INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Although numerous studies have shown that ventilator-associated pneumonia (VAP) frequently complicates the course of patients in the intensive care unit (ICU) (1), the question of whether or not critically ill patients are at increased risk of death or prolonged ICU stay as a result of acquiring pneumonia remains controversial. Studies of fatality rates among patients with VAP as compared with patients without VAP have generated the contradictory conclusions that VAP does (6) and does not (1, 9) increase morbidity and mortality. Other studies, using univariate analysis, have suggested that VAP is associated with an increased risk of fatality, but this association was not evident in multivariate analyses (3, 10). The use of stepwise logistic regression to control for effects of confounding variables has shown VAP to significantly increase the risk of death in the ICU (5, 11). Using a rigorous method of matching patients with pneumonia to patients without pneumonia, one study concluded that VAP probably does increase the risk of death (12), whereas another study found that VAP does not increase mortality (13).

Previous analyses of the attributable morbidity and mortality of VAP have yielded conflicting results. Diverse study designs, modest sample sizes, and different definitions of VAP have made interpretation of this literature challenging. Variables that may influence the extent to which VAP increases morbidity or mortality include the affected patient population, timing of onset of pneumonia, diagnostic strategy, causative organism, and adequacy of initial therapy. We determined the clinical consequences of VAP in a large, prospectively collected data base of ICU patients at risk for VAP. Our objective was to determine, with a rigorous matching scheme, the attributable length of stay and mortality associated with VAP.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Our study was conducted in the context of a multicenter, randomized trial in which the effects of sucralfate and those of ranitidine were compared in mechanically ventilated, critically ill patients (14). In 15 university affiliated ICUs across Canada, patients ventilated for more than 48 h who had no evidence of pneumonia or gastrointestinal bleeding on admission were enrolled, and were followed daily until death or discharge. Age, sex, admission diagnosis, location prior to the ICU, and admission acute physiology and chronic health evaluation II (APACHE II) score (15) were recorded at baseline. Patients were monitored daily for the development of gastrointestinal bleeding and VAP; multiple organ dysfunction (MOD) scores were also recorded daily (16). Ethics approval was obtained from the institutional review board of each participating site, and informed consent was obtained from patients or their surrogates.

All study patients who had clinically suspected pneumonia were to undergo bronchoscopy with either collection of a protected brush catheter (PBC) specimen or bronchoaveloar lavage (BAL) to diagnose VAP. Clinically suspected pneumonia was defined as the finding of radiographic features suggestive of pneumonia with no other obvious cause, and the presence of any two of the following: fever > 38° C, leukocytosis (> 11.0 × 10 ⁹/L), or relative neutropenia (< 3.5 × 10 ⁹/L), purulent sputum or a change in character of sputum, isolation of pathogenic bacteria from sputum, and an increasing alveolar-arterial oxygen gradient. An adjudication committee, blinded to patients' treatment allocation and study outcomes, reviewed pertinent clinical, microbiologic, laboratory, and pathologic reports to determine the final outcome status of all patients with clinically suspected pneumonia. In some cases, patients were assessed as having VAP even though bronchoscopic techniques yielded less than significant growth on culture. Decisions of the adjudication committee about the presence or absence of VAP were based on the total constellation of signs, symptoms, and ancillary data. All events were given duplicate, independent reviews; differences were resolved by consensus. All decisions about VAP and general preventive and therapeutic patient management were made by the ICU team at each center.

Statistical Analysis

To determine the attributable morbidity and mortality of VAP, we utilized a matching procedure (similar to the procedure we used in a previous study [17]) with a weighting schema to ensure that patients with an adjudicated diagnosis of pneumonia were similar in every way except for the event of interest to patients who did not have VAP. Only patients who did not develop clinically suspected pneumonia were considered for the control group. The criteria used in our matching procedure were based on variables considered to be important determinants of length of stay and mortality. Given that the conflicting results of previous analyses (12, 13) may be partly explicable on the basis of differences in patient population, we also matched patients and controls on medical or surgical status upon admission to the ICU.

