Is Ventilator-associated Pneumonia an Independent Risk Factor for Death?

This study was conducted in the medico-surgical ICU of Sainte Marguerite University Hospital, which is an 850-bed teaching hospital. Each year, an average of 455 patients are admitted to the ICU. The ICU is divided in 12 separated rooms and 3 isolation rooms. During a 3-yr period (from July 1, 1993, to June 30, 1996), the charts of 1,552 patients admitted to the ICU were reviewed. Patients included in the study were selected from a cohort of 475 patients who were not immunosuppressed and who required mechanical ventilation for more than 48 h. During the study period, 135 consecutive patients who died represented the case group. Control patients were selected among the 340 survivors. Ventilator circuit changes were performed every week for all patients. The patients were all ventilated in a semirecumbent position, and sucralfate was routinely used for stress-ulcer prophylaxis.

For all patients, data were prospectively entered into a computer program. Each episode of VAP was identified and recorded prospectively. The prospective surveillance using a nosocomial infection surveillance program is routinely performed in this ward and has previously been described in detail elsewhere. 13Briefly, for respiratory tract infections, a bacteriologic examination is routinely performed three times a week, even in the absence of suspicion of any infection, using quantitative endotracheal aspirate (EA) analysis. In case of suspicion of VAP, additional examinations are performed using PSB, bronchoalveolar lavage (BAL), mini-BAL, or protected telescoping catheter.

A pulmonary infection was defined as VAP if the onset was recorded after at least 48 h of mechanical ventilation and if the following criteria were present 14: (1) chest radiograph showing a new and persistent infiltrate consistent with pneumonia; (2) purulent sputum; (3) elevated temperature greater than 38.3°C or leukocyte count more than or equal to 10,000/μl or less than or equal to 1,500/μl; (4) alteration of gas exchange; (5) microorganisms isolated from at least one of the following samples: PSB, BAL, mini-BAL, protected telescoping catheter, and EA. The threshold for considering a sample positive was 10³colony-forming units per milliliter (CFU/ml) for PSB and mini-BAL and 10⁴CFU/ml for BAL and EA. Patients were considered as not having VAP when at least one clinical sign of VAP was missing with no positive culture and no curative antibiotic prescribed. When patients presented with a clinical picture of VAP but with microbiologic results under the necessary thresholds, they were not considered as having a VAP. To avoid delaying the treatment of suspected VAP, antibiotics were administered just after completion of sampling procedures according to previous routine EA cultures if positive (directed antibiotherapy). In case of sterile culture of previous systematic EA, empirical treatment was started. After 48 h, antibiotics were adapted to the definite results of the sampling procedures performed on the day of clinical suspicion. If at least one of the five techniques was positive, antibiotics were continued with possible modification according to the antibiogram. If all sampling procedures were negative, antibiotics were stopped.

Cases were matched to controls without knowledge of the presence or absence of VAP. The conditions necessary to constitute a matched pair were as follows: (1) same diagnosis on admission; (2) age difference less than or equal to 10 yr; (3) same sex; (4) admission within 1 yr; (5) APACHE II 15score ± 7 points; (6) ventilation of control patients for at least as long as the cases. The following system was used to select an adequate control for each case: the whole cohort was stratified according to the date of admission to the ICU and according to the diagnosis on admission. Patients were then stratified by age, using decades, and by gender. Finally, APACHE II score calculated within the first 24-h period in the ICU and duration of respiratory support were examined. When more than one potential control on the list appeared to be well matched with a case patient, the closest date of admission to the case was used to assign the control patient. When the case and the control differed by more than one matching criteria, the control was rejected.

Concerning VAP, the following variables were noted: date of onset of VAP, temperature, leukocyte count, results of sampling techniques, antibiotic use, and arterial oxygen pressure–inspired oxygen fraction ratio. We also recorded other factors that could have introduced bias into the matching method because they are recognized or are supposed to be risk factors for VAP, i.e. , patient preadmission status according to McCabe and Jackson classification, 16organ-system failures evaluated by the Organ System Failure score calculated during each day during hospitalization in the unit, 17previous antibiotherapy in the 10 days preceding the pulmonary infection, stress-ulcer prophylaxis, steroid treatment, patient preadmission origin, and type of surgery (scheduled or emergency). Appropriate initial antimicrobial therapy was considered when at least one effective drug was included in the initial antibiotic treatment of VAP. We also recorded the presence or absence of continuous hemofiltration, coinfection with cytomegalovirus, duration of mechanical ventilation, duration of hospitalization in the ICU, and the date of hospital discharge.

