Galactomannan detection in broncho‐alveolar lavage fluid for invasive aspergillosis in immunocompromised patients

Koen de Heer; Marije G Gerritsen; Caroline E Visser; Mariska MG Leeflang

doi:10.1002/14651858.CD012399.pub2

Detección de galactomananos en el líquido de lavado broncoalveolar para la aspergilosis invasiva en pacientes inmunocomprometidos

Declaraciones de intereses de los autores

Versión publicada: 20 mayo 2019 Historial de versiones

https://doi.org/10.1002/14651858.CD012399.pub2

Contraer todo Desplegar todo

Resumen

disponible en

Antecedentes

La aspergilosis invasiva (AI) es una micosis oportunista potencialmente mortal que se presenta en algunos individuos con un sistema inmunitario comprometido. El análisis de inmunoadsorción enzimática (ELISA) de galactomanano en suero fue aceptado rápidamente y de manera generalizada como parte de los estudios de diagnóstico en un paciente con AI presunta. Debido a la naturaleza no invasiva, puede utilizarse como prueba de detección sistemática. El ELISA también puede realizarse en el lavado broncoalveolar (LBA), lo que permite el muestreo de las áreas próximas a la infección. Sin embargo, la naturaleza invasiva del LBA cambia la función de la prueba de galactomanano de manera significativa, por ejemplo, al impedir su uso como prueba de detección sistemática.

Objetivos

Evaluar la exactitud diagnóstica de la detección de galactomanano en el LBA para el diagnóstico de AI en pacientes inmunocomprometidos, a diferentes valores de corte para la positividad de la prueba, en conformidad con el Manual Cochrane de exactitud de pruebas diagnósticas (Cochrane Diagnostic Test Accuracy Handbook).

Métodos de búsqueda

Se efectuaron búsquedas en tres bases de datos bibliográficas, incluido MEDLINE, el 9 de septiembre de 2016, con los términos "aspergilosis" y "galactomannan" como palabras de texto y títulos cuando correspondía. Se revisaron las listas de referencias de los estudios incluidos para obtener estudios adicionales.

Criterios de selección

Se incluyeron estudios de cohortes que examinaron la exactitud del galactomanano en LBA para el diagnóstico de AI en pacientes inmunocomprometidos si utilizaban la clasificación de la European Organization for Research and Treatment of Cancer/Invasive Fungal Infections Cooperative Group y el National Institute of Allergy and Infectious Diseases Mycoses Study Group (EORTC/MSG) como valor de referencia.

Obtención y análisis de los datos

Dos autores de la revisión evaluaron la calidad de los estudios y extrajeron los datos. Para la evaluación de la calidad se utilizó la Quality Assessment of Diagnostic Accuracy Studies‐2 (QUADAS‐2).

Resultados principales

Se incluyeron 17 estudios en esta revisión. El riesgo de sesgo fue alto en dos o más dominios de todos los estudios excepto uno. El rendimiento diagnóstico de un índice de densidad óptica (IDO) de 0,5 como valor de corte se informó en 12 estudios (con 1123 pacientes). La sensibilidad estimada fue de 0,88 (intervalo de confianza [IC] del 95%: 0,75 a 1,00) y la especificidad fue de 0,81 (IC del 95%: 0,71 a 0,91). El rendimiento de un IDO de 1,0 como valor de corte pudo determinarse en 11 estudios (con 648 pacientes). La sensibilidad fue de 0,78% (IC del 95%: 0,61% a 0,95%), y la especificidad fue de 0,93% (IC del 95%: 0,87% a 0,98). Con un IDO límite de 1,5 o mayor, la heterogeneidad en la especificidad disminuyó de modo significativo y fue invariablemente > 90%.

Conclusiones de los autores

El valor de corte óptimo depende de la incidencia local y la vía de atención clínica. Con una prevalencia de un 12%, una población hipotética de 1000 pacientes incluye a 120 pacientes con AI. A un valor de corte de 0,5 serán omitidos 14 pacientes con AI y habrá 167 pacientes con un diagnóstico incorrecto de AI. Si se utiliza la prueba a un valor de corte de 1,0; se omiten 26 pacientes con AI. y 62 pacientes recibirán un diagnóstico incorrecto de aspergilosis invasiva. Las poblaciones y los resultados fueron muy heterogéneos. Por lo tanto, la interpretación y la extrapolación de estos resultados se deben realizar con cuidado. Un resultado de 1,5 en el IDO o mayor parece ser un indicador fuerte de AI.

Resumen en términos sencillos

disponible en

Medición de galactomananos en el líquido de lavado pulmonar para el diagnóstico de la aspergilosis invasiva en pacientes con deterioro del sistema inmunitario

¿Por qué es importante mejorar la exactitud diagnóstica de las micosis invasivas?

Algunas personas tienen un sistema inmunológico debilitado y en esos casos se dice que están "inmunocomprometidas". Las personas pueden estar inmunocomprometidas a causa de presentar leucemia o VIH/SIDA y por el uso de algunas formas de medicación.

En el paciente inmunocomprometido, se ve disminuida la capacidad del sistema inmunitario para combatir las infecciones. Esto puede dar lugar a una infección que no se presenta en condiciones normales. la denominada infección oportunista. Un ejemplo es la aspergilosis, una infección causada por un hongo. Como el sistema inmunitario no la bloquea, la infección se disemina por el organismo. El hongo "invade" el cuerpo y entonces se habla de aspergilosis invasiva. que puede ser letal. El diagnóstico precoz permite un tratamiento oportuno e impide que el hongo siga diseminándose por el organismo.

¿Qué prueba diagnóstica se estudió en la revisión?

Se buscó determinar la efectividad de la "detección de galactomanano" para diagnosticar una infección de aspergilosis invasiva.

