ABSTRACT
Combined biomarker screening is increasingly used to diagnose invasive aspergillosis (IA) in high-risk patients. In adults, the combination of galactomannan (GM) and fungal DNA detection has proven to be beneficial in the diagnosis of IA. Data in purely pediatric cohorts are scarce. Here, we monitored 39 children shortly before and after allogeneic stem cell transplantation twice weekly by use of a commercial GM enzyme-linked immunosorbent assay (ELISA) and a PCR assay based on amplification of the pan-Aspergillus ITS1/5.8S ribosomal operon. In addition, clinical data were recorded and classification of IA was performed according to the European Organization for the Research and Treatment of Cancer/Mycoses Study Group (EORTC/MSG) criteria. Among the 39 high-risk children, we identified 4 patients (10.3%) with probable and 2 (5.1%) with possible IA. All patients with probable IA were repeatedly positive for both tests (means of 9.5 and 6.8 positive GM and PCR samples, respectively), whereas both possible IA cases were detected by PCR. The sensitivity and specificity were, respectively, 67% and 89% for GM and 100% and 63% for PCR. Positive and negative predictive values were, respectively, 50% and 100% for GM and 27% and 100% for PCR. For the combined testing approach, both values were 100%. The number of positive samples seemed to be lower in patients undergoing antifungal therapy. Sporadically positive tests occurred in 12% (GM) and 42% (PCR) of unclassified patients. In summary, our data show that combined monitoring for GM and fungal DNA also results in a high diagnostic accuracy in pediatric patients. Future studies have to determine whether combined testing is suitable for early detection of subclinical disease and how antifungal prophylaxis impacts assay performance.
INTRODUCTION
Invasive aspergillosis (IA) is the most significant opportunistic fungal infection in neutropenic adult and pediatric patients following allogeneic hematopoietic stem cell transplantation (alloHSCT) (1). Diagnosis of IA is challenging, as clinical symptoms are often nonspecific and classical diagnosis is poor (2). Methods such as high-resolution computed tomography (CT) scans show only typical signs once the infection is established, and even then specific signs can be transient (2, 3). Serological tests that detect galactomannan (GM) and β-d-glucan have low positive predictive values (PPVs), better used for the exclusion rather than the diagnosis of IA (2, 3). Such limitations have led to the development of PCR-based assays to detect fungal DNA in patient specimens. However, such assays operate at the very limit of detection due to the small amounts of fungal DNA recovered from blood samples and in pediatrics is further compounded by the relatively small blood volumes typically available. Recent studies in adults indicate that combining the PCR-based assays with antigen testing for the diagnosis of IA may be beneficial (4–6). However, few data exist for these combined strategies in high-risk pediatric patients. This article describes a highly standardized diagnostic schedule involving twice weekly systematic screening of high-risk children by GM and PCR assays around alloHSCT. In parallel, a large variety of clinical signs and symptoms and microbiological data were collected. These data were retrospectively compared to those of a recently described adult cohort (7), and the findings are discussed below.
RESULTS
The entire set of samples comprised 543 blood specimens collected from 39 high-risk pediatric patients shortly before or after alloHSCT. Among all patients, 4 cases of probable IA (10%), 2 cases of possible IA (5%), and 33 unclassified patients with no specific radiological signs of IA (85%) were classified according to the current European Organization for the Research and Treatment of Cancer/Mycoses Study Group (EORTC/MSG) criteria (8). No proven IA was documented.
Biomarker assay performance.As outlined in Table 1, both GM and PCR were positive in 4/4 patients with probable IA, generating a sensitivity of 100%. In these four patients with probable IA, 38/74 (51% [95% confidence interval {CI}, 40 to 62%]) and 27/77 (35% [95% CI, 25 to 46%]) samples proved to be GM and PCR positive, respectively. In 2 patients with possible IA, PCR was positive in 2/22 samples, whereas no positive GM ELISA occurred in these cases. In the unclassified patients, 4/33 and 13/33 patients showed either a positive GM or a positive PCR result, respectively, representing 9/437 (2.1% [95% CI, 1.1 to 3.9%]) and 17/441 (3.9% [95% CI, 2.4 to 6.1%]) samples positive by GM and PCR, respectively. No unclassified patients were positive by both PCR and GM, generating a combined specificity of 100%. The PPVs were 50% for GM ELISA and 27% for PCR; the negative predictive values (NPVs) were 100% for GM and 100% for PCR testing. When the GM ELISA and PCR were considered a combined test system, the PPV and NPV values were both 100% for the detection of probable IA. One patient with a pulmonary aspergilloma never had any positive GM or PCR test result.