To determine attributable length of stay, we used 10 criteria to determine the best match for a case. A control had to match a patient in the following criteria in order to be considered a possible match: mortality status, medical/surgical status, time in ICU prior to the development of VAP, duration of mechanical ventilation prior to VAP, Day 1 APACHE II score (± 4 points), and MOD score on the day prior to development of VAP (± 3 points). For each control, the remaining four criteria were weighted with regard to their importance (to reflect the extent to which each criterion [covariate] might influence outcome): (1) ICU admitting diagnosis: 8 points; (2) age ± 15 yr: 4 points; (3 ) center at which treated: 2 points; and (4) gender: 1 point. The overall score was the sum of the points for which the control matched the characteristics of the case. Among the potential control subjects who met the "must match" criteria, the subject with the highest score was matched to the case. Thus, if one potential control subject had the same admission diagnosis as a case (8 points) and another matched the case on all the less important criteria (a total of 7 points), the control subject matching by admission diagnosis status had the higher score. In the case of a tie, the control subject with the closest APACHE II score was chosen. If there was still a tie, the control closest in age to the case was chosen. Once the controls were chosen, they were taken out of the pool of possible matches for the remaining cases. The additional length of stay attributable to VAP was estimated from the average difference in stay for all matched pairs (case versus control), which was tested statistically with a paired t test; this was also done separately for pairs that survived and those that died.

To determine mortality attributable to VAP, we used the same matching procedure as described previously, except that we deleted mortality status as a matching criterion. The attributable mortality was estimated as the proportion of the crude mortality that was attributable to VAP (i.e., (crude mortality_cases  Crude Mortality_controls)/ crude mortality_controls) reported as the relative risk increase (RRI) in mortality. The confidence interval (CI) for the attributable mortality and the statistical test for determining whether it was different from zero were based on the large sample variance proposed by Leu and coworkers (18), as derived from the work of Kuritz and Landis (19).

We also performed several a priori sensitivity analyses. To determine the influence of differing patient populations on VAP outcome, we performed two separate subgroup analyses in which we compared the results for medical and surgical patients and trauma and nontrauma patients. To determine the influence of timing of onset of pneumonia, we compared the results for patients with early-onset VAP (< 7 d) with those of patients with late-onset pneumonia (> 7 d after ICU admission). We used a cutoff of 7 d because it corresponds to the median time of occurence of pneumonia in our data base, and because mechanical ventilation lasting 7 d or more is associated with increased likelihood of infection with a "high risk" organism (20). To explore the effect of different diagnostic strategies for determining the presence or absence of VAP, we examined the results of the analysis when cases were defined by: (1) the adjudication committee; (2) a positive culture with a significant concentration of organisms from invasive bronchoscopic testing (> 10³ for the PBC technique and > 10⁴ for BAL); and (3) criteria for clinically suspected pneumonia as determined by the bedside intensivist. To determine the influence of different organisms associated with VAP, we classified causative organisms on the basis of their potential to develop multidrug resistance (as previously defined [20]). "High risk" organisms included Pseudomonas species, Acinetobacter, Stenotrophomonas, and methicillin-resistant Staphylococcus aureus. "Low risk" organisms included all others (we did not consider Candida a pathogen in our immunocompetent population). Additionally, to determine whether the adequacy of empiric antibiotic therapy (21) influenced attributable risk, we compared patients receiving "appropriate" and "inappropriate" empiric antibiotics. To determine the adequacy of empiric antibiotics, charts were reviewed independently and categorizations were made after dicussions among three of us (D.J.C., S.K., D.K.H.). In this determination we focused on whether a patient's empiric antibiotics had activity against each organism isolated, rather than examining dosing schedules. Initial antibiotics (prescribed within 48 h after the onset of clinically suspected of pneumonia) were considered appropriate if at least one such antibiotic had activity against the organism(s) that subsequently grew on cultures (except in the case of Pseudomonas, for which appropriate initial therapy required two antibiotics with activity against the organism).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Twelve hundred critically ill patients, recruited to participate in a randomized trial of stress ulcer prophylaxis, were followed daily for the development of VAP. We excluded 186 patients who died, were discharged, or had VAP occurring within the first 48 h after ICU admission. Of the remaining 1,014 patients, 250 (24.7%) developed clinically suspected VAP. Of these, 186 (74.4%) underwent bronchoscopy and had a specimen collected with the PBC technique or underwent BAL. After a review of pertinent patient data, 177 patients were judged to have VAP. The median duration from admission to the onset of VAP was 7 d (interquartile range: 5 to 10 d). Organisms implicated in the etiology of VAP are shown in Table 1.

