Introduction

Ventilator-associated pneumonia (VAP) is the most frequent nosocomial infection in the intensive care unit (ICU) [1–4]. Although its attributable mortality remains debated, VAP is associated with increased lengths of ICU stay and mechanical ventilation [2, 4, 5]. It is then of interest to diagnose and to treat VAP adequately, as early as possible. The “clinical option” [1] is to treat every patient clinically suspected of VAP with broad-spectrum antibiotics, given clinical signs and positive endotracheal aspiration (non-quantitative culture). Indeed, any delay in treating severe sepsis results in increased mortality [6–8]. However, the “clinical option” [9–12] is poorly specific and may result in excessive antibiotic use and in increased prevalence of antibiotic resistance [13].

The “bacteriological option” [1] is based on the quantitative culture of a distal sample of alveolar fluid, which can be obtained with bronchoscopic [bronchoalveolar lavage (BAL) or protected specimen brush] or non-bronchoscopic methods (protected telescoping catheter, PTC). Although the introduction of antibiotics in this option is still based on clinical grounds and is sometimes difficult to perform in a 24-h workday, it improves the specificity of the diagnosis and allows more changes of antibiotic management [14].

Simple criteria allowing pneumonia to be confirmed or ruled out at an early stage of management would thus be clinically useful. An option combining clinical score and early bacteriological results may improve the sensitivity/specificity of VAP diagnosis [15]. Procalcitonin, the prohormone of calcitonin, is ubiquitously secreted in case of bacterial infection [16]. In case of suspected community-acquired pneumonia, procalcitonin assays serve to confirm the diagnosis but also to guide the duration of treatment [16–18]. In VAP, procalcitonin seems to be a strong indicator of recovery [19].

We therefore designed this prospective study to evaluate the respective and combined usefulness of the Clinical Pulmonary Infection Score (CPIS), BAL gram staining, endotracheal aspirate surveillance culture, and BAL and serum procalcitonin for ICU patients who were suspected of VAP. Our hypothesis was that combining clinical and microbiological parameters were the best tests to anticipate the diagnosis of VAP.

Methods

This prospective study was performed in the medical ICU of Sainte-Marguerite University Hospital in Marseille, France, during an 8-month period. Because no supplementary blood samples were drawn and investigations (BAL) were considered part of routine clinical practice, no written informed consent was required, as confirmed by the Clinical Research Ethics Committee of the Institut Fédératif de Recherche 48, Marseille, France. Patients or their next of kin were informed of their inclusion in this study and could refuse to participate.

Study setting and population

All consecutive patients (18 years or older) who were ventilated 48 h or more were prospectively included if VAP was clinically suspected: presence of a new infiltrate or progression of a prior stable infiltrate on the chest radiograph in association with two of the four following criteria: fever >38°C or hypothermia <36°C; leukocytosis >10 × 109/l or <4 × 109/l; purulent tracheal secretions; decrease in the PaO2/fraction of inspired oxygen (FiO2) ratio [1]. Fiberoptic bronchoscopy examination was performed in each patient within 12 h of the suspicion of VAP.

Diagnosis of VAP

One of the investigators (NE, BJ, or LP) made daily rounds in the ICU to identify eligible patients, to determine the onset of VAP based on the diagnostic criteria described below, and to record relevant data from the medical records, bedside flow sheets, and the hospital’s mainframe computer for reports of microbiological studies and bacterial antibiotic sensitivity profiles. When VAP was suspected, it is our routine care to perform fiberoptic bronchoscopy with BAL before any modification or introduction of new antibiotics.

All chest radiographs were analyzed prospectively by two (BJ, NE) of the investigators. The tip of the bronchoscope (Olympus BF P20 D, Olympus, New Hyde Park, NY) was introduced without any bronchial suctioning and was positioned in the most prominent radiographic infiltrate. The bronchoscope was wedged in the bronchial orifice of a lung segment, and BAL was performed by infusing three 50-ml aliquots of sterile 0.9% NaCl. The first aliquot, which was used as a bronchial wash, was discarded [20, 21], and subsequent aliquots were pooled, mixed, measured and then strained through cotton gauze to remove mucus. A gram stain of a cytocentrifuged layer was performed, and organisms were identified as gram negative bacilli and/or gram positive cocci. Bacterial identification and antibiotic susceptibility tests were performed using standard methods. Diagnosis of VAP was established when BAL quantitative culture grew with at least one microorganism at a concentration ≥104 colony-forming units (cfu) ml−1 associated with clinical signs (all included in the CPIS) that could be related to VAP (Table 1; [22]). Viral and fungal pneumonias were excluded from the analysis. All episodes of bacterial VAP in a same patient were taken into account. Early onset VAP (≤5 days) was differentiated from late-onset pneumonia (>5 days after intubation and mechanical ventilation).

