INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Clinical and bacteriologic diagnosis of nosocomial pneumonia (NP) in mechanically ventilated (MV) patients is still difficult (1, 2). Different clinical criteria have been tested and have been proven to be inaccurate in predicting pneumonia (3, 4). Simple diagnostic procedures (such as endotracheal aspiration) give good qualitative information on tracheal ecology, but are unusable because of the frequency of the colonization of the proximal airways (5). Therefore, the past years have seen the development of new diagnostic methods such as protected specimen brush (PSB) and bronchoalveolar lavage (BAL), in order to approach the reality of the pulmonary infection. Provided the use of threshold (10³ cfu/ml for the PSB and 10⁴ cfu/ml for the BAL), these new procedures are now recommended as reference methods in the diagnosis of NP in MV patients (6, 7, 11). However, autopsic studies report the persistence of false positive and negative results (15). According to scientific evaluation criteria, rigorous assessment of diagnostic methods is needed for efficient use (20). In this hierarchic procedure, the first of the five steps is the technologic capability (i.e., the ability of the test to perform to specification in a laboratory setting). BAL and PSB have both been tested under laboratory settings with good results. Moreover, the clinical evaluation has raised the problem of the repeatability of these procedures. A study on the reliability of the PSB showed that cautious intepretation of the results is advisable in clinical practice, especially if a decision (to treat or not to treat) is to be made (21). These results show how deep the gap can be between laboratory studies and clinical application. The clinical repeatability of the BAL has not been tested yet (20). Therefore, we decided to assess the repeatability of BAL in the diagnosis of NP in MV patients.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Patients

Forty-four patients (32 men, 12 women) undergoing mechanical ventilation (Servo 900 C, Siemens, Sweden) for more than two days and meeting criteria for suspicious pneumonia (i.e., new or progressive or persistent pulmonary infiltrate, purulent tracheal secretions, fever > 38° C) were studied. Mean age was 59 ± 18 (SEM) years (range, 19 to 89). Mean SAPS was 11 ± 4 (SEM) (range 4 to 24). The average time between starting of mechanical ventilation and the study was 12 ± 11 (SEM) days (range, 3 to 57). The primary indication for ventilatory support included (Table 1): acute respiratory failure (n = 20), exacerbation of chronic obstructive pulmonary disease (n = 11), acute cardiac failure (n = 6), postoperative respiratory failure (n = 4) and miscellaneous (n = 3). For all the patients, the diameter of the tracheal prosthesis was superior to 8. Antibiotics were administrated to all patients except for five of them, and for more than 3 days at the time of the study. All patients had a negative HIV status. This prospective study was approved by the Institutional Review Board (Comité Consultatif de Protection des Personnes dans le Recherche Biomedicale). Informed consent was obtained from the patients or the nearest relatives.

Bronchoalveolar Lavage

Patients received intravenous premedication with midazolam and vecuronium. Thirty minutes before the beginning of the first BAL (BAL1), the FI_O2 was adjusted to 100% and was kept to this level until thirty minutes after the end of the second BAL (BAL2). During the same period, the positive end-expiratory pressure (PEEP) applied by the ventilator was equal to zero. No topical anesthesics were used. The trachea was suctioned before introducing the bronchoscope. The fiberoptic bronchoscope (FB 18X; Pentax, Tokyo, Japan) was introduced through a special adaptator (double rotule; Europe Médicale, Bourg, France) and advanced under direct vision next to the lobe setting the radiographic signs, without suctioning to avoid contamination. The bronchoscope was then positioned in a segmental bronchus of the lobe. Three aliquots of 40 ml sterile normal saline were separately infused through the working channel of the bronchoscope. Hand suction was applied and the first aliquot withdrawn as bronchial lavage. The two other aliquots were pooled for bacteriologic and cytologic analysis. The samples were transported to the laboratory within 30 min of collection. The second bronchoscopy was performed under the same conditions 30 min after the end of the first one. The 30-min delay was necessary to sterilize the bronchoscope. Location of the second bronchoscopy was in the same lobe, adjacent segment. All the sampling procedures were performed by the same bronchoscopist in order to avoid operator-dependent variability.