Results are expressed as mean ± SD. Statistical calculations were performed using the Statistical Analysis System software package (SAS version 5, SAS Institute, Cary, NC). 18The chi-square test was used to compare categoric variables. One-way analysis of variance or Wilcoxon signed-rank test (when analysis of variance could not be applied) was performed to compare demographic characteristics. Continuous variables were compared using the Student t  test for normally distributed variables and Wilcoxon rank sum test for non-normally distributed variables. Only VAP diagnosed during the same duration of mechanical ventilation as cases were taken into account in paired controls. In other words, if the case was ventilated for 14 days and the matched control for 30 days, only VAP diagnosed during the first 14 days was taken into account for the control patient. The McNemar test with continuity correction was used to compare the incidence rate of VAP between cases and controls. 19Two multivariate analyses, one using mortality as the outcome and one using the development of VAP, were performed using the logistic regression technique. A stepwise approach was used for entering terms into the model, with 0.10 as the limit for accepting and removing newly entered terms. For all statistical tests used, a P  value < 0.05 was considered statistically significant.

Among the 135 case patients, 108 were matched with a control, with 27 cases for whom no compatible control was found. These 27 unmatched case patients presented the following characteristics: age 66 ± 16 yr; APACHE II score on admission, 25 ± 9; and length of stay in the ICU, 27 ± 29 days. Details concerning the successful matching are reported in table 1. The overall effectiveness of matching was 91.3% (592 of the 648 matching criteria). Fifty-two patients were matched successfully on all six factors. Ten pairs consisted of control patients who were ventilated during a shorter period (at least 24 h) than cases. Only four of these pairs were discordant for the occurrence of VAP. However, VAP occurred in the case patient in three of these four pairs. The APACHE II score difference was greater than 7 in 9 case patients (difference of 8, n = 7 pairs; difference of 9, n = 2 pairs) and in 7 control patients (difference of 8, n = 6 pairs; difference of 10, n = 1 pair) (table 2). Unmatched characteristics are presented in table 3.

There were 39 (36.1%) case patients who developed at least one episode of VAP. The analysis of the controls over a period corresponding to the same duration of mechanical ventilation revealed exactly the same number of patients who developed VAP. Thirteen of the 27 (48%) unmatched case patients presented at least one episode of VAP during their stay in the ICU. The reason for the inability to find a control was the diagnosis in 12 cases, the duration of mechanical ventilation in 8 cases, and the APACHE II score in 7 cases. If 27 controls not developing a VAP during their ICU stay had been found for these 27 unpaired cases, the incidence of VAP would have been 38.5% for the cases and 28.9% for the controls. The difference would not have been statistically significant. If the entire period of mechanical ventilation of controls was considered, 16 more patients developed VAP. During the study period, 25% of the 340 patients mechanically ventilated more than 48 h and who survived developed at least one episode of VAP.

There were 23 concordant pairs (both case and control agreed concerning the presence or absence of VAP) and 32 discordant pairs. In 16 of these discordant pairs, VAP was diagnosed in the control, and in the 16 other pairs, VAP was diagnosed in the case. Statistical analyses confirmed the absence of difference (odds ratio = 1), i.e. , VAP did not appear as a risk factor for death.

Multivariate analysis took into account all of the variables used for the matching process and some additional variables that could affect outcome such as all organ failures that appeared during ICU stay, the development of VAP, appropriate initial antimicrobial therapy for the treatment of VAP, prior corticosteroid treatment, cancer, immunosuppressive agents, the use of antibiotics for infections other than VAP (before the diagnosis of VAP in patients developing VAP), and the presence of potentially resistant bacteria responsible for VAP. This stepwise logistic regression showed that renal failure, bone marrow failure, and treatment with corticosteroids were independently associated with patient mortality (table 4).