Table 1 Modified Clinical Pulmonary Infection Score, adapted from [22]
Full size table

Baseline assessment and data collection

Each patient’s hospital chart was constituted prospectively, and the following data were recorded on ICU admission: age, sex, severity of underlying medical condition stratified according to the criteria of McCabe and Jackson, Simplified Acute Physiology Score II (SAPS II) [23], Sepsis-related Organ Failure Assessment (SOFA) score [24], the presence of co-morbidities and reason for mechanical ventilation.

In addition, on the day of sampling, we collected SOFA score and the results of routine surveillance culture of endotracheal aspirate (EA) performed twice weekly (Tuesday and Friday morning) in all mechanically ventilated patients as previously described [25]. We also calculated the CPIS score modified by Luna (Table 1). We formulated the CPIS-EA score, which combined the CPIS score modified by Luna et al. (mCPIS) [22] and the quantitative culture of routinely performed endotracheal aspirate. No point was added to mCPIS when EA culture was <105 cfu ml−1, and 2 points were added when EA culture was ≥105 cfu ml−1. Total points for CPIS-EA score therefore varied from 1 to 12 points. We also formulated the CPIS-BAL gram score, which was a combination of mCPIS and direct examination of BAL fluid. No point was added when BAL fluid gram staining was negative, and 2 points were added when BAL fluid gram staining was positive. Total points for CPIS-BAL gram score also varied from 1 to 12 points.

Estimation of the probability of VAP

The probability of pneumonia using the clinical picture and the mCPIS was prospectively estimated by two ICU senior physicians at the time of sampling on a 0–100% scale. They were blinded to the definitive results of the BAL culture at that time. The potential improvement over this estimate that was provided by serum procalcitonin was assessed. Finally, the improvement in diagnostic accuracy was assessed after obtaining BAL direct examination results. This estimate was subsequently categorized as “very low” when the probability was scored less than 20%, “low” between 20 and 40%, “moderate” between 40 and 60%, high between 60 and 80%, and very high when between 80 and 100% [15]. Patients were classified as probably having a VAP according to the physicians’ opinion when the combination of the two ICU physicians’ evaluations was higher than 60%. Patients under this threshold were classified as probably not having VAP.

Procalcitonin assay

For all patients, serum collected from blood drawn the day of suspicion of VAP was frozen for further analysis. BAL samples for procalcitonin determination were centrifuged (1,500 rpm, 10 min), and supernatants were immediately stored at −80°C until measurements. An investigator blinded to clinical data used TRACE (time-resolved amplified cryptate emission) technology on a Kryptor analyzer (Brahms Diagnostica, Berlin, Germany). The Kryptor analyzer detection limit in 100 μl of both serum and alveolar fluid was 0.019 ng ml−1, and sensitivity (interassay variation coefficient, 20%) was 0.06 ng ml−1. The 95th percentile reference was 0.064 ng ml−1.

Statistical analysis

Data are expressed as mean ± SD for normally distributed data and median with interquartile range (IQR) for non-normally distributed data. Continuous variables were compared using Student’s t test for normally distributed variables and the Mann-Withney rank-sum test for non-normally distributed variables. The chi-square test or Fisher's exact test was used to compare categorical variables. To evaluate the amount of procalcitonin in BAL and blood, we individualized two groups of patients: patients with a suspicion of VAP but a negative BAL, and patients with a confirmed VAP (BAL positive).

Results

Patient population

Over an 8-month period, 86 BAL procedures for suspicion of VAP were performed in 57 prospectively included consecutive patients. The clinical characteristics of the 57 patients are reported in Table 2. One episode of VAP was suspected in 35 patients, two episodes were suspected in 17 patients, three episodes in 3 patients and four episodes in 2 patients.

Table 2 Characteristics of the patients on ICU admission
Full size table

While a VAP was clinically suspected in 86 cases, it was bacteriologically confirmed by BAL culture results in 48 cases (56%). Fourteen of these 48 episodes were classified as early onset. Characteristics of the patients at the time of BAL are reported in Table 3. Patients were already on antibiotics in 36 of the 86 cases (42%) when BAL samples were obtained. In only six cases had antibiotic change occurred in the preceding 3 days. Fifty-four microorganisms were considered responsible for VAP. Forty-one (85%) were gram negative bacilli (P.aeruginosa in 37% of the cases and Klebsiella sp. or Enterobacter sp. in 17 and 10% of the cases, respectively). Thirteen gram positive cocci were isolated (11 methicillin-susceptible S.aureus, 1 methicillin-resistant S.aureus and 1 coagulase negative Staphylococcus). There was no difference for length of mechanical ventilation after the first suspicion of VAP according to the result of BAL (22 [13–30] days and 13 [7–25] in the VAP group and in the no VAP group, respectively). Among the 38 patients considered not to have VAP, antibiotics were discontinued when BAL culture was available in 13 cases (34%), never introduced in 8 cases (21%), continued for extrapulmonary superinfection in 12 cases (32%) and stopped because of dying patients in 5 cases (13%) (none because of identified VAP).