Cytology

The differential count of macrophages, neutrophils, eosinophils, mastocytes, and contaminant elements (bronchial and upper aerodigestive cells) was made on centrifugation preparations of the BAL sample. Two spots were prepared, in a Shandon spin, at 2,000 rpm for 2 min, one was stained with May-Grunwald-Giemsa, the other with Papanicolaou techniques; 200 cells were numbered on each spot. BAL samples were considered invalid when more than 1% of the cells counted were ciliated bronchial cells. For numeration counts, results are expressed as cells/ml of BAL.

Bacteriology

One aliquot of each BAL fluid was plated on MacConkey agar, chocolate agar, and 5% sheep blood Columbia colistin-nalidixic acid agar (BioMérieux, Marcy l'Etoile, France). After 48-h incubation, colonies were counted and identified.

Statistical Analysis

The study tests the degree of agreement between two BAL (BAL1 and BAL2) performed on the same subject. Absence of systematic difference (i.e., bias) between the two BAL for the same patient was looked for using the MacNemar test (p > 0.05 meant no bias). The results were then compared using the kappa test for agreement and repeatability; p < 0.05 meant the test was repeatable (22).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The 88 BAL performed had less than 1% ciliated cells, assessing their alveolar origin. Using the Wilcoxon test, no statistical difference was found between BAL1 and BAL2 for the total numerations (1,826,000 ± 2,416,000 versus 1,036,000 ± 1,737,000, respectively; z = 1.55, p > 0.1) or neutrophilic PMN counts (1,491,934 ± 2,123,469 versus 854,632 ± 1,677,096, respectively; z = 1.49, p > 0.1). Considering the qualitative repeatability of the cultures (i.e., the presence or absence of bacterial colonies), 42 patients (95.4%) had the same results on both BAL (Table 2). Twenty-eight (65.1%) were both negatives and 14 (31.8%) were both positives and showed the same bacteria. One patient had BAL1 positive and BAL2 negative, and one had BAL1 negative and BAL2 positive. The agreement level was 90% with no bias (p > 0.3 for MacNemar test, p < 10⁸ for kappa test). Applying the 10⁴ cfu/ml threshold to the 44 pairs of BAL, 40 (90.9%) gave the same results, and the kappa test showed good agreement (72%, p < 10⁵) (Table 3).

The same statistical analysis performed upon the 16 BAL being at least once positive shows less significant results: 11 (78.6%) did not have the same log₁₀ cfu/ml. The quantitative results were spread out on both sides of the 10⁴ cfu/ml threshold in 4 (25%) of the 16 patients (Table 4). For these four patients, numeration counts were controlled using the Wilcoxon test to assess absence of bias (i.e., one BAL was in pneumonia and not the other). Results showed no statistical difference between positive and negative BAL for total numeration and neutrophilic PMN counts (Table 5). Upon the 16 patients, the agreement level was 45.5% with no bias (p > 0.3 for MacNemar test). The kappa test showed limited significance (test = 1.97; p = 0.05 for test = 1.96). The intrasubject variability has been studied for agreement (Figure 1). The same log₁₀ cfu/ml was found in five patients (agreement: 31.3%, p > 0.15 for kappa test). The difference in log₁₀ cfu/ml was one (or less) in 11 patients (agreement: 75%, p < 10⁹), two (or less) in 13 (agreement: 87.5%, p < 10⁹) and three (or less) in 16 (agreement: 100%, p < 10⁹). The order of passage of the BAL did not influence the quantitative results (BAL1 > BAL2: 7, BAL1 = BAL2: 5, BAL1 < BAL2: 4, MacNemar test: p > 0.85). No complication or respiratory distress was seen.