La detección del galactomanano es una técnica diagnóstica. Consiste en hacer pasar un instrumento a través de la boca o la nariz hacia las vías respiratorias. Se inyecta líquido hacia una región del pulmón. Este líquido remueve los hongos y otros microbios del pulmón y luego se recoge y analiza. Este proceso se llama lavado broncoalveolar (lavado pulmonar) (LBA). El galactomanano, un componente de la pared celular del hongo, es uno de los productos que pueden recogerse en el líquido del LBA.

El resultado de la prueba de galactomanano en LBA se incluye en los criterios internacionales usados para diferenciar entre los pacientes con probable aspergilosis invasiva y los pacientes en quienes es menos probable la infección. Si se satisfacen criterios múltiples (p.ej. estudios por imágenes y síntomas y signos clínicos), entonces los pacientes se tratan con medicación antimicótica.

¿Cuál era el objetivo de esta revisión?

El objetivo de esta revisión era determinar la tasa de error de la prueba de galactomanano en el LBA. Para esto, se buscaron estudios sobre este tema. El resultado de la prueba de galactomanano consiste en un "índice de densidad óptica" (IDO) (y no en un "sí" o "no"). Hubo diferencias entre los estudios en cuanto al valor de corte por encima del cual un resultado se consideró positivo. La mayoría consideró un resultado por encima de un IDO de 0,5 o 1,0 como positivo.

¿Cuáles son los principales resultados de esta revisión?

Considérese un grupo de 1000 personas con un deterioro del sistema inmunitario en que 120 (12%) presentan aspergilosis invasiva. Los resultados de esta revisión indican que, en teoría, si se utilizara la prueba de galactomanano en el LBA CON un IDO de 0,5 o mayor para decidir qué resultados son positivos:

‐ unos 273 tendrán un resultado que indica que probablemente presenten aspergilosis invasiva, y de estos, 167 se clasificarán de modo incorrecto como con aspergilosis invasiva; y

‐ de los 727 individuos con un resultado que indica que es probable que no haya aspergilosis invasiva, 14 se clasificarán de modo incorrecto como sin aspergilosis invasiva.

Si se utiliza un valor de corte mayor para determinar la positividad de la prueba, entonces más pacientes tendrán un resultado falso negativo, y menos pacientes, un resultado falso positivo.

¿Cuán confiables son los resultados de los estudios en esta revisión?

En los estudios incluidos, el diagnóstico de aspergilosis invasiva se realizó con la evaluación de todos los pacientes con criterios internacionales (valores de referencia). Es probable que haya sido un método fiable para decidir si los pacientes realmente presentaban aspergilosis invasiva. Hubo variabilidad en los resultados de los diversos estudios, y había pocos pacientes con aspergilosis invasiva. Por lo tanto, las cifras representan un promedio entre los estudios de la revisión y no se sabe si siempre se obtendrán estos resultados.

¿Para quiénes son relevantes los resultados de esta revisión?

Los resultados son aplicables sobre todo a pacientes adultos que reciben quimioterapia intensiva o un trasplante de células madre para una forma de cáncer de células de la sangre.

¿Cuál es el grado de actualización de esta revisión?

Los autores de la revisión buscaron y utilizaron los estudios publicados hasta el 9 de septiembre 2016.

Authors' conclusions

Implications for practice

Institutions can use our results to decide on the optimal cut‐off level, taking their local incidence and the clinical pathway into consideration. Although the sensitivities of the two cut‐off levels seems nearly equal, assuming a cohort of 1000 patients and a disease incidence of 12%, at a cut‐off level of 0.5, the test will on average be falsely negative in 14 patients with invasive aspergillosis (IA), and will falsely diagnose IA in 167 patients. If we use the test at a cut‐off value of 1.0, we will miss 26 patients with IA and will incorrectly diagnose IA in 62 patients (see Summary of main results).

Based on our reviews, it is not possible to determine whether the serum or bronchoalveolar lavage (BAL) galactomannan is a better test, as the diagnostic accuracies were determined in different studies with a large heterogeneity (See the Summary of main results). Probably, the tests complement each other, as the other test might be positive in cases the first test is negative in. Although the diagnostic accuracy found in our review is not as high as in other reviews, it does support its prominent role in the current reference standard.

Implications for research

Research on IA is associated with a large number of methodological difficulties, as described in Applicability of findings to the review question, most importantly incorporation and misclassification bias. However, the generally low‐methodological quality of the studies in our review should be improved upon. Many of the theoretically plausible forms of bias were not dealt with in the design of most studies. Although there was a significant effect of study design on the performance statistics, the current body of evidence did not allow us to determine which forms of bias did and did not play a role.

A number of questions remains unanswered. It would be interesting if future research would focus more on clinical pathways and the relative value of other tests, such as serum galactomannan and the various polymerase chain reactions (PCR's). To this end it would be helpful if studies reported all components of the reference standard clearly. More studies on homogeneous and well‐described patient populations would also be welcome. Reporting incorporation is paramount in determining the effect and amount of incorporation. Reporting all cut‐off values would be helpful in determining the optimal cut‐off value. The quality of systematic reviews would improve if they could be performed as an individual patient data (IPD) analysis instead of a meta‐analysis.