Number of positive GM or PCR assays with respective EORTC/MSG classification of patientsa
If cases of both probable and possible IA were considered positive for the calculation of the clinical performance of the assay, the sensitivity and NPV for the GM test were diminished to 67% and 94%, respectively. While sensitivity and NPV for the combined test system were reduced to 67% and 94% as well, the PPV for the PCR assay increased to 35%.
Investigating the effect of antifungal therapy on biomarker assay performance.The effect of using Aspergillus-active prophylaxis or empirical therapy on biomarker assay positivity testing in unclassified patients was investigated. As shown in Table 1, during the use of prophylaxis or empirical therapy, there was a trend (P = 0.0519) toward a reduced rate of positive test results (14% [95% CI, 6 to 32%] compared to 60% [95% CI, 23 to 88%] positive samples in the untreated unclassified population). Interestingly, 8 of 28 (29% [95% CI, 15 to 41] of these treated patients had positive test results outside their treatment or prophylaxis period, resulting in a doubling in the percentage of positive test results compared to that during the prophylaxis/treatment period (29% versus 14%, respectively). This lower incidence of positive biomarkers during antifungal prophylaxis or empirical therapy might indicate a positive clinical effect of antifungal therapy, reducing the release of mold into the blood circulation. The times from initiation of screening to biomarker positivity in patients with probable disease were comparable, with 49 ± 39 days versus 46 ± 40 days (mean ± standard deviation [SD]) for GM versus PCR testing, respectively.
Associations between CT, PCR, and GM positivity.Sixty-seven CT scans were available from 33 patients; in 6/33 patients, CT scans displayed lesions compatible with IA (Fig. 1). PCR was positive in 6/6 patients (sensitivity, 100%), compared to 10 PCR-positive patients in 27 CT-negative patients (specificity, 63%). The Pearson chi-square test of independence applied to the association between a positive CT scan and a positive PCR result suggested a positive correlation between the two tests (chi-square [1 degree of freedom] = 7.7917, P = 0.005).
Venn diagram illustrating the relationship between the CT-based detection of aspergillosis (CT), the change from a negative to a positive PCR result (asp) in a given patient, and a positive GM assay result (gm). Out of 16 positive Aspergillus PCR events, 6 were truly positive and 10 were false positive (PPV, 37.5%). All positive CT scans were detected by Aspergillus PCR (sensitivity, 100%). Out of 7 positive GM events, 4 were truly positive and 3 showed a false-positive GM assay result (PPV, 57.1%). The GM assay failed to detect 2 out of 6 positive CT scans (sensitivity, 66.7%).
The GM ELISA was positive in 4/6 CT-positive patients (sensitivity, 67%) and in 3/27 CT-negative patients (specificity = 89%). The same chi-square test of independence between a positive CT scan and a positive GM ELISA was significant (chi-square [1 degree of freedom] = 9.0659, P = 0.003). Using the GM index as the classifying variable and the Aspergillus PCR assay as the reference variable, receiver operating characteristic (ROC) analysis revealed that the area under curve (AUC) was 0.7335 (95% CI, 0.64233 to 0.82475), with a standard error of 0.0465. The maximal Youden index, J, was 49.9%, corresponding to a GM index of 0.4 (Fig. 2), demonstrating that a GM index of 0.4 was the best cutoff point to predict when the Aspergillus PCR assay will be positive.
ROC curve analysis demonstrating the capability of the GM ELISA index predicting a positive Aspergillus PCR assay. The Youden index, J (red line), suggests a GM cut-point of 0.4 as the best for a discriminative purpose. This corresponds to a sensitivity of 55% and a specificity of 95%.