Using our matching algorithm, we found suitable matches for 164 of the 177 (93%) cases of VAP in the attributable length-of-stay analysis, and for 173 of the 177 (98%) cases in the attributable mortality analysis. Baseline characteristics of patients with VAP and their matched controls were similar, and are shown in Table 2.

Our matching procedure estimated the ICU length of stay attributable to VAP to be 4.7 d (95% CI: 2.0 to 7.5 d) in those patients who survived, and 2.6 d (95% CI: 5.6 to 10.7 d) in patients who died, for an average of 4.3 d in all patients (95% CI: 1.5 to 7.0 d, p < 0.001).

The crude mortality rates for cases and controls were 41 of 173 (23.7%) and 31 of 173 (17.9%), respectively (absolute attributable mortality of 5.8%, 95% CI: 2.4 to 14.0). The RRI attributable to VAP was 32.3% (95% CI: 20.6 to 85.1%). The crude mortality rates for each subgroup are shown in Table 3.

In medical patients, the attributable ICU length of stay was greater (6.5 d versus 0.7 d, p < 0.04) and the attributable mortality was greater (65% versus 27.3% RRI, p < 0.04) than in surgical patients. Differences in baseline demographics and study outcomes between medical and surgical patients are shown in Table 4. There was no difference in attributable length of stay or mortality between trauma and nontrauma patients or between those with and early-onset and late-onset pneumonia (Table 5).

When different diagnostic criteria were used to ascertain the presence or absence of pneumonia, there was no statistically significant change in the results (Table 5). The attributable length of stay due to VAP associated with high risk organisms was longer than the attributable length of stay for VAP due to "low risk" organisms (9.1 d versus 2.9 d, p = 0.06), although the attributable mortality was similar in the two groups. There were no differences in attributable morbidity or mortality from VAP treated empirically with appropriate versus inappropriate antibiotics (Table 5).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In this prospective study, we compared patients with VAP and matched controls, and found that VAP is associated with an excess ICU length of stay and a trend toward increased risk of death. Our sample size allowed us to explore the influence of several covariates on attributable morbidity and mortality of VAP. The ICU length of stay and mortality were greater for medical than for surgical patients with VAP. The attributable length of ICU stay was greater for patients with high risk organisms than for those with low risk organisms. Results were similar for trauma and nontrauma patients, for patients with early versus late onset pneumonia, for three different VAP diagnostic strategies, and for VAP treated empirically with appropriate and inappropriate antibiotics.

The relationship between late-onset pneumonia, high-risk organisms, and inappropriate empiric antibiotics is difficult to clarify. Late-onset pneumonia may carry a higher risk of morbidity and mortality, probably because cases are often caused by resistant organisms, which are difficult to treat, and may result in delayed or ineffective antibiotic therapy. In several studies, VAP caused by Pseudomonas, Acinetobacter, and methicillin-resistant S. aureus has been associated with increased mortality (12, 22, 23). Likewise, recent observational studies suggest that the appropriateness of early empiric antibiotic therapy is an important prognostic factor in VAP (21, 24). However, the influence of time of onset of pneumonia per se, relative to other confounding factors such as multidrug resistance and/or inappropriate antibiotics, is unclear (9). In our analysis we were able to find an effect of high-risk organisms on outcome; however, the results of our sensitivity analysis do not show that patients with late-onset pneumonia had greater mortality than those with early-onset pneumonia. This latter finding is consistent with the findings reported by Mosconi and colleagues, who compared patients with early- and late-onset pneumonia and found a similar risk of death in both groups (25). Likewise, patients given inappropriate empiric antibiotics had outcomes similar to those of as patients receiving appropriate empiric therapy. However, given the small number of patients receiving inappropriate antibiotics (n = 31) in our study, the inferences from analysis of this subgroup weak and inconclusive. A multivariate analysis of a large cohort of patients with VAP would better determine the respective influence of time of onset of pneumonia, appropriateness of antibiotic therapy, and high risk organisms on outcome.