Summary of findings

Open in table viewer

Summary of findings Summary of findings table

What is the diagnostic accuracy of the Platelia© Aspergillus ELISA for invasive aspergillosis in immunocompromised patients?
Patients/population: immunocompromised patients, mostly neutropenic haematology patients. Prior testing: varied. Index test: the Platelia© ELISA, a sandwich ELISA for galactomannan, an Aspergillus antigen. Importance: currently regarded as the most accurate test to diagnose invasive aspergillosis. Reference standard: the EORTC/MSG criteria. Studies: cohort studies reporting all EORTC‐MSG categories and of which the 2 × 2 or 4 × 2 table contained not more than one index test per participant.
Cut‐off value	No. of participants	Summary estimates (95% CI)	Implications	Quality and Comments
0.5 ODI (12 studies)	229 with proven or probable IA	Sensitivity: 0.88 (95% CI 0.75 to 1.00)	At a prevalence of 12%, 120 out of 1000 patients will develop proven or probable IA. Of these, 14 will be missed. Of the 880 patients with possible or no IA, 167 will have a false‐positive test result and may be unnecessarily treated. Of all 273 patients with a positive test, 106 will indeed have proven or probable IA; of all 727 negative tests, 14 will have IA after all (2%).	All studies except one had a high risk of bias in two or more domains. For all studies the concerns regarding applicability were low.
	894 with possible or no IA	Specificity: 0.81 (95% CI 0.71 to 0.91)
1.0 ODI (11 studies)	177 with proven or probable IA	Sensitivity: 0.78 (95% CI 0.61 to 0.95)	At a prevalence of 12%, 120 out of 1000 patients will develop proven or probable IA. Of these, 26 will be missed. Of the 880 patients with possible or no IA, 62 will have a false‐positive test result and may be unnecessarily treated. Of all 156 patients with a positive test, 94 will indeed have proven or probable IA; of all 844 negative tests, 26 will have IA after all (3%).
	594 with possible or no IA	Specificity: 0.93 (95% CI 0.87 to 0.98)
Note, the populations and results were very heterogeneous. Therefore, interpretation and extrapolation of these results has to be performed with caution.
CI: confidence interval;IA: invasive aspergillosis; ODI: optical density index

Background

Target condition being diagnosed

Although it has been suggested that the mortality rate has decreased over recent years due to more effective therapy (Kwon 2012; Sun 2017), invasive aspergillosis (IA) is a life‐threatening opportunistic mycosis. Published mortality rates range from 30% to 70% (Aisner 1977; Robenshtok 2007; von Eiff 1995). Invasive aspergillosis is caused by ubiquitous spores of Aspergillus species, inhaled by everyone on a daily basis, which can cause invasive disease in the case of a failing immune system (Chotirmall 2013). As IA mainly occurs in people with immune deficiency, the incidence varies depending on the degree of immune deficiency. In people who are highly immune deficient, such as people with acute myeloid leukaemia (AML) receiving highly intensive chemotherapy, the incidence during their treatment can exceed 5% (Robenshtok 2007). Mainly because of the increasing application of immunosuppressive treatments, especially highly intensive chemotherapy, prolonged corticosteroid therapy and stem cell transplantation (Kaneko 2001; Rubio 2009), the incidence of IA is increasing. Most often the lungs or airways (or both) are the site of infection. But IA can also occur in the nasal sinus, central nervous system, skin, gastro‐intestinal tract, the deep structures of the eye, and prosthetic heart valves; or may give rise to a disseminated infection.

Index test(s)

Galactomannan is a cell wall component of Aspergillus species and Penicillium species (Latgé 1994). It is excreted by the fungus during the growth phase. It is suggested that the level of galactomannan antigenaemia is proportional to the fungal tissue load and has a prognostic value (Koo 2010).

Currently, galactomannan detection in bronchoalveolar lavage (BAL), usually performed as a result of a chest computerised tomography (CT) scan consistent with IA, is considered the most accurate test to diagnose or exclude IA. A meta‐analysis in 2010 reported a sensitivity of 90% and a specificity of 94% (Guo 2010).

Currently, there is one commercially available assay for the detection of galactomannan in BAL, the Platelia™ sandwich enzyme‐linked immunosorbent assay (ELISA) test. It can also be applied on serum, cerebrospinal fluid, pleural fluid and urine (Ascioglu 2001; Hope 2005). Results of the ELISA are given as an optical density index (ODI), which is the ratio of the optical density of (usually) 1 ng/mL galactomannan versus the optical density of the sample. In order to enhance the specificity of the test, the Food and Drug Aministration (FDA) changed the recommended cut‐off for positivity in the USA from an ODI of 0.5 to 1.0 (U. S. Food and Drug Administration 2011).

Clinical pathway

Indication

There is no generally accepted indication for galactomannan testing in BAL, and no generally accepted consequence of a specific test result.

Clinical signs and symptoms of IA are non‐specific at an early stage. Establishing a diagnosis at an early stage of infection improves the chances of survival through subsequent earlier treatment (Upton 2007; von Eiff 1995). A definitive diagnosis can be made by two means. By autopsy, or "by a positive culture of A.fumigatus in a specimen obtained with a sterile procedure from a normally sterile site that is clinically or radiologically abnormal and consistent with an infectious disease process (excluding bronchoalveolar lavage fluid, a cranial sinus cavity specimen, and urine)" (De Pauw 2008). However, the person's clinical condition (frequently including thrombocytopenia (reduced numbers of platelets) because of preceding chemotherapy) often prohibits invasive techniques. Cultures are frequently false‐negative. In fact, definitive diagnoses of IA are rare. In clinical practice one usually has to work with presumed diagnoses with a varying level of suspicion. In 2008, a committee consisting of the Invasive Fungal Infections Cooperative Group of the European Organization for Research and Treatment of Cancer (EORTC), the Mycoses Study Group (MSG) and the National Institute of Allergy and Infectious Diseases proposed to grade the diagnosis of IA with three levels of probability: 'proven', 'probable' and 'possible' IA (see Table 1; De Pauw 2008). There is also a fourth category of people without IA. The divisions are based on host factor criteria (a set of risk factors, which are designed to almost preclude IA if all are absent, an example being chemotherapy‐induced prolonged neutropenia, clinical criteria (specific clinical and radiological signs such as a chest CT consistent with IA), and microbiological criteria (for example, Aspergillus species cultured from BAL). This consensus was a revision of a consensus reached in 2002, see Table 2. A positive galactomannan test, whichever fluid it is applied on, is a microbiological criterion in both versions. Clinical studies have shown that these criteria do not match autopsy results perfectly. This holds especially true for the 'possible' category. For clinical trials investigating the effect of treatment of IA, it is recommended that only people in 'proven' or 'probable' categories are considered to have IA (Borlenghi 2007; Subirà 2003).

