Detection of Galactomannan in Bronchoalveolar Lavage Fluid Samples of Patients at Risk for Invasive Pulmonary Aspergillosis: Analytical and Clinical Validity

INTRODUCTION

Angio-invasive pulmonary aspergillosis (IPA) is the most common mold infection in immunocompromised patients and causes significant morbidity and mortality. Solid organ transplant recipients and patients with hematologic malignancies, including allogeneic hematopoietic stem cell transplant recipients, are the best-known risk populations (31). However, recently, other frail patient groups have also been identified as risk groups, including critically ill, intensive care unit (ICU) patients. At particular risk are those with advanced liver cirrhosis, those who suffer from chronic obstructive pulmonary disease (COPD) for which they receive corticosteroids, and patients receiving immunomodulating drugs for the treatment of autoimmune disorders (16).

The crude mortality rate of IPA is considerable, though largely influenced by the status of the underlying disease (29, 32). In order to improve the dismal outcome, adequate antifungal therapy needs to be started as soon as possible (33). Hence, it is important to recognize this condition in a timely fashion. However, securing a firm diagnosis of fungal disease is difficult as patients may not exhibit reliable symptoms and signs in the presence of neutropenia and immunomodulating drugs, such as corticosteroids (31). Many centers have now adopted an antifungal treatment approach based on the presence of suggestive lesions on computed tomography scans (such as nodules, with or without a surrounding halo) although this strategy is hampered by a lack of specificity of these radiological features for IPA, especially in nonneutropenic patients (1, 9, 25).

Over the last 2 decades the diagnostic capability for detection of IPA has improved through the use of Aspergillus-specific antigens (5). Galactomannan (GM), a major component of the Aspergillus cell wall, is released during invasive disease, and the level of circulating serum or plasma GM is indicative of the fungal burden in the host. GM can be detected by the commercially available Platelia Aspergillus enzyme immunoassay (GM EIA; Bio-Rad Laboratories, Marnes-la-coquette, France), and results are reported as the optical density (OD) galactomannan index (GMI) (10). Serum or plasma GMI measurements have been studied extensively, gaining widespread acceptance as a sensitive method for prospective surveillance in hematology patients but not in other at-risk groups (17).

More recently, detection of GM in bronchoalveolar lavage (BAL) fluid has been advocated as a sensitive test for diagnosing IPA, both in hematology and nonhematology patients, including solid-organ transplant recipients and critically ill patients (2–4, 7, 8, 13–15, 18, 19, 21–24, 26–28, 30). However, most of the published studies have been restricted to a specific risk group. In addition, the reported literature includes only small numbers of proven and probable cases of IPA. The primary aim of this study was to evaluate the performance characteristics and clinical usefulness of GM detection in BAL fluid for diagnosing IPA in a wide range of underlying at-risk conditions. In addition, we examined the performance of this test using different cutoff values for test positivity.

MATERIALS AND METHODS

We conducted a retrospective cohort study at the University Hospitals Leuven (Leuven, Belgium), a tertiary-care university center. All adult patients (≥18 years of age) who underwent a bronchoscopy with lavage between 1 January 2009 and 31 December 2009 and who had BAL fluid tested for GM were reviewed. The study protocol was approved by the Institutional Review Board. All patients gave informed consent before the procedure.

Bronchoscopy was performed using a flexible fiberoptic bronchoscope following local anesthesia with 2% lidocaine (Xylocaine). All procedures were performed with sterile isotonic saline (20 ml twice); no Plasmalyte solution was used (11). The sampling area was determined on the basis of the lesion location on chest imaging (X-ray or computed tomography [CT] scan). Following aspiration, aliquots were immediately sent to the microbiology and pathology laboratories. BAL fluid samples were submitted for cytology assessment, Gram staining, auramine staining, and calcofluor white fungal staining and for aerobic bacterial, fungal, viral, and mycobacterial cultures. In addition, PCR testing for respiratory viruses, Pneumocystis jirovecii, Mycoplasma, and Legionella was performed. During this study period, integrated-care pathways for fungal diagnosis and treatment were operational for hematology patients, solid-organ transplant recipients, and patients admitted to the medical and surgical ICU. These care pathways clearly identified triggers for requesting additional radiological evaluation (including CT scan) and, if indicated, for requesting a bronchoscopy with lavage. As per protocol, detection of galactomannan in lavage fluid was a standard test. For the remaining patients, the test was usually requested by the chest physician who performed the bronchoscopy.

The Platelia Aspergillus GM EIA (Bio-Rad Laboratories, Marnes-la-Coquette, France) was used to detect the presence of GM on uncentrifuged BAL fluid specimens. The test was performed according to the manufacturer's recommendations for testing serum samples. The assay was run three times weekly in the routine microbiology laboratory. Pretesting samples were stored at 2 to 8°C. All tests were performed by technicians who were unaware of the clinical condition of the patient.