View larger version (12K): [in this window] [in a new window]   Figure 1.   Variability of log between BAL (16 patients). Differences in the log between the two BAL of the same patient. Definitions: pts = number of patients; Diff = 0 means log BAL1 = log BAL2; * kappa test nonsignificant (i.e., BAL nonreproducible).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The BAL is one of the two methods recommended to assess the diagnosis of nosocomial pneumonia in ventilated patients (2, 18, 20). Its accuracy is dependent on the patients' characteristics, the operator and sampling methods, and the interpretation of the results (25). Because of the bronchial contamination, a recommended diagnostic threshold has been established (10⁴ cfu/ml) for the bacteriologic results (6, 7, 11- 14). One could debate this 10⁴ cfu/ml threshold: 10³ cfu/ml is more sensitive (but less specific), while 10⁵ cfu/ml is more specific (but less sensitive). The 10⁴ cfu/ml threshold appears to be the best compromise between sensitivity and specificity. In any case, the use of these measurements in daily practice (for diagnostic purposes and therapeutic decisions) assumes that their repeatability is acceptable. Bacteriologic results of the BAL (using the 10⁴ cfu/ml threshold) are important for the clinician, for therapic decisions (to treat or not to treat). And that decision is an important one. Treating someone who does not need it can lead to nosocomial infection with antibioticresistant bacteria, or to important side effects. Not treating someone who needs it can lead to death. So this result has to be reliable in daily practice. Many variations among the positive results have been noted. The explanations for these variations are not clear. All the procedures were performed by the same operator, in order to avoid operator-dependent factors. All the samples were studied in the same laboratory as usual, the only difference being that nobody but the fibroscopist knew which BAL was the first one or the second one. Intrasubject variability can be ruled out because the patients were their own control. Repercussion of the first BAL upon the quality of the second can be discarded because of the absence of bias. Variations of dilution of the bronchial secretions could account for the variability of the qualitative or quantitative culture results. However, total cellular counts and neutrophilic PMN counts were not statistically different between BAL1 and 2. Moreover, BAL were performed as they are in daily practice, and these results seem to us applicable to clinical practice. As in daily practice, most of the patients were receiving antibiotic therapy (88%). It is important to notice that those antibiotic therapies were the same for at least 3 days, so that it was licit to suspect a resistant-bacteria pneumonia, and to perform a BAL. The aim of this study is to test the repeatability of the BAL by repeating the sampling procedure twice in the same lobe (contiguous segments) in MV patients with suspected NP. The results were compared using the MacNemar test, which looks for a systematic difference between two results. In this study, the MacNemar test appreciates the presence of a bias. The MacNemar test is not significant in our results. This fact allows us to say that there is no bias reliable to the technicality in this study. As there is no bias, the kappa test may be used (22). It looks for repeatability. It calculates the agreement level (in percentage), but as it is dependent on the prevalence of the factor studied, it is not a good statistical result. Using the test itself, it gives a statistically useful result. In this study, the repeatability is statistically demonstrated when p < 0.05 for kappa test. As shown in this study, absence of pneumonia (based on presence of bacteria in the BAL with less than 10⁴ cfu/ml) is a highly reproducible result (p < 10⁵). But presence of pneumonia is a poorly reliable result (0.04 < p < 0.05). If one considers that, given the importance of the result of BAL, on therapic decision, the threshold for significance has to be 1% (p < 0.01), it appears that BAL is not reproducible when positive. The intrasubject variability shows that only 31.25% of the positive BAL have the same log₁₀ cfu/ml. A variability of one log₁₀ cfu/ml has to be allowed for the BAL to be reliable. That low degree of repeatability could account for the different results that different authors express in the literature. However, this result is limited because of the small number of positive patients. BAL seems to be a reproducible diagnostic method, especially when bacterial cultures are negative. The limited significant threshold when cultures are positive leads us to recommend cautious interpretation in daily practice. Further studies are needed before definite conclusions can be drawn.