Open in table viewer

Table 1. EORTC/MSG criteria, 2008 version

	Host criterion	Clinical criterion	Microbiological criterion
No IA	Not required	Not required	Not required
Possible IA	Required	Required	Not required
Probable IA	Required	Required	Required

IA: invasive aspergillosis

Open in table viewer

Table 2. EORTC/MSG criteria, 2002 version

	Host criterion	Clinical criterion (1 major or 2 minor)	Microbiological criterion
No IA	Not required	Not required	Not required
Possible IA	Required	Either one of these is required
Probable IA	Required	Required	Required

IA: invasive aspergillosis

Because of its invasiveness, the test is usually employed in the clinical pathway of a person at risk of IA, who is also suspected of having IA based on clinical and/or radiological grounds. The test is rarely used as a screening test in asymptomatic persons, whether they are at risk or not. In the setting of chemotherapy‐induced prolonged neutropenia, the host factor that causes the majority of cases of IA, a full work‐up is performed in most centres as a result of a number of indications, including most notably: fever not responsive to broad spectrum antibiotics, a positive galactomannan test on serum, or hyphae or moulds found in a respiratory specimen. This work‐up will include a chest CT. In case it is consistent with IA, a BAL and the index test are performed, if the clinical condition permits.

Clinical consequence

The consequences of a positive or negative index test vary among hospitals and will usually depend on the EORTC/MSG classification after the full diagnostic work‐up.

Despite a negative index test, a person can still be classified as having 'probable' IA (based on another microbiological criterion) or as having 'proven' IA. Also, he or she can still be classified as having 'possible' IA due to a clinical criterion. As BAL is usually performed because of a chest CT consistent with IA (a clinical criterion) in a patient at risk of IA (a host criterion), a classification of 'no' IA will not occur very frequently. For the same reason, in case of a positive index test a person will often be classified as having 'probable' IA.

Someone with 'probable' IA will usually be considered to have IA and treated accordingly. Although it is very well possible that a person classified as having 'possible' IA in fact has 'no' IA, the likelihood rises, and in light of the high mortality many clinicians will also prefer to treat such a person for IA. So it could seem the test has no clinical consequence. However, the duration of therapy with expensive and toxic antimycotics is usually longer in probable aspergillosis than in possible aspergillosis. It can also influence the choice of antimycotic, as the odds of an invasive mould infection other than aspergillosis decrease with a positive test result. Moreover, in case of side effects it is more likely a clinician would continue therapy in a patient with probable aspergillosis. Moreover, further immunosuppressive treatment such as chemotherapy or stem cell transplantation, which is often essential in curing a patient, is contra‐indicated in people with IA.

Risk factors

There are other risk factors for IA besides chemotherapy‐induced prolonged neutropenia that increasingly warrant the index test. Most notable are:

allogeneic stem cell transplantation and/or graft‐versus‐host disease;
intensive care unit (ICU) admittance;
chronic pulmonary disease, such as chronic obstructive pulmonary disease (COPD) and bronchiectasis;
acquired immune deficiency syndrome (AIDS);
immunosuppressive drugs, such as tumour necrosis factor‐alpha (TNF‐a) inhibitors;
long‐term steroid use;
inherited severe immunodeficiency;
decompensated cirrhosis; and
influenza.

The 2002 version of the EORTC/MSG criteria also contained the following host criteria:

fever refractory to broad‐spectrum antibacterial treatment in high‐risk patients; and
'proven' or 'probable' invasive fungal infection during previous episode of neutropenia.

Alternative test(s)

There are many alternative tests, such as direct microscopy, histopathological examination, microbiological culture, and less frequently performed: beta‐D‐glucan testing and various polymerase chain reactions (PCRs). These can be performed on a number of specimens, such as respiratory samples, serum, BAL or tissue samples. The galactomannan ELISA can also be applied on other fluids, such as cerebrospinal fluid (CSF), pleural fluid and urine. None has a reported sensitivity or specificity comparable with the index test (Cuenca‐Estrella 2011; Guo 2010; Karageorgopoulos 2011; Lamoth 2012; Pickering 2005; Zou 2012). There are published Cochrane Reviews on the index test performed on serum (Leeflang 2015), and on PCR blood tests (Cruciani 2015).

Rationale

There are already a number of meta‐analyses on the diagnostic value of BAL galactomannan. However, the fact that the index test plays an important role in the set of criteria that forms the reference standard confers a difficulty in interpreting such analyses. Our meta‐analysis adds an evaluation of this and other sources of heterogeneity to the previous meta‐analyses.

Objectives

To assess the diagnostic accuracy of galactomannan detection in bronchoalveolar lavage (BAL) for the diagnosis of invasive aspergillosis (IA) in people who are immunocompromised, at different cut‐off values for test positivity, in accordance with the Cochrane Diagnostic Test Accuracy Handbook.

Secondary objectives

Our secondary objective was to study several possible sources of heterogeneity:

the effect of cut‐off value,
the effect of incorporation bias (the incorporation of the index test in the reference standard),
the effect of pre‐selection based on a positive serum galactomannan test,
the effect of the revision of the reference standard,
the effect of a protocol that defines the exact and explicit indication of the index test,
study design,
the effect of sponsoring of the study by the manufacturer of the commercially available galactomannan test,
clinical subgroups.

Methods

Criteria for considering studies for this review

Types of studies

Eligible studies were those that assessed the diagnostic accuracy of galactomannan detection in bronchoalveolar lavage (BAL). The galactomannan enzyme‐linked immunosorbent assay (ELISA) could be assessed alone or in comparison with (an)other test(s).