In patients with probable IA, a mean of 9.5 GM samples (range, 6 to 14 samples; mean optical density [OD], 3.1) and 6.8 PCR samples (range, 3 to 14 samples) showed positive results.
Next, binary and continuous covariates were analyzed using a time-to-event approach, running the nonparametric log rank method. Time-event variables included the day of binary Aspergillus PCR status change, as first conversion from negativity to positivity, and the day of first GM conversion to a positive index. We revealed that liposomal amphotericin B use reduced the risk of a positive Aspergillus biomarker test result, whereas CT positivity was associated with a significant increase of risk (P = 0.042). Using the GM conversion to positivity as the endpoint, CT positivity was associated with a markedly increased risk for IA (P = 0.0375). Four covariates (use of corticosteroids, other herpesviruses, caspofungin, CT positivity) were significantly associated with an increased risk for IA.
Longitudinal investigations.For all 4 patients with probable IA, a longitudinally detailed analysis was performed. Figure 3 shows as examples the characteristic time courses for 2/4 patients (see Fig. S1 in the supplemental material for data for the 2 other patients).
Representative medical histories of two patients suffering from probable IA with different courses and outcomes. Graphs display the course of C-reactive protein (CrP, red) and neutrophils counts (blue) over time, together with the results of the GM/PCR assays, presence of clinical symptoms, CT scans (bars above the graph), and antifungal treatment (bars below the graph). For CT scans and GM/PCR assays, red indicates a positive test result and blue indicates a negative test result. (A) A 12-year-old boy who received an alloHSCT due to a refractory AML. He developed de novo IA during aplasia but recovered. Finally, he died from further disease progression but without clinical signs of IA. (B) A 15-year-old boy with refractory ALL who was referred to our center for ALL immunotherapy. Despite intensified antifungal treatment, his preexisting IA further progressed, from which he ultimately died. During treatment, all GM/PCR tests remained continuously positive with increasing values (GM index in the final three samples is above the upper limit of 8).
Patient 1 (Fig. 3A), a 12-year-old boy with refractory acute myeloid leukemia (AML) who had received a matched unrelated donor transplant, was given prophylactic caspofungin due to prolonged neutropenia. During aplasia after alloHSCT, he developed fever, cough, tachycardia, and elevated C-reactive protein (CRP) levels and oxygen demand. CT scans revealed a new 1.4-cm nodule in the left upper lobe, compatible with pulmonary IA. After neutrophil regeneration and intensified antifungal treatment, he cleared his IA within 24 days, as confirmed by repeat CT. Interestingly, GM samples (1 test) and PCR samples (3 tests) became positive again, later during the reconvalescence, without clinical symptoms but in association with neutropenia.
Patient 2 (Fig. 3B), a 15-year-old boy who suffered from refractory acute lymphocytic leukemia (ALL), was referred to our center to receive bispecific antibody treatment and subsequent alloHSCT. The referring institution had already suspected a pulmonary IA, which was confirmed by double-positive GM and PCR tests as well as radiologically by the presence of diffuse nodules up to 2.3 cm in diameter in both lungs. Although antifungal treatment was further intensified, he progressed clinically and showed 14 double-positive samples in one series with increasing GM indices and DNA load in the real-time PCR assay. This patient ultimately died from IA, prior to alloHSCT.