Table 3 Characteristics of patients the day of BAL
Full size table

Operative indices of the tested criteria of VAP

As shown in Table 4, the diagnostic accuracy of the mCPIS ≥5 was only 58% [95% confidence interval (CI) 48–68] with a sensitivity of 75% and a poor specificity. Gram stains of BAL specimens were positive in only 19 of 48 (40%) confirmed VAPs and in 4 of 38 (10.5%) non-confirmed VAPs. Serum and BAL procalcitonin had poor diagnostic accuracy because of the very low specificity for reduced thresholds and the rapid decline in sensitivity while increasing threshold (Fig. 1).

Table 4 Operative characteristics for the diagnostic markers of VAP in patients presenting a suspicion of VAP
Full size table
Fig. 1
figure 1

Procalcitonin in the diagnosis of ventilator-associated pneumonia. No significant difference in plasma (Fig. 1a) or alveolar procalcitonin (Fig. 1b) was observed among groups

Full size image

Combining mCPIS with either a positive direct examination of the BAL (mCPIS-BAL-gram) using a threshold of 6 or a previous EA culture ≥105 cfu ml−1 (mCPIS-EA) improved specificity from 37% for mCPIS to 61% for the two combined indices (P = 0.001), while sensitivity remained acceptable (71% and 83%, respectively) (Table 4).

Diagnostic accuracy of mCPIS-EA at a threshold of 6 was higher than that of the combination of mCPIS ≥5 with a serum procalcitonin level of at least 0.2 ng ml−1 (P = 0.026), but did not differ from mCPIS-BAL-gram (P = 0.25). Finally, mCPIS-EA and mCPIS-BAL-gram areas under the ROC curves were significantly larger than the areas of both serum (P = 0.001 and P = 0.002, respectively) and BAL procalcitonin levels (P = 0.001 for both) (Fig. 2). There was no difference between gram negative and gram positive VAP regarding alveolar or serum procalcitonin concentrations (data not shown).

Fig. 2
figure 2

Receiver-operating characteristics of modified CPIS, alveolar and plasma procalcitonin, modified CPIS combined with both BAL gram staining (CPIS-BAL gram) and endotracheal aspirate (CPIS-EA). Modified CPIS combined with both the results of gram staining and endotracheal aspirate have the best area under the ROC curve to predict a VAP episode

Full size image

The probability of pneumonia was estimated by physicians at the time of sampling (blinded to the definitive results of BAL culture) according to the clinical picture with the modified CPIS with an accuracy of 57% (49 of 86 VAP suspicions). This diagnostic performance was not improved by the addition of the serum procalcitonin result [accuracy 55% (47 of 86)]. When the clinical picture including modified CPIS and procalcitonin was combined with direct examination of BAL, it significantly improved the diagnostic performance (accuracy of 79%, 68 of 86 VAP suspicions; P = 0.003 vs. clinical picture with modified CPIS, P = 0.001 vs. clinical picture with modified CPIS, and serum procalcitonin). However, the diagnostic accuracy of the association of clinical picture with modified CPIS, serum procalcitonin and direct examination of BAL was similar to that of the last routinely performed endotracheal aspirate at a threshold of 105 cfu ml−1 (77%). When clinicians evaluated the risk of VAP with clinical picture with the mCPIS, 23 suspicions among 86 (27%) were classified as at intermediate risk (40–60%). Fifteen of these 23 patients (65%) were treated immediately after the BAL procedure because of their hemodynamic status and/or respiratory status and/or an impaired immune status.

Discussion

The main result of this study is that addition of microbiological data (direct BAL examination or culture results of a recent endotracheal aspirate) to the clinical suspicion rather than serum or alveolar PCT values is more helpful in the increasing accuracy of the early diagnosis of VAP.

In 2008, diagnosis of VAP remains a challenge for intensivists [26, 27]. The microbiological techniques require 24–48 h for complete analysis, which can be too long for the sickest patients, but could also result in overprescription of broad-spectrum antibiotics while waiting for final microbiological results. Those considerations have led some investigators to test the accuracy of biomarkers to determine whether patients need antibiotics [16, 17, 28]. Procalcitonin, the prohormone of calcitonin, is a specific biomarker of bacterial infection [29, 30]. Its secretion is ubiquitous in case of sepsis, and its level in blood is increased by community-acquired pneumonia [16–18]. However, few studies have evaluated its usefulness in diagnosing VAP [19, 28, 31–34].