Participants

Immunocompromised participants who have been tested to diagnose, exclude or screen for invasive aspergillosis (IA) using galactomannan detection in BAL. We defined immunocompromised as the presence of one of the EORTC/MSG host criteria, either of the 2002 or 2008 version. The current version of the EORTC/MSG criteria was anticipated to be updated with two extra host criteria: chronic airway abnormalities and decompensated cirrhosis. We also included studies on participants with immune deficiencies as per the anticipated version of the EORTC/MSG criteria. If in more than 10% of the participants a EORTC/MSG host criterion was not present, a study was excluded.

Index tests

The commercially available galactomannan sandwich ELISA (Platelia™) performed on BAL was the test under evaluation. Studies addressing detection in serum and other body fluids were excluded.

Target conditions

The target condition of this review was invasive pulmonary aspergillosis.

Reference standards

The following reference standards were used to define the target condition:

autopsy specimens, with histopathological evidence of aspergillosis and a positive culture;
the EORTC/MSG criteria, see Table 3 and Table 4 for the criteria in full detail.

Open in table viewer

Table 3. Full EORTC/MSG criteria, 2008 version

Possible / probable invasive aspergillosis
Host factors	Recent history of neutropenia (< 0.5 x 10 9 neutrophils/L [< 500 neutrophils/mm³] for > 10 days) temporally related to the onset of fungal disease. Receipt of an allogeneic stem cell transplant Prolonged use of corticosteroids (excluding among patients with allergic bronchopulmonary aspergillosis) at a mean minimum dose of 0.3 mg/kg/day of prednisone equivalent for > 3 weeks Treatment with other recognized T cell immunosuppressants, such as cyclosporine, TNF‐a blockers, specific monoclonal antibodies (such as alemtuzumab), or nucleoside analogues during the past 90 days Inherited severe immunodeficiency (such as chronic granulomatous disease or severe combined immunodeficiency
Clinical criteria	Lower respiratory tract fungal disease (every reasonable attempt should be made to exclude an alternative etiology), with the presence of 1 of the following 3 signs on CT Dense, well‐circumscribed lesions(s) with or without a halo sign Air‐crescent sign Cavity Tracheobronchitis: tracheobronchial ulceration, nodule, pseudomembrane, plaque, or eschar seen on bronchoscopic analysis Sinonasal infection: Imaging showing sinusitis plus at least 1 of the following 3 signs: Acute localized pain (including pain radiating to the eye) Nasal ulcer with black eschar Extension from the paranasal sinus across bony barriers, including into the orbit CNS infection: 1 of the following 2 signs: Focal lesions on imaging Meningeal enhancement on MRI
Mycological criteria	Direct test (cytology, direct microscopy, or culture) showing mold in sputum, bronchoalveolar lavage fluid, bronchial brush, or sinus aspirate samples, indicated by 1 of the following: Presence of fungal elements indicating a mold Recovery by culture of a mold (e.g., Aspergillus, Fusarium, Zygomycetes, orScedosporium species) Indirect tests (detection of antigen or cell‐wall constituents) Aspergillosis: Galactomannan antigen detected in plasma, serum, bronchoalveolar lavage fluid, or CSF Invasive fungal disease other than cryptococcosis and zygomycoses: b‐d‐glucan detected in serum
Proven aspergillosis
Criteria	Histopathologic, cytopathologic, or direct microscopic examination of a specimen obtained by needle aspiration or biopsy in which hyphae or melanized yeast‐like forms are seen accompanied by evidence of associated tissue damage; or, Recovery of a mold or “black yeast” by culture of a specimen obtained by a sterile procedure from a normally sterile and clinically or radiologically abnormal site consistent with an infectious disease process, excluding bronchoalveolar lavage fluid, a cranial sinus cavity specimen, and urine.

Open in table viewer

Table 4. Full EORTC/MSG criteria, 2002 version

Possible / Probable invasive aspergillosis
Host factors	Neutropenia (< 500 neutrophils/mm³ for > 10 days) Persistent fever for > 96 h refractory to appropriate broad‐spectrum antibacterial treatment in high‐risk patients Body temperature either > 38°C or < 36°C and any of the following predisposing conditions: prolonged neutropenia (> 10 days) in previous 60 days, recent or current use of significant immunosuppressive agents in previous 30 days, proven or probable invasive fungal infection during previous episode of neutropenia, or coexistence of symptomatic AIDS Signs and symptoms indicating graft‐versus‐host disease, particularly severe (grade ≥ 2) or chronic extensive disease Prolonged (> 3 weeks) use of corticosteroids in previous 60 days
Clinical criteria	Must be related to site of microbiological criteria and temporally related to current episode Lower respiratory tract infection Major: Any of the following new infiltrates on CT imaging: halo sign, air‐crescent sign, or cavity within area of consolidation, in absence of infection by organisms that may lead to similar radiological findings including cavitation, such as Mycobacterium, Legionella, and Nocardia species. Minor: Symptoms of lower respiratory tract infection (cough, chest pain, haemoptysis, dyspnoea); physical finding of pleural rub; any new infiltrate not fulfilling major criterion; pleural effusion Sinonasal infection Major: Suggestive radiological evidence of invasive infection in sinuses (i.e., erosion of sinus walls or extension of infection to neighbouring structures, extensive skull base destruction) Minor: Upper respiratory symptoms (e.g. nasal discharge, stuffiness); nose ulceration or eschar of nasal mucosa or epistaxis; periorbital swelling; maxillary tenderness; black necrotic lesions or perforation of hard palate CNS infection Major: Radiological evidence suggesting CNS infection (e.g. mastoiditis or other parameningeal foci, extradural empyema, intraparenchymal brain or spinal cord mass lesion) Minor: Focal neurological symptoms and signs (including focal seizures, hemiparesis, and cranial nerve palsies); mental changes; meningeal irritation findings; abnormalities in CSF biochemistry and cell count (provided that CSF is negative for other pathogens by culture or microscopy and negative for malignant cell
Microbiological criteria	Positive result of culture for mold (including Aspergillus, Fusarium, or Scedosporium species or a Zygomycetes) or Cryptococcus neoformans or an endemic fungal pathogen from sputum or bronchoalveolar lavage fluid samples Positive result of culture or findings of cytologic/direct microscopic evaluation for mold from sinus aspirate specimen Positive findings of cytologic/direct microscopic evaluation for mold or Cryptococcus species from sputum or bronchoalveolar lavage fluid samples Positive result for Aspergillus antigen in specimens of bronchoalveolar lavage fluid, CSF, or ≥ 2 blood samples Positive result for cryptococcal antigen in blood sample Positive findings of cytologic or direct microscopic examination for fungal elements in sterile body fluid samples (e.g. Cryptococcus species in CSF) Positive result for Histoplasma capsulatum antigen in blood, urine, or CSF specimens Two positive results of culture of urine samples for yeasts in absence of urinary catheter Candida casts in urine in absence of urinary catheter Positive result of blood culture for Candida species
Proven aspergillosis
Criterion	Histopathologic or cytopathologic examination showing hyphae from needle aspiration or biopsy specimen with evidence of associated tissue damage (either microscopically or unequivocally by imaging); or positive culture result for a sample obtained by sterile procedure from normally sterile and clinically or radiologically abnormal site consistent with infection, excluding urine and mucous membranes.