DISCUSSION
The aim of this study was to determine whether intense (twice weekly) and standardized GM ELISA and PCR screening could increase the diagnostic accuracy for the detection of IA in high-risk pediatric patients. For children, the number of existing studies on this matter is scarce, patient populations are often heterogeneous, and currently no methodological standards similar to the recommendations of the European Aspergillus PCR Initiative recommendations (EAPCRI) exist. As discussed recently by Lehrnbecher et al., there are only 19 studies reporting on GM ELISA testing and 11 studies reporting on PCR testing in pediatric cancer or HSCT patients (9). In addition, disease pathology and the physiology of children differ from those of adults (e.g., children-specific diseases, immunology, metabolism, body weight, and blood volume). In contrast to findings in adults, the typical signs of IA (halo and air crescent signs) are rarely seen in young children and radiographic findings are often unspecific (10). In addition, the nutrition of babies and children as well as the maturation state of the mucosa might be responsible for significantly higher rates of false-positive GM ELISAs (11). Similarly, the antibiotics piperacillin-taxobactam and amoxicillin-clavulanate have been reported to yield false-positive results in GM assays, although recently this finding has been questioned (12) (13). Recently, Buchheidt et al. presented a review focusing on studies of PCR-based assays to diagnose IA in immunocompromised pediatric patients (14). The authors concluded that the value of PCR in pediatric patients has not yet been defined and that a large variety of assay protocols exist. These previous studies relied on nested-PCR assays (15, 16), did not include any cases of probable IA (16, 17), and used extremely small blood volumes (17, 18). In addition, study sizes were very small (17) and sampling frequencies were low (e.g., once per week [16]). Nevertheless, authors found high sensitivities of up to 96% (19) and/or high NPVs of up to 100% (17, 19). Here, we used an established PCR assay which has been validated in an adult cohort upfront. Our data from close longitudinal follow-up combined with a detailed clinical history clearly show that PCR assays are also of diagnostic value in pediatric patients. As expected, on the single-test level, PCR was more sensitive than GM testing, whereas specificity was superior for GM testing. It should be kept in mind that by the EORTC/MSG definitions, all patients with probable IA must be positive for a mycological criterion, which in this study was GM, whereas all patients with possible IA are subsequently negative. This incorporation bias may lead to an overestimation of performance measures for GM and in parallel might negatively influence the performance measures of PCR.
For both PCR and GM assays, a positive CT scan was associated with a higher risk for test conversion. PCR conversion was negatively associated with the use of liposomal amphotericin B, which is in line with the lower rate of PCR positivity when antifungal prophylaxis or therapy is used (Table 1). As reported before (20), both tests were unable to detect noninvasive forms of IA.
The number of unclassified patients with at least one positive PCR specimen in this study of children was higher than expected. However, this could not be attributed to the prophylactic or empirical use of Aspergillus-active agents, since the majority of positive test results occurred in patients without prophylaxis or outside the treatment or prophylaxis period. This observation deviates from our former cohort analysis of 213 adult hematological patients (7). There, we showed that PCR screening is of limited value in patients receiving mold prophylaxis, with 72% (53/73 patients) receiving antifungal prophylaxis generating a potentially false-positive PCR result, compared to only 14% in this pediatric cohort. The reason for these conflicting data are unknown. However, we hypothesize that in pediatric patients with mold prophylaxis, subclinical Aspergillus infections might be less frequent than those in adults and, in consequence, fewer positive PCR tests occur in children receiving mold prophylaxis (4/28 patients) (Table 1). This again argues for more controlled trials on the role of Aspergillus-active prophylaxis in allogeneic transplant patients.
Furthermore, we demonstrated that in adult patients suffering from probable IA, only means of 2.1 and 1.9 samples were positive for GM and PCR, respectively (7), compared to the 9.5 and 6.8 positive samples per affected patient in the pediatric cohort described herein. While the mean OD value of all positive GM tests in pediatric patients with probable IA was 3.1, the corresponding OD value in adults was only 1.5 (7). The reason for this observation is not entirely clear; however, it may well be that in a clinically overt IA, fungal components are more concentrated in pediatric blood samples, since total body blood volumes of children can be up to 10 times smaller than those in adults, while the blood volumes sampled for GM and PCR assays did not markedly differ between the two cohorts.
A closer look at the individual clinical courses of patients provides new insights but also raises several questions. First, positivity of both tests significantly suggests IA, and the level of the GM index or the PCR quantification cycle (Cq) values may eventually serve as a surrogate marker of the clinical course. However, it remains unclear whether single positive tests in cases of possible or unclassifiable IA actually represent false-positive results or whether they truly indicate fungal cell components in the bloodstream without any disease pathology. Furthermore, due to the relatively low number of probable IA cases, we were not able to determine whether the combined testing approach also allows a preemptive detection of IA, before clinical symptoms become apparent. It is worth noting that all high-risk patients in our cohort received antifungal prophylaxis, including the 4 patients developing probable IA, independent of their neutrophil counts. This prophylaxis included posaconazole (10 mg/kg of body weight), caspofungin (50 mg/m2 of body surface area), or liposomal amphotericin B (1 mg/kg of body weight). The impact of antifungal prophylaxis on diagnostic test systems as well as on disease development in the pediatric setting remains elusive and requires further investigations focused on this question.