Six studies have utilized a matching procedure to determine the morbidity and mortality attributable to pneumonia in ICU patients (Table 6). Our results differ from those of Papazian and coworkers (13) and Baker and associates (27), but are consistent with those of other studies (12, 28) (Table 6). Fagon and colleagues (12) matched 48 ICU patients (predominantly medical) with VAP (diagnosed through BAL or the PBC technique) with controls for age, simplified acute physiology score, indication for ventilatory support, date of admission, and duration of ICU stay prior to onset of VAP. In this retrospective study, VAP was associated with an increased risk of death (odds ratio [OR]: 2.0; 95%CI: 1.61 to 2.49), and the mean length of stay of cases was 34 d, as compared with 21 d for controls (p < 0.02). The increased attributable mortality in Fagon and colleagues' study relative to the results of our study may be due to differences in empiric treatments and causative organisms. Previous research by Fagon and associates found that patients with VAP were more likely to receive inappropriate initial therapy (26) and/or to be infected with high risk organisms (20) than the patients in our setting.

Baker and colleagues retrospectively identified 62 trauma patients with clinically suspected pneumonia who underwent bronchoscopy with the PBC technique or BAL (27). Thirty patients with a positive BAL or PBC result were matched with patients of similar age, sex, severity of illness, and number of discharge diagnoses. In this population there was a trend toward prolonged ICU stay (20 d versus 15 d, p = 0.16), but no difference in mortality (24% versus 24%).

In another study, patients admitted to a surgical ICU and a medical/respiratory ICU who developed VAP were identified prospectively and matched with control patients. The only criteria used to match patients was admission to the same ICU during the same time period (28). Attributable duration of ICU stay was not reported. Mortality rates were significantly higher among patients with VAP than among controls (55% versus 5% for surgical ICU patients and 55% versus 7.5% for medical/respiratory ICU patients, p < 0.001 for both comparisons). The lack of adequate matching limits the inferences that can be drawn from this study.

Craig and coworkers (29) studied a large group of ICU patients (n = 670) and matched those that developed pneumonia in the ICU (n = 54) with those that did not on the basis of the following criteria: age (± 2 yr), sex, admission diagnosis, risk factors for pneumonia, and surgical procedure. They found that patients with pneumonia stayed three times longer in the ICU (12 d versus 4.3 d, p < 0.001) and had a 4-fold greater mortality than controls (20.3% versus 5.6%, p < 0.001). Not all study patients were ventilated, and the duration of ICU stay prior to onset of pneumonia was not considered in this study. The adequacy of the matching was questionable, since patients with pneumonia had more risk factors for pneumonia than did patients who did not develop pneumonia. In addition, only clinical criteria were used to diagnose pneumonia.

Kappstein and colleagues (30) prospectively followed 270 consecutive ICU patients for the development of pneumonia, matching cases (patients who developed VAP) with controls (patients who did not). However, they excluded cases in which the patient died (21 of 78, or 26.9%) and cases for whom no suitable control could be found (23% or 29.5%). The mean excess length of ICU stay was 10 d, but mortality attributable to VAP was not reported. That over half of Kappstein and colleagues' cases were not matched weakens the inferences one can make from their study.

Papazian and colleagues (13) examined the excess mortality and length of both ICU and hospital stay for 85 patients with VAP diagnosed through the PBC technique compared with controls matched on the basis of diagnosis, indication for ventilation, age, sex, and APACHE score. In addition, the control had to have been ventilated for at least as long as the case patient prior to the onset of VAP. Mortality was similar among cases (40%) and controls (38.8%). The duration of mechanical ventilation and ICU stay was significantly longer in cases than in controls (21 d versus 16 d and 26 d versus 21 d, respectively).