The results within each possible EORTC classification (‘no’, ‘possible’, ‘probable’ and ‘proven’ IA) had to be reported, or the results within the combined groups ‘no/possible’ and ‘probable/proven’ IA.

Search methods for identification of studies

Electronic searches

We searched Embase and MEDLINE through Ovid for aspergillosis and galactomannan as Medical Heading and text word, see Appendix 1 for the exact terms. The appendix also lists the terms that were used to search the Web of Science Core Collection (1970 to date of search). The version of MEDLINE we searched was Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations, Ovid MEDLINE(R) Daily and Ovid MEDLINE(R) 1946 to 9th September 2016. We searched the literature from 2000 onwards, as the reference standard was published in 2002.

We conducted the searches on 9 September 2016. All studies published after the date of the initial search will be included in the next update of the systematic review. The search for the update is planned for September 2021.

Searching other resources

To identify additional published studies, we:

entered the included studies into PubMed and then used the Related Articles feature;
searched the Science Citation Index to identify articles that cite the included studies;
checked the reference lists of all included studies.

Data collection and analysis

Selection of studies

The first selection, based on title and abstract, was performed by two review authors (KdH, MG). We excluded articles that were judged to be highly unlikely to report the test performance of galactomannan detection in bronchoalveolar lavage, such as animal studies, plant studies, or studies on fungi other than Aspergillus species. Afterwards, we located the full paper of each potentially eligible article. Two of four review authors (KdH, MG, CV, ML) independently assessed all eligible articles for inclusion. We resolved disagreement by discussion. We included articles when we could not resolve disagreement.

We excluded studies:

that were not published in an English peer‐reviewed journal (we will include these in the first update of this review);
that did not enrol patients consecutively;
that were case‐control studies;
of which the 2 × 2 or 4 × 2 table contained more than one index test per participant;
that (partially) contained the data of another study/publication;
that did not report all index test results of the studied cohort, e.g. a study that does not report the results of the 'possible' IA and/or 'no' IA group;
that contained an ostensible calculation error in the 2 × 2 or 4 × 2 table.

We contacted the study authors for more information before excluding studies on the last two grounds.

Data extraction and management

The following data were extracted:

author, year of publication and journal;
our application of the QUADAS‐2 tool, see Appendix 2, for which a data‐extraction form and background document had been designed;
the data for a 4 × 2 or 2 × 2 table.

The data‐extraction form was accompanied by a background document that stated how each item on the form was to be interpreted. We standardised the form and piloted it on two studies. Four review authors (KdH, MG, CV, ML) extracted data and assessed quality. One review author (KdH) has a haematological background, one (MG) a medical background, one (CV) a microbiological background, and one (ML) a methodological background. The articles were randomly allocated to a pair of review authors. Disagreements were resolved by discussion.

Assessment of methodological quality

Study quality was assessed using the QUADAS‐2 data‐extraction form. For four domains a number of pre‐defined signalling questions were scored as 'yes', 'no' or 'unclear', or had to be answered with free text. Then, per domain the risk of bias was judged to be either 'low', 'high' or 'unclear', based on the answers to these signalling questions (Whiting 2011). Results are presented in text, in a graph and in a table. We did not calculate a summary score estimating the overall quality of an article since the interpretation of such summary scores is problematic and potentially misleading (Jüni 1999; Whiting 2005). We did not assess the concern about applicability for the patient domain, as we did not have a specific population of interest in mind and expected to find studies with a wide range of patients. After all, one cannot say that in a study on lung transplant patients (in which, quite understandably, the diagnostic accuracy is very different from the diagnostic accuracy which is found in a study on allogeneic stem cell transplantation patients) there is a problem with applicability. However, the results of the study are not applicable to patients undergoing allogeneic stem cell transplantation. Instead, we chose to carefully document the type of participants. For the index test and reference standard domains, we classified all studies to have a low applicability concern because our inclusion criteria stipulated that we only include studies that used the galactomannan ELISA performed on BAL as the index test and the EORTC/MSG criteria as reference standard. Therefore no studies that were subject to applicability concerns on these two domains were included.

