appeared at other upper airway sites before or at the same time it was cultured from the trachea. Patterns of colonization for P. aeruginosa were more variable.

Risk Factors for VAP

Variables examined as possible risk factors for VAP by univariate analysis are shown in Table 3. Results of the logistic regression analysis are shown in Table 4. Results of the univariate analysis and logistic regression analysis did not change when repeated after exclusion of the 15 patients without pneumonia who were receiving antibiotics at the time of bronchoscopy.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Nosocomial pneumonia is a common complication in medical ICU patients, primarily those requiring endotracheal intubation and mechanical ventilation. The cumulative incidence of VAP in our study was 12.1%. Fagon and colleagues (31) noted a similar cumulative incidence of 9% using a PSB to diagnose VAP (31). These rates are much lower than the 21 to 24% reported by most studies, which used less specific diagnostic criteria (22, 24, 43).

The accuracy and reliability of incidence rates are improved when factors for time and device are incorporated into the denominator. We observed an incidence rate of 20.5 cases of pneumonia per 1,000 ventilator days, or about 2% per day of mechanical ventilation. This is about twice the rate of pneumonia observed by the National Nosocomial Infection Surveillance (NNIS) system of the Centers for Disease Control and Prevention using clinical and roentgenographic diagnostic criteria in similar ICUs (Coronary, 9.8; Medical, 9.6 and Medical/ Surgical, 12.7 cases per 1,000 ventilator days) (Robert Gaynes, personal communication, 1995).

The most common pathogens causing VAP in our patients were S. aureus, P. aeruginosa, S. pneumoniae, and Hemophilus species; together, these microorganisms accounted for nearly two thirds of the isolates. Enteric gram-negative bacilli were relatively uncommon, accounting for only 12.5% of pathogenic isolates identified by protected bronchoscopic sampling. The majority of other studies using quantitative culture techniques (5, 12, 13, 15, 21, 24, 25, 31, 39, 40, 42) have also found that enteric gram-negative bacilli make up  25% of causative microorganisms in VAP (44), despite the finding that enteric gram-negative bacilli and yeast make up two thirds to three fourths of the flora in the stomach of patients receiving stress ulcer prophylaxis (45). Thus, focus on patterns of colonization for microorganisms other than enteric gram-negative bacilli may be important to define the microbiology of nosocomial pneumonia in critically ill patients.

Our surveillance culture data showed that tracheal colonization with the eventual pathogen(s) consistently preceded the development of VAP, usually by at least 3 d. Based on this observation, tracheal colonization may serve as the usual proximate reservoir from which pathogens are aspirated into the distal lung to cause pneumonia. The pathways by which these microorganisms reach the trachea appear to be variable, depending on the type of microorganism. Our data do not support the stomach as a common reservoir for pathogens causing VAP, consistent with other studies using bronchoscopic diagnosis (46). Tracheal colonization was preceded by gastric colonization for only four of 31 eventual pathogens. In six other instances, gastric colonization occurred before the diagnosis of pneumonia but at the same time or after tracheal colonization. Twenty of the 31 pathogens that were isolated from surveillance cultures before pneumonia were never isolated from the stomach. Our epidemiologic analyses also could not demonstrate a relationship between stress ulcer prophylaxis with antacids or H2 blockers and VAP (Table 3) (22, 47, 48).

Avoidance of misclassification is essential to clarify the epidemiology and microbiology of VAP in MICU patients. Our study attempted to achieve correct classification by utilizing protected bronchoscopic sampling and quantitative cultures for diagnosis. Because no diagnostic criteria for nosocomial pneumonia is 100% accurate, the protected bronchoscopic techniques used also may lead to misclassification bias. The most likely effect is incorrect classification of a patient as not having pneumonia, especially if receiving antibiotics. The majority of our patients with a clinical suspicion of VAP but negative bronchoscopy cultures had been receiving antibiotics for  3 d, decreasing the likelihood that the results represented false negative classification (39, 40). We did attempt to control for this effect by first including patients with suspected VAP but negative cultures receiving antibiotics in the analysis and repeating the analysis with these patients excluded. Results of the two analyses did not differ. The effect of this type of nondifferential misclassification on risk factor analysis is always to push conclusions toward the null hypothesis. Risk factors found in this study are therefore applicable even if a less stringent diagnosis is used. In addition, our results are directly applicable for ICUs that do use quantitative culture techniques to diagnose and manage VAP.