Despite antifungal prophylaxis, 4 patients developed probable IA, and in parallel, several others frequently had single positive GM or PCR assay results. Although our PCR assay covers a broad range of Aspergillus species, we could not isolate the responsible mold. Thus, we do not have any data on antibiotic resistance in these cases. However, this circumstance also points to the need for more conclusive clinical data and a global consensus on antifungal prophylaxis and coordinated stewardship programs to promote the consistent and appropriate use of antifungal drugs, especially in neonates and children (21).
In summary, we were able to show that in pediatric patients, a combined PCR and GM screening method yielded greater diagnostic accuracy than single testing, providing 100% specificity/PPV, and detected 4 out of 4 patients with probable IA. A single positive test result occurrs frequently without indicating overt disease, while dual testing plays a decisive role in the diagnosis and management of IA and allows the identification of pediatric patients with IA. The role of antifungal prophylaxis and its interaction with diagnostic test systems in pediatric patients remains to be further evaluated in larger multicenter trials.
MATERIALS AND METHODS
Patients.Since 2012, all patients from the University Children's Hospital Würzburg were screened twice weekly by GM ELISA and PCR-based diagnostic assays. In July 2015, all patients to date (n = 39) were selected for retrospective data analysis. All patients (24 males, median age of 9.5 years [range, 4 to 21 years]; 15 females, median age of 10 years [3 to 19 years]) had alloHSCT. Patients were diagnosed with the following underlying diseases: acute lymphoblastic leukemia (n = 22), myelodysplastic syndrome (n = 5), acute myeloid leukemia (n = 3), and other diseases, including chronic myeloid leukemia, Hodgkin lymphoma, Diamond-Blackfan anemia, thalassemia, neuroblastoma, Fanconi anemia, adrenoleukodystrophy, severe aplastic anemia, and Omenn syndrome (n = 1 each).
In this cohort, 85% of the patients received one or two mold-active antifungal drugs as prophylaxis or empirical therapy prior to commencing or during diagnostic testing (among them, 97%, 13%, 13%, and 17% received liposomal amphotericin B, voriconazole, posaconazole, and caspofungin, respectively). The durations of antifungal prophylaxis and empirical therapy were 27 ± 28 days, 35 ± 33 days, and 99 ± 84 days (mean ± SD) in patients with unclassified, possible, and probable disease, respectively.
In total, 543 blood samples (mean, 13.9 samples per patient; range, 6 to 33 samples) were collected. Sampling started in the preparative phase for alloHSCT (median, 14 days [range, 0 to 74 days] before transplantation) and continued for all patients until day +100 after alloHSCT or, if before day +100, until the patient died.
Diagnostic molecular assays.Samples were tested as part of routine diagnostic care involving concomitant GM ELISA (Platelia; Bio-Rad, Munich, Germany) and by PCR testing. We used an index of ≥0.5 to define a positive GM ELISA. A single positive test was required to define a positive GM ELISA. For the PCR, DNA was extracted from a 1-ml cell-free blood fraction using the QIAamp UltraSens virus kit (Qiagen, Hilden, Germany) (22). The protocol used was compliant with the European Aspergillus PCR Initiative recommendations (EAPCRI) for serum to guarantee the highest diagnostic standards (23). The study was designed and performed on test serum samples. If occasionally serum samples were not available, plasma specimens were used instead to continue consecutive sampling for all patients. Real-time PCR, involving an internal control, was performed as previously described (7). Clinical signs and microbiological data were recorded for each individual patient, with IA defined according to revised European Organization for the Research and Treatment of Cancer/Mycoses Study Group ((EORTC/MSG) criteria (8). Data were compared with those for a previously published adult hematology cohort that was screened by Aspergillus-specific biomarkers (7).