The stepwise logistic regression analyzing risk factors of VAP such as age, sex, duration of mechanical ventilation, diagnosis, APACHE II score, Organ System Failure score, sucralfate, prior corticosteroid treatment, cancer, immunosuppressive agents, and the use of antibiotics for infections other than VAP (before the diagnosis of VAP in patients developing VAP) concluded that only the use of antibiotics was independently associated with the risk of developing a VAP (odds ratio, 2.69; 95% confidence interval, 1.44–5.03;P = 0.002).

There was no difference between cases and controls concerning the diagnostic criteria for VAP (table 5). Appropriate initial antimicrobial therapy was administered to the patients during the 48-h period after the suspicion of VAP in 79% of the cases and in 82% of the controls. Initial antimicrobial therapy was based on the results of the previously performed routine EA in 28 cases and in 27 controls presenting with VAP. Empirical treatment was therefore prescribed in the remaining 11 cases and 12 controls presenting with VAP. All patients initially received a combination of antibiotics. This antimicrobial therapy was considered inappropriate in 8 of the 39 cases and 7 of the 39 controls presenting with VAP. If only initial antimicrobial therapy based on the results of the previously performed routine EA was considered, antibiotherapy was appropriate in 26 of the 28 cases and 24 of the 27 controls receiving an EA-directed antimicrobial treatment. The duration of antimicrobial therapy was greater than 2 weeks in 35% of the cases and 41% of the controls. EA was performed in all cases and was always associated with at least one of the other sampling techniques. We recorded a positive result for EA in 28 of the 39 incidences of VAP in the case group compared with 29 of the 39 incidences of VAP in the control group. There was also no statistically significant difference between cases and controls concerning the number of positive mini-BAL (14 and 14, respectively), protected telescoping catheter (18 and 16, respectively), PSB (19 and 21, respectively), and BAL (17 and 16, respectively). Details on culture results of the sampling procedures at different thresholds are reported in table 6.

Concerning VAP diagnosed in cases, a total of 56 microorganisms were identified (table 7). The mean delay between the beginning of mechanical ventilation and the onset of infection was 12.5 ± 12.8 days. Forty-one percent were gram-negative bacilli. Potentially resistant bacteria (Pseudomonas  aeruginosa , Acinetobacter baumanii,  methicillin-resistant Staphylococcus aureus , or Stenotrophomonas maltophilia ) represented 34% of the bacteria recovered in cases and 26% in controls (nonsignificant). Microorganisms producing extended spectrum β-lactamases were identified in 4 cases and 8 controls.

A review of the literature on VAP reveals that only a few investigations focusing on patient prognosis have sufficiently taken into account the numerous risk factors for death present in this kind of severely ill population. However, only studies that consider such “confounding factors” (i.e. , using a matching design or a multivariate analysis) can really make interpretable conclusions on prognosis. To our knowledge, all of the studies using this kind of method and a precise definition of VAP except one group 6,12found no overmortality attributable to VAP. Numerous studies that were not especially designed to analyze patient outcome reported a higher fatality rate for patients developing VAP. However, these studies did not take into account confounding factors on prognosis and should therefore not be used to argue this point. Among studies using a matching procedure, all case-control studies matched patients with VAP to patients without VAP. 5,12,20,21 

The current case-control study on VAP is the first that matches deceased patients and survivors. The advantage of such a design is that the case refers to a specific outcome that does not suffer from the drawback of a controversial definition such as the definition of VAP. VAP can therefore be described in detail. Thus, if different readers have different definitions of VAP, particularly concerning the positive threshold of microbiologic samples, they can find food for thought in the details of our results (table 6). When a cohort analysis is performed, it is necessary to classify the patients (with VAP, without VAP) using a more or less pertinent diagnostic criteria before the beginning of the matching process. When a case control is designed with the outcome as the criteria differentiating cases from controls, it is possible to analyze the incidence of VAP after the matching process, taking into account more or less pertinent criteria.