The following baseline findings were retrieved from the hospital chart: age, gender, underlying disease, department of hospitalization, antifungal use (prophylaxis and treatment), use of corticosteroids, administration of piperacillin-tazobactam or amoxicillin-clavulanate, and neutrophil count at the day of testing (± 1 day). Results of bacterial, fungal, and mycobacterial cultures and of PCR tests performed on BAL fluid samples were retrieved from the laboratory information system; results of direct microscopic examination of BAL fluid and respiratory tissue biopsy samples were collected from the pathology department. All chest CT scans (or chest X-rays when CT scan was not available [mainly for ICU patients]) were reviewed.

The BAL fluid GM EIA results were benchmarked against the revised European Organization for Research and Treatment of Cancer/Mycoses Study Group (EORTC/MSG) consensus definitions for invasive aspergillosis (IA) (6). In the absence of histopathology findings, these definitions are considered to be the most appropriate reference standard for assessing the diagnostic performance of new diagnostic tests in medical mycology. The EORTC/MSG categorization classifies the target population as having proven IPA, probable IPA, possible invasive fungal disease (IFD), and no IA. All assessments were done by two independent investigators who were also blinded to the BAL fluid sample GM EIA results.

Proven IPA was defined by positive histopathology from lung tissue. These samples could have been obtained premortem or postmortem. For all patients fulfilling the revised EORTC/MSG host criteria (neutropenia, allogeneic stem cell transplant, prolonged use of corticosteroids, use of immunosuppressive drugs, or inherited severe immunodeficiency), probable IPA was defined by the presence of features of a lower respiratory tract fungal disease on CT scan (only dense, well-circumscribed lesions with or without a halo sign, an air crescent sign, or a cavitated lesion were accepted as radiological abnormalities) and at least one mycological criterion (e.g., positive culture or microscopy for mold from a respiratory sample or at least two positive [OD of ≥0.5] serum GM detections). To avoid incorporation bias, GM detection in BAL fluid was not included as a mycological criterion. As previously stated, critically ill patients with chronic obstructive pulmonary disease, autoimmune disorders, or liver cirrhosis were added as host factors (21). However, because a CT scan is less feasible and because the radiological consensus criteria are less specific for IPA in these groups of patients, the presence of pulmonary infiltrates or consolidations not responding to broad-spectrum antibacterial coverage was considered a sufficient radiological criterion (21). Cases with a host factor and compatible radiological features but without a mycological criterion were considered possible IFD. All other patients were considered IPA negative.

In an attempt to minimize spectrum bias (i.e., excluding important patient subgroups), we have included BAL fluid specimens from all patients for whom the test is intended to be used (the target population); this includes all patients fulfilling the host criteria as defined per revised EORTC/MSG consensus definitions and presenting with unexplained nodular lesions or pulmonary consolidations/infiltrates on chest X-ray or (preferably) chest CT scan. In addition, specific subgroups of ICU patients, namely, those with advanced liver cirrhosis, chronic obstructive pulmonary disease on corticosteroids, and autoimmune disorders receiving immunomodulating therapy with the presence of pulmonary infiltrates or consolidations not responding to broad-spectrum antibacterial coverage were also included. Hence, BAL fluid samples of patients not fulfilling the above-mentioned host criteria, patients without radiological abnormalities (e.g., BAL fluid examined as part of a protocol-defined follow-up procedure), and patients without recent radiological examinations were excluded from the analysis. To avoid disease progression bias for patients with multiple BAL fluid samples available, only the first sample was included in the analysis. Finally, only BAL fluid samples taken before the start of mold-active antifungal therapy were included in the analysis.

For performance analysis, proven and probable IPA cases were grouped together as IPA. Diagnostic performance was expressed as sensitivity, specificity, positive and negative predictive value (PPV and NPV, respectively), positive and negative likelihood ratio (+LH and −LH, respectively), diagnostic odds ratio, and error odds ratio relative to different optical density (OD) index cutoff values by using two-way contingency tables. A 95% confidence interval (CI) was calculated for each value (using the calculator at http://statpages.org/ctab2x2.html). The area under the receiver operating characteristics (ROC) curve was constructed to assess how changes in the OD index cutoff for the GM EIA assay altered sensitivity and the value of 1 minus specificity. Fagan's nomograms were constructed for calculating posttest probabilities (using the calculator at http://araw.mede.uic.edu/cgi-bin/testcalc.pl).

RESULTS

DISCUSSION

Pulmonary involvement is a hallmark of invasive aspergillosis. Bronchoscopy with examination of BAL fluid is therefore widely used for the evaluation of patients with suspected invasive aspergillosis, relying on diagnosis by culture and visualization of fungal elements microscopically using specific stains (20). However, these methods are often falsely negative and time-consuming in the case of isolation and identification by culture (12). Hence, a culture-independent method may increase the yield of BAL fluid samples. Recently, the clinical utility of GM EIA detection using BAL fluid samples has been explored in a variety of adult and pediatric populations; these studies, which use various OD cutoff indices for positivity, have reported high sensitivities in hematology patients (2, 3, 7, 13, 19, 22, 24, 26, 28, 30), most solid-organ transplant recipients (4), critically ill patients (21), and nonimmunocompromised patients (23). However, significantly lower sensitivities have been reported in lung transplant recipients (4, 15, 18, 27). A recent meta-analysis determined the sensitivity to be 90% and the specificity to be 94%. Moreover, a positive result increased the probability of having IPA about 6-fold, whereas a negative result decreased the probability to 1% (10).