Differences in results across these studies may be partly explained by differences in patients, methods, and diagnostic strategies. As with to our approach, Papazian and coworkers did not include patients in their control group with clinically suspected pneumonia that did not have a confirmed diagnosis, whereas Fagon and associates included such patients as potential controls. Our matching strategy was similar to that of others (12, 13) in that we included prior duration of stay in the ICU as a matching criterion. However, our approach differed from that in all other studies in that we matched cases and controls according to daily assessment of organ dysfunction. We used the MOD score, a time-dependent variable that has been shown to discriminate patients who are likely to survive from those who are not (16). Matching only on the basis of variables measured at baseline may result in unequal groups, given that pneumonia, on average, occurs approximately 7 d after admission to the ICU. Patients developing VAP after their ICU admission may be at greater risk of infection because of persisting organ failure and severity of disease, whereas controls matched only according to disease severity at admission may have a more favorable evolution; thus assessment of the morbidity caused by an event occurring 7 d after admission may lead to overestimation of VAP-associated morbidity. Therefore, we used a time-dependent variable and measured its value more proximate to the onset of pneumonia in order to enhance the comparability of cases and controls. In this study we matched cases and controls according to severity of organ dysfunction on the day prior to clinical suspicion of pneumonia. Conversely, the selection of the day before development of clinically suspected pneumonia may be viewed as too close to the event studied, since cases may already have an increased risk of poor outcome because of incipient pneumonia, which may lead to underestimation of attributable risk. However, our results were similar when we matched cases and controls on the basis of severity of organ dysfunction 3 d before the onset of VAP (data not shown).

An important difference among the studies being discussed here is in the criteria used to diagnose pneumonia. In both the French studies (12, 13) and the study by Baker and colleagues (27), bronchoscopy with a protected brush catheter and a specimen that yielded > 10³ cfu/ml were required to make the diagnosis. In our study, 187 of 250 (75%) patients with clinically suspected pneumonia underwent bronchoscopy with either a protected brush catheter or BAL, of whom 87 (46%) had VAP confirmed according to the usual criteria for these techniques. However, the majority of patients were already receiving antibiotics at the time of diagnosis. Given the uncertainty about the accuracy of invasive diagnostic techniques in patients already receiving antibiotics (31, 32), we used an adjudication committee to determine the final diagnosis of pneumonia, but examined attributable length of stay and mortality according to three different definitions in the sensitivity analysis. We found that results for attributable length of stay and mortality were similar regardless of diagnostic approach. With a larger sample size, differences between the different diagnostic methods might be discernible. However, our findings are consistent with the finding by Timsit and colleagues that attributable morbidity and mortality were the same in patients with clinically suspected pneumonia and patients with a bacteriologically confirmed diagnosis (33).

Differences in the proportion of medical and surgical patients in each study may also explain discrepant results. The two studies that showed no attributable risk of dying enrolled no medical patients (27) and only 26% medical patients (13), respectively. This contrasts with 44% medical patients in Fagon and associates' study (12) and 61% in our study. On the basis of our sensitivity analysis, the attributable increase in length of ICU stay and mortality is largely seen in medical patients, with essentially no effect seen in surgical patients. This may be due to differences in the underlying disease processes and chronic health status of surgical and medical patients.

In summary, five of the six studies that used matching procedures to examine the attributable morbidity and mortality of VAP concluded that VAP does increase ICU length of stay. Only two of the studies did not find an excess mortality attributable to VAP (13, 27). In our study, the largest case-control study of mortality attributable to VAP, we found that VAP may increase the risk of death by about 33% (an absolute risk increase of 5%). Our sensitivity analysis suggests that attributable risk may be greater in medical patients and in patients infected with high risk organisms. The multicenter nature of our study enhances the validity and generalizability of our findings.

Understanding the true attributable mortality and length of stay associated with VAP allows physicians to provide accurate prognostic information to families of critically ill patients, identifies appropriate endpoints to be used in therapeutic and preventive trials in VAP, and better informs economic evaluations of interventions related to VAP. Given that VAP is associated with a substantial burden of illness for critically ill patients, studies aimed at explaining its underlying pathogenesis and evaluating approaches to its diagnosis, prevention, and treatment are warranted.