Statistical analysis and data synthesis

The EORTC/MSG criteria were our reference standards. They classify participants into one of four groups: 'proven' IA, 'probable' IA, 'possible' IA and 'no' IA. This results in a two‐by‐four table: a positive or negative galactomannan test result versus each of the four reference groups. To calculate test accuracy, we defined the 'proven' and 'probable' participants as having IA; and we defined 'no' IA and 'possible' IA participants as not having IA, in order to construct two‐by‐two tables. Other classifications of participants having and not having IA were assessed in a sensitivity analysis.

We calculated the sensitivity and specificity for each study per reported cut‐off value. We also plotted the individual study results in forest plots and the receiver operating characteristic (ROC) space.

We used a bivariate random‐effects approach for a meta‐analysis of the pairs of sensitivity and specificity (Reitsma 2005; Cochrane Diagnostic Test Accuracy Handbook). Covariates were incorporated in the bivariate model to examine the effect of potential sources of bias and variation across subgroups of studies. Due to the bivariate nature of the model, effects of covariates on sensitivity and specificity could be modelled separately.

For the meta‐analysis we used the two‐by‐two table of one cut‐off value per study. Otherwise, the more cut‐off values a study reports, the larger its impact on the outcome of the analysis becomes. We chose the thresholds of 0.5 or 1.0, if reported, because these are the positivity thresholds currently and formerly recommended. We expected most studies in our analysis to report either an optical density index (ODI) of 1.0 or 0.5. If both were reported, we selected one at random. As the random inclusion could have had an effect on the outcome of the meta‐analysis, we performed a sensitivity analysis by including the two‐by‐two tables of an ODI of 0.5 as well as an ODI of 1.0, if both were reported.

All analyses were performed in SAS, version 9.4, using Proc NLMIXED.

Investigations of heterogeneity

Heterogeneity was investigated first through visual examination of the forest plots and the raw ROC plot of all data in the review, at all cut‐off values. Then, the potential sources of heterogeneity were included in the meta‐analysis as a covariate. We assumed no interaction between the sources of heterogeneity. We addressed the following possible sources.

a. Effect of cut‐off value

A main source of heterogeneity in diagnostic test accuracy reviews are the differences in the applied cut‐off values between studies. We first investigated what the influence of these cut‐off values was on the sensitivity and specificity by including the cut‐off value as a covariate in the meta‐analysis.

b. Effect of incorporation bias (incorporation of the index test in the reference standard)

A positive galactomannan test is incorporated in the EORTC/MSG criteria. Studies that do not exclude this criterion from the reference standard might overestimate the diagnostic accuracy of the index test. Vice versa, a study that does exclude this criterion from the reference standard might underestimate the diagnostic accuracy. We studied the effect of incorporation of the BAL and serum galactomannan test, as these are dependent tests.

c. Effect of pre‐selection based on a positive serum galactomannan test

The indication for the index test in a study, and if applicable the clinical pathway in which it is embedded, are major determinants of the accuracy of the index test. We expected many studies not to contain a clinical pathway and the triggers to vary greatly, or even to be documented unclearly. We extracted as much detail as possible on the indication for the index test, the associated clinical pathway and the tests performed prior to the index test, but we expected that analysis of heterogeneity on this form of bias would not be feasible.

We did expect all studies to report whether a positive serum galactomannan could have been the indication for the index test. We studied the effect of this possibility.

d. Effect of the reference standard

The criteria of the EORTC/MSG were our reference standard. We studied the effect of the 2008 update of these criteria.

e. Effect of a protocol that defines the exact and explicit indication of the index test

We studied the effect of the presence of a formal indication for performing the index test in the study protocol.

f. Study design

We defined a study that uses a data set not created for the purpose of a research question as retrospective. We defined a study that includes patients consecutively after the formulation of a research question as prospective. We studied the heterogeneity as an effect of study design.

g. Effect of sponsoring

We studied the effect of sponsoring of the study by the manufacturer of the commercially available galactomannan test.

h. Clinical subgroups

We explored the possible influence of clinical subgroups by including the following additional covariates in the meta‐analysis.

children versus adult participants — we assumed a cut‐off at age 16 to 18 to differentiate between children and adults;
participants with neutropenia versus all other participant groups; we also performed a subgroup analysis comparing groups based on the different host criteria;
routine use of anti‐mould prophylaxis in a study (yes versus no);
the department(s) where the patients were recruited.

Sensitivity analyses

a. Methodological quality

To assess whether methodological quality influenced the results, we performed a sensitivity analysis by leaving out the studies with a relatively low methodological quality.

b. Effect of positive and negative reference standard definition

Although it is uncertain whether a participant with 'possible' IA has invasive aspergillosis or not, we considered a participant with 'possible' IA as having no invasive aspergillosis and we combined these participants with those with 'no' IA when calculating sensitivity and specificity. We acknowledge the methodological difficulty of this choice and therefore studied its effect in a sensitivity analysis. We also examined the effect of ignoring 'possible' IA ('proven' and 'probable' versus 'no' IA).

Results

Results of the search

Our search was performed on 9 September 2016 and resulted in 5844 hits. The initial screening, excluding the studies that were judged beyond any doubt not to contain original data on the diagnostic value of bronchoalveolar lavage (BAL) galactomannan in diagnosing invasive aspergillosis (IA), or that were not published in a peer‐reviewed journal, resulted in 103 eligible reports of which the full text was retrieved (Figure 1). We excluded two non‐English reports from the assessment which were studies eligible for inclusion. We will assess the translated versions at the first update of this review.

Figure 1

Study flow diagram

We sent corresponding authors an email if we thought it possible that a study could be included with additional information or data, i.e. if not all index test results were reported or if the report contained an ostensible error. This resulted in sending 29 emails, from which we received 10 replies.