This study finally showed the same VAP rate for cases and controls (36.1%). Using a multivariate analysis, Kollef 22showed from a cohort of 277 patients that mortality was independently associated with the number of organ system dysfunctions, Patient Preadmission Lifestyle score, and head positioning, but not with VAP. Another multivariate analysis has been conducted by Timsit et al.  23in 387 patients. The investigators introduced both clinically suspected VAP and bacteriologically confirmed VAP in a Cox model and failed to demonstrate an overmortality for bacteriologically documented VAP. The case-control analysis conducted by Baker et al.  20matched 29 trauma patients with VAP documented on BAL or PSB with 58 controls without VAP. No mortality was attributable to pneumonia in that work, whereas age and severity of injury were controlled using a matching process. Fagon et al.  6,12reported an increased fatality rate attributable to VAP using two different methodologies: a multivariate analysis and a case-control study. They described VAP documented on PSB at a threshold of 10³CFU/ml or BAL at a threshold of 10⁴CFU/ml, which were the same criteria used in other studies that found opposite results. 5,20,21,23Nevertheless, several differences can be found that can explain that this group found an increased mortality attributable to VAP. Their populations appeared to be severely ill according to their APACHE II scores, which were more than 24, compared with other study populations in which this score was approximately 20. They included more postoperative patients than we did and did not exclude immunosuppressed patients. Patients with postoperative complications after cardiac surgery are known to represent a selected population commonly excluded from studies on prognosis 15and have been shown in a specific study to have an increased mortality rate if nosocomial infection develops. 11Such patients represented a large part of the studied population (20%) in the reports by Fagon et al.  6,12In addition, it is possible that some differences concerning patient characteristics or therapeutic management not specified in these studies could explain these conflicting results.

In addition to studies especially designed to investigate patient prognosis, numerous other well-conducted reports lead readers to conclude on the absence of increased mortality linked to VAP. For example, several prospective randomized controlled trials evaluating some therapeutic approaches for the prevention of VAP 24–27or the prevention of stress ulcer in ventilated patients 28have reported similar outcome for patients with and without VAP.

From the initial cohort of 475 patients who required more than 48 h of mechanical ventilation, our analysis concerned 108 pairs. Concerning risk factors for death, some have been shown in well-designed studies as major elements in ICU patients’ prognosis: APACHE II score, 15duration of mechanical ventilation, 8age, 10Organ System Failure score, 17,22and nonscheduled surgical status. Some factors have been reported to be prognostic factors for patients who developed VAP. They include age, 2underlying condition, 6shock, noncardiac surgery, worsening acute respiratory failure, 29clinical presentation (severe or uncomplicated), 21and Pseudomonas  or Acinetobacter  species. 2,12Several investigators have suggested that the adequacy of initial antimicrobial therapy is an important factor determining outcome. 30–33In two of these latter studies, 30,31fewer than 30% of the patients received appropriate initial antimicrobial therapy. In the current study, 79% of the cases and 82% of the controls presenting VAP received appropriate initial antimicrobial therapy. We found that the adequacy of appropriate initial antimicrobial therapy was not an independent risk factor for death. The wide discrepancy concerning the adequacy of initial antimicrobial therapy observed between these previously published studies and the current work could be explained by the diagnostic strategy. Indeed, the choice of initial antimicrobial therapy is based on routine endotracheal aspirates analysis performed before or during the appearance of clinical signs of pneumonia. Therefore, the initial antimicrobial therapy is not empirical for the great majority of patients.

The diagnostic criteria proposed to define VAP are still a controversial issue. We used all of the commonly accepted blind and directed, invasive and noninvasive, protected and nonprotected techniques. For the definition of a positive culture, we used the thresholds that were validated by experts in reference studies using a histologic gold standard. 34–38 

In conclusion, it seems that there are enough arguments in the literature today to say that VAP does not have an overall effect on patient outcome. This could be a result of the effectiveness of antibiotics, management of mechanical ventilation, or improvement in hemodynamic and other organ-failure supports. Nevertheless, the fact that VAP does not appear as a risk factor for death must be interpreted with caution. Indeed, in some subgroups, such as immunosuppressed patients or patients suffering complication after cardiac surgery, VAP could worsen outcome per se .