Using logistic regression analysis, we identified six risk factors for the development of VAP (Table 4). The association between mechanical ventilation and pneumonia was so strong that multivariable analysis of risk factors for the entire MICU population was precluded. The duration of mechanical ventilation was identified as an independent risk factor, similar to other multivariable (24, 27) and observational (31) studies. We found a dose-response effect for duration of mechanical ventilation, rather than a single threshold duration above which risk of VAP increased, suggesting a continuing risk for developing VAP so long as a patient continues to receive mechanical ventilation. Cumulative pack-years of smoking has not previously been identified as a risk factor, but it may reflect impairment of local host defenses that occurs with chronic lung disease, which has been reported as a risk factor (23, 24). Cumulative pack-years may provide a useful quantitation of a progressively more significant risk that is not provided by the simple presence or absence of a sometimes subjective diagnosis of chronic lung disease. PEEP has previously been reported as a risk factor (24), although only analyzed as present or absent and with borderline significance in the logistic regression model. High level PEEP as a risk factor for VAP may reflect the severity of respiratory failure, the presence of ARDS, or possibly the mechanical effects of PEEP on local defenses in the lung. A low serum albumin level probably reflects the adverse influence of a poor nutritional state on host defenses. Poor nutritional status has previously been reported as a risk factor for nosocomial pneumonia in hospitalized patients (49, 50) but not specifically in ventilated patients.

Absence of antimicrobial therapy as a risk factor for VAP has not been identified previously and, in fact, Kollef (27) found prior antimicrobial use was a risk factor for VAP. In addition to using different diagnostic criteria, the discrepancy may also be explained by a synergistic relationship with another previously unreported risk factor, colonization of the upper respiratory tract with respiratory gram-negative bacilli, primarily Hemophilus species. Patients receiving antibiotics had a significantly lower risk for VAP caused by gram-positive cocci and H. influenzae than did patients not receiving antibiotics (42). Antimicrobial therapy may be protective, particularly for early-onset VAP, by decreasing the local resevoirs of respiratory gram-negative bacilli, and perhaps pneumococci as well. Because Kollef (27) documented only two cases of Hemophilus and no pneumococcal VAP, the beneficial effect of antibiotics was outweighed by a predisposition to select for bacteria with a high degree of antimicrobial resistance characteristic of late-onset VAP (31, 42). Interestingly, Sirvent and colleagues (29) found that colonization at the time of intubation by Hemophilus or pneumococci was the only significant risk factor for early-onset VAP in a population of patients with structural coma and that antibiotic prophylaxis prevented early-onset VAP.

It is also noteworthy that several risk factors proposed in previous studies were not significantly related to VAP in our study. These included age (25, 27), naso-oroenteric tubes (26), depressed consciousness (25, 28), fall-winter season (22), severity of illness (31), and use of H2 blockers (22, 48). This discrepancy may be due to the use of different diagnostic criteria and study of different patient populations.

Design of appropriate prevention strategies is one of the main goals of epidemiologic studies, and results serve to validate the findings of the initial epidemiologic study. In a selected subpopulation but using similar diagnostic criteria to ours, Sirvent and colleagues (29) validated two of the results found in our epidemiologic studyabsence of antibiotic therapy and the presence of respiratory gram-negative organisms. The randomized component of their study demonstrated that two doses of cefuroxime prevented early-onset VAP in patients with structural coma. Their results contrast with the negative study of Mandelli and colleagues (51) in an overlapping population of patients with impaired airway reflexes. Although the differences may reflect slightly different patient populations and different antibiotic regimens, the contrasting studies support both the results and the need for an epidemiologic study based on quantitative cultures. Mandelli and colleagues did not use quantitative cultures to define their VAPs and may have obscured benefit by including patients with chemical aspiration pneumonitis as infectious pneumonia cases, which was avoided by Sirvent and colleagues by the use of protected NBBAL cultures.

Prevention of VAP by antimicrobial prophylaxis remains controversial. Our data suggest that antimicrobial prophylaxis may have a role. No prophylactic antibiotic strategy should be widely employed before the implications for colonization with resistant microorganisms have been fully explored. Emergence of colonization with antibiotic-resistant microorganisms has occasionally become a problem with other prophylactic antibiotic regimens (52). On the basis of the risk factors found in our study and the results of the study of Sirvent and colleagues (29), prophylactic strategies should probably be restricted to ventilated patients at high risk for early-onset VAP using short courses of narrow spectrum agents active against Hemophilus species and pneumococci. Extension of the results of Sirvent and colleagues (29) to other populations would still require separate randomized, controlled trials.

Several other risk factors found in our study may be amenable to intervention. Low serum albumin may be countered by optimizing the nutritional status of critically ill ventilated patients. Whether correction of hypoalbuminemia specificly or by nutritional intervention will change the risk of VAP remains to be proven by prospective randomized trials. Neither enteral nutrition (53) nor enteric feeding tubes (26) independently predisposed patients to VAP in our study, suggesting that the enteral route is preferred.

Decreasing the risks for pneumonia related directly to mechanical ventilation may be difficult. Intubation itself is the greatest risk factor and can be partially avoided with the use of noninvasive positive pressure ventilation (54), especially in patients with chronic lung disease who are at additional risk for VAP and are frequently colonized with respiratory gram-negative bacilli. New pharmacologic and ventilatory manipulations offer the potential to not only lower the level of PEEP but also shorten the duration of ventilation in patients with ARDS (55, 56). The length of ventilation also appears to vary from hospital to hospital, even when adjusted for equivalent predicted duration of ventilation (57). Any strategy to accelerate the process of weaning from mechanical ventilation may lead to fewer VAPs.