Our study shows that on the day of suspicion of VAP, neither the serum nor BAL level of procalcitonin is a reliable biomarker of VAP. Our results are similar to those reported by Gibot et al. [31], who showed that the blood level of procalcitonin was not different in VAP patients as compared with patients without pneumonia. More recently, two studies reported that the crude value of plasma procalcitonin the day of the VAP suspicion had poor diagnostic value for VAP [33, 34]. However, none of these three studies evaluated the diagnostic value of alveolar levels of procalcitonin [31, 33, 34]. Duflo et al. reported that using mini-BAL, the procalcitonin alveolar level was not different in patients presenting or not a VAP at day 0, 3 and 6 after the onset of the suspicion [31, 33, 34]. We found similar results using BAL and a more recent fully automated homogeneous sandwich immunoassay (TRACE® technology, Kryptor® analyzer).

The secretion of procalcitonin in response to sepsis seems to be initiated by adherent monocytes that act as a trigger during this initial period [35]. After a short crosstalk time between adherent monocytes and tissues, cells of the tissue start to produce procalcitonin [35, 36]. Although adherent monocytes seem to be the cornerstone of the ubiquitous secretion of procalcitonin, the pathway of its excretion to the serum compartment and its apparently poor secretion to tissue compartment are not clearly understood. Moreover, previous studies reported that procalcitonin was rarely elevated in cerebrospinal fluid [37], in ascites [38] or in synovial fluid [39] during localized infection. Regarding our results and those of Duflo et al. [28], alveolar procalcitonin assay cannot be recommended as a guide for diagnosis of VAP (Fig. 1b). In addition, serum assay as a single value the day of the suspicion of VAP does not appear to be a reliable marker of pneumonia (Fig. 1a) [28, 31, 33, 34].

In the present study, the diagnostic accuracy of the mCPIS ≥5 was only 58% with a sensitivity of 75% and a poor specificity (Table 4). Our results are similar to those of other studies [40, 41] demonstrating that mCPIS has a higher sensitivity but a lower specificity than BAL.

Combining a CPIS ≥5 with a serum procalcitonin level of at least 0.2 ng ml−1 did not strongly increase the diagnostic accuracy (57%) or the other operative indices (sensitivity of 65%, specificity of 47%, predictive positive value of 61% and negative predictive value of 51%) (Fig. 2). In contrast with a recently published paper [15], but like others [26, 42], the sensitivity of gram staining of BAL as an independent parameter was low. In our clinical practice, the introduction of new antibiotics is not guided by the BAL gram staining, but is targeted on the previous EA culture result and then adapted when the final BAL culture is available [25, 43]. Combining clinical score such as CPIS and routine surveillance of airway colonization improved VAP diagnosis. Diagnostic accuracy of mCPIS-EA at a threshold of 6 was higher than that of the combination of mCPIS ≥5 with a serum procalcitonin level of at least 0.2 ng ml−1 (P = 0.026), but we observed no difference between mCPIS-EA and mCPIS-direct examination of the BAL. Indeed, combining mCPIS with direct examination of the BAL improved specificity and sensitivity. This latter result is similar to those in other studies [15, 44]. The results also agree with previous studies by our team, where we reported 95% of adequate initial antibiotic therapy for VAP when it was guided by the last systematically, twice-weekly performed quantitative endotracheal aspiration [25] and 85% when it was guided by only once weekly EA [43]. Furthermore, these findings are reinforced by those of Depuydt et al. [45], who showed that systematic surveillance cultures help to predict pathogens responsible for VAP.

Study limitations

Our study presents several limitations. Firstly, 40 to 50 percent of patients were already on antibiotic therapy the day of sampling, which may have decreased the sensitivity of BAL sampling. However, this phenomenon, if it existed, must be marginal in our study because less than 10 percent had newly (less than 72 h) introduced antibiotics [46, 47]. It should be pointed out that there is no gold standard for the routine diagnosis of VAP. Secondly, this was a single center study performed with TRACE technology, which has not been developed to analyze procalcitonin in body fluids, but in blood. Thirdly, we did not evaluate the interest of repeated determinations of procalcitonin or CPIS score, although it has been shown that monitoring the kinetics improves the specificity of those tests in some studies [48] but not in others [33, 40, 41]. However, the aim of our study was to evaluate the combination of clinical scores and biomarkers to anticipate the diagnosis of VAP and not to follow the treatment and/or to anticipate the prognosis in patients presenting a VAP. Finally, we did not evaluate other biomarkers, such as CRP or sTREM-1, as proposed by Gibot and co-workers [31].

Conclusions

The present study showed that microbiological resources (the last available endotracheal aspirate or direct examination performed on BAL at the moment of suspicion of VAP) combined with CPIS are helpful in distinguishing VAP diagnosed by BAL from patients with a negative BAL.

In contrast, a single value of serum procalcitonin is not an accurate marker of VAP. Alveolar procalcitonin, even using the TRACE technique, does not help clinicians to identify VAP.