This large-cohort study assessed the clinical utility of the BAL fluid GM EIA in a mixed population of at-risk patients who underwent bronchoscopy with BAL to evaluate the etiological nature of unexplained pulmonary abnormalities. The study represents real-world observations. The sensitivity of the BAL fluid GM EIA using an OD cutoff value of ≥0.5 was 93.2%; as such, a negative BAL fluid GM assay value virtually excludes the presence of IPA, irrespective of the pretest probability of disease. This high sensitivity is, however, offset by a reduction in test specificity. Using an OD cutoff value of 0.5, at least in patients not receiving mold-active drugs, results in a significant rate of false-positive results. As such, an OD index of ≥0.5 does not confirm the diagnosis of IPA and requires further investigation. However, a negative result makes the diagnosis very unlikely. This is furthermore underscored by the low negative likelihood ratio (below 0.1) at this cutoff, providing convincing evidence to rule out diagnosis in most circumstances. On the other hand, an OD index value of ≥3.0 corresponds with an excellent specificity, securing the diagnosis of IPA, but at the expense of a reduced sensitivity. Again, the high positive likelihood ratio (above 10) at this cutoff provides convincing evidence to rule in diagnosis in most circumstances. We therefore propose to use this OD index cutoff to include patients in clinical studies evaluating the efficacy of antifungal drugs in proven and probable cases of IPA. However, patients with possible IFD may also be candidates for antifungal therapy. Hence, for daily clinical practice, a lower OD index cutoff value (0.8) seems to be much more appropriate, as evidenced by the ROC analysis. This lower cutoff is also in line with previously published experiences (24). Of note, at these lower values, the pretest probability of disease remains crucially important for determining the posttest probability (Fig. 2).

Our study has several limitations. First, physician-driven decisions for performing bronchoscopy and the timing of bronchoscopy are potential sources of bias. This is certainly the case for patients not treated according to integrated care pathways (1). As a consequence, samples may have been collected at various time points during the course of the infection and at various stages of fungal disease. Second, BAL protocols, contrary to serum sampling, are not standardized; the variable amount of instilled and collected fluid may affect GM concentrations and, consequently, the sensitivity of the assay, as evidenced in a recent study (28). Normalizing BAL fluid GM values to another lung fluid analyte (such as urea) could be a way of circumventing this problem in future studies. Third, by using the EORTC/MSG consensus definitions, we used a very conservative (but generally accepted) reference standard to estimate the performance characteristics of this assay. Unfortunately, the 2008 revised definitions fail to capture all cases of IPA, both in patients with the appropriate host factors (e.g., patients with airway-invasive aspergillosis presenting with radiological features not fulfilling the EORTC/MSG criteria) and in patients without appropriate host factors and clinical features (in particular, ICU patients and many solid-organ transplant recipients) (9, 21, 25); therefore, the specificity of BAL fluid GM detection at various cutoff values may be underestimated. Fourth, this study cannot assess the impact of antifungal prophylaxis or therapy on the test performance since too few patients received mold-active prophylaxis and since BAL fluid samples were collected before the start of antifungal therapy. However, according to recent data, treatment with antifungal therapy for two or more days promptly decreases the assay sensitivity in BAL fluid samples as well as in serum samples (28). Fifth, because of the retrospective nature of the study, we cannot provide data on the diagnostic performance of the GM assay in BAL fluid samples in comparison with paired serum assays.

The strength of this study could be the high number of documented cases, the minimization of spectrum bias, and the external validity of the findings. Although the number of patients with proven IPA is small, this study included a substantial number of patients with well-documented probable IPA. Estimates of diagnostic accuracy are usually subject to spectrum bias because most studies do not include the complete spectrum of subgroups at risk, frequently omitting the intermediate and typically more difficult cases to diagnose (e.g., critically ill patients in intensive care units and COPD patients). Eliminating these cases produces an overly optimistic picture of how the assay performs in clinical practice. So, we aimed at estimating the performance of a BAL fluid GM EIA in a population sufficiently representative of the intended-use population, including traditional as well as emerging risk groups for IPA. As such, we could also show that the performance of the assay in BAL fluid samples—in contrast to results with blood samples—was not influenced by the neutropenic status of the host. Finally, we believe that these results are reflective of the real-world performance of the BAL fluid GM EIA. Indeed, over this 1-year period, several batches of the assay have been used by multiple trained and experienced technicians in a nonresearch, clinical microbiology laboratory.

In conclusion, detection of GM in BAL fluid samples of patients at risk of invasive pulmonary aspergillosis (and before receiving mold-active drugs) can be used as a diagnostic method with excellent accuracy, provided results are interpreted in parallel with clinico-radiological findings and pretest probabilities. Future research should primarily focus on the impact of preanalytic variables (e.g., volume of BAL fluid) and standardization of the bronchoscopic procedure.