CT technology and reading process.Sixty-seven CT scans were performed as part of the clinical management. CT scans of the thorax were acquired as low-dose examinations during inspiration with institutional age- and weight-adapted parameters (80 to 120 kV, 30 to 50 mA, Care Dose 4D). All scans were performed on a 64-slice state-of-the-art scanner (Siemens Sensation 64; Siemens Healthcare, Erlangen, Germany). The scans were reconstructed iteratively in thin slices (1 mm). In addition to the original analysis, two independent experts who were blind to clinical and mycological information retrospectively reviewed the CT images for confirmation. The definition of Aspergillus-specific lung lesions was based on published EORTC/MSG criteria (8).
Covariate analysis.Individual patient-specific covariates were recorded, including gender, neutropenia (neutrophil count of <500/μl for >10 days), corticosteroids (0.3 mg/kg of body weight for more than 21 days), T-cell-immunosuppressive medication (cyclosporine, tacrolimus, or mycophenolate mofetil [MMF] at any dose), cytomegalovirus (DNA detection in plasma), diagnosis of other herpesviridae (herpes simplex virus 1 or 2, human herpesvirus 6, Epstein-Barr virus, varicella-zoster virus) or adenoviridae, detection of respiratory viruses (e.g., influenza virus, parainfluenza virus, coronavirus, or rhinovirus), use of itraconazole, fluconazole, liposomal amphotericin B, voriconazole, posaconazole, or caspofungin (all in therapeutic dosages), CT evidence of IA, fever (>38.5°C, or >38°C for >4 h), cough, tachycardia, oxygen demand, and tachypnea. In addition, the continuous covariates of body weight (kg), days of neutropenia (neutrophil count of <500/μl for >10 days), days of corticosteroid use (0.3 mg/kg body weight), age, GM index, numbers of CD3+, CD4+, CD8+, CD16/CD56+, and CD19+ cells per μl of blood, and numbers of leukocytes, monocytes, neutrophils, and lymphocytes per μl of blood were recorded.
Statistics.A frequency analysis test was performed: any patient with at least one positive CT scan was considered CT positive, whereas any patient with no positivity in CT scans was considered CT negative. The same scheme was performed for Aspergillus PCR and GM assays. Then, the association between CT positivity, Aspergillus positivity, and GM positivity was explored using the Pearson chi-square test of independence.
The relationship between the Aspergillus PCR assay result (as a binary reference variable) and the GM index (as a continuous classifying variable) was explored by means of the ROC curve method for all observations reporting both assays.
Moreover, a nonparametric time-to-event (log rank) analysis was performed. Two main event variables were identified and used to perform two corresponding approaches to time-to-event analysis: (i) the day of the binary Aspergillus PCR status change from negative to positive, and (ii) the day of the first conversion of the serum GM assay to a positive result (galactomannan assay index [GMI] ≥ 0.5). The effect of each binary and continuous covariate was estimated singly for both outcomes.
Statistical methods were performed with Stata 14.1 (StataCorp, College Station, TX, USA) and R version 3.2.3 (R Foundation, Vienna, Austria).
ACKNOWLEDGMENTS
This work was supported by Pfizer Pharma GmbH, Berlin, Germany (grant no. IIR-WI-187426).
C. P. Heussel has received lecture and consultation fees from Schering-Plough, Pfizer, Basilea, Boehringer Ingelheim, Novartis, Roche, Astellas, Gilead, Merck Sharp & Dohme, Lilly, Intermune, Fresenius, Olympus, Essex, AstraZeneca, Bracco, MEDA Pharma, Intermune, Chiesi, Siemens, Covidien, Pierre Fabre, Bayer, and Grifols, as well as research funding from Siemens, Pfizer, Boehringer Ingelheim, and MeVis. The other authors report no conflicts of interest.
FOOTNOTES
- Received 11 August 2016.
- Returned for modification 1 September 2016.
- Accepted 12 October 2016.
- Accepted manuscript posted online 19 October 2016.
Supplemental material for this article may be found at https://doi.org/10.1128/JCM.01682-16.
- Copyright © 2016 American Society for Microbiology.