After considering these 103 reports:

33 reports were judged not to be a study containing data on the diagnostic value of BAL galactomannan in the diagnosis of IA, or were not published in a peer‐reviewed journal, or were a case‐control study,
23 reports were judged not to use the EORTC/MSG criteria without significant modification as reference standard,
28 reports were judged not to report all EORTC/MSG categories (proven, probable, possible, no IA),
for 4 reports, we were unable to repeat the statistical calculations,
2 reports did not report the index test specifically (but for example, as a composite with another test such as serum galactomannan),
3 reports were judged to be a double publication,
3 reports did not report on a per‐patient basis,
1 report did not use the Platelia galactomannan assay as the galactomannan test.

This led to the exclusion of 87 reports, as more criteria could be applicable per study, and the inclusion of 16 reports. As one report contained two methodologically different studies, in total 17 studies were included.

No extra studies were found through additional searches or reference checking. Two studies were included after the receipt of additional data per email.

The studies comprised 1634 patients in total. 'Probable' or 'proven' IA was diagnosed in 366 patients, of which 34 were specifically denoted as having 'proven' IA and 177 as having 'probable' IA. Of the remaining patients, 222 were specifically denoted as having 'possible' IA and 538 as having 'no' IA. There were 479 patients with a positive BAL galactomannan and 1155 with a negative BAL galactomannan. On average, 90.8 patients were included in a study (range 30‐210).

Methodological quality of included studies

Of the 17 studies, 10 were judged to have a high risk of bias with respect to the domain of Patient Selection. These studies were unclear about whether they enrolled participants consecutively.

Sixteen studies were judged to have a high risk of bias regarding the Index Test domain. In these studies the index test was either interpreted with knowledge of the reference standard, or using a cut‐off value that was not predefined, or this was not specified.

Based on the signalling questions in the Reference Standard domain, 15 studies were considered at a large risk of bias. In these studies either the reference standard was interpreted with knowledge of the index test, or the reference standard was not strictly adhered to (e.g., the presence of a host criterion was not confirmed), or the authors were not specific on these points.

Finally, with respect to the domain of Flow and Timing, 10 studies were considered to be associated with a high risk of containing bias. In these studies the review authors judged that there was (a high probability of) a significant exclusion of study participants from the analysis.

These data are summarised in Figure 2 and Figure 3.

Figure 2

Risk of bias and applicability concerns graph: review authors' judgements about each domain presented as percentages across included studies

Figure 3

Risk of bias and applicability concerns summary: review authors' judgements about each domain for each included study

Findings

Raw data

In seven studies both the 0.5 optical density index (ODI) and 1.0 ODI data were reported (see Figure 4). For three of these, we even had the individual patient data (IPD) available, allowing us to investigate every possible cut‐off value. The IPD of two studies were kindly sent to us by email. In five studies the diagnostic performance of an ODI of 0.5 was reported, but not of an ODI of 1.0. Four studies only reported the 1.0 ODI data but not the 0.5 ODI data. One study reported only data at an ODI of 0.8.

Figure 4

ROC plot; all studies reporting more than 1 cut‐off value.

Thus, the diagnostic performance of an ODI of 0.5 was reported in 12 studies (with 1123 patients), with sensitivity ranging between 52.2% and 100% and specificity ranging between 40.4% and 94.3%. The performance of an ODI of 1.0 could be determined in 11 studies (with 648 patients), with sensitivity ranging between 42.9% and 100% and specificity ranging between 75.6% and 100%.

There was considerable variation in the reported sensitivity and specificity (see Figure 5 and Figure 6).

Figure 5

Forest plot of tests: 1 ODI 0.5, 2 ODI 0.8, 3 ODI 0.85 and 0.87, 4 ODI 1.0, 5 ODI 1.5, 6 ODI 2.0, 7 ODI 2.5.

Figure 6

ROC Plot of all data in the review; all cut‐off values.

Figure 5 shows the diagnostic accuracy estimates of the individual studies in tables and Forest plots. Figure 6 shows the summary ROC plot. In Figure 4, a ROC plot is shown containing all studies that reported the diagnostic accuracy statistics of more than one cut‐off value.

The variation in the reported specificity might partially be explained by chance variation, because of the small number of patients with the target condition in the various studies, on average 23.0% with a minimum of 6.2%, a maximum of 45.3% and a standard deviation of 12.4%. Above a cut‐off ODI of 1.5, the heterogeneity in specificity decreased significantly and specificity was invariably > 90%. At an ODI of 1.5 or more the index test thus seems a reliable test to rule in the index disease (see Figure 5 and Figure 4).

Meta‐analysis

We performed a meta‐analysis of the pairs of sensitivity and specificity using a bivariate random‐effects approach, with cut‐off as a covariate, see summary of findings Table and Figure 7. One included study did not report accuracy data at a cut‐off ODI of 0.5 or 1.0 (Heng 2014). We did not foresee this possibility and did not address this situation in our protocol. We chose to exclude this study from the analyses.

Figure 7

Plain language summary test consequence figure

At a cut‐off level of 0.5, this resulted in an estimated sensitivity and specificity of 0.88 (95% confidence interval (CI): 0.75 to 1.00) and 0.81 (95% CI 0.71 to 0.91), respectively. The sensitivity for cut‐off 1.0 was lower: 0.78 (95% CI 0.61 to 0.95), but this was not statistically significant (P = 0.36). The specificity for cut‐off 1.0 was higher than for cut‐off 0.5: 0.93 (95% CI 0.87 to 0.98), which was statistically significant. These summary estimates and confidence regions are graphically shown in Figure 8.

Figure 8

Summary estimates and confidence regions at cut‐off ODI 0.5 and 1.0.

We performed a sensitivity analysis by including from one study both the accuracy data at the ODI cut‐off level of 0.5 as well as the accuracy data at the ODI cut‐off level of 1.0, if both were reported, see Table 5. The sensitivity and specificity were 0.84 (95% CI 0.74 to 0.94) and 0.83 (95% CI 0.76 to 0.90), respectively, at a cut‐off ODI of 0.5. At a cut‐off ODI of 1.0 the sensitivity was 0.80 (95% CI 0.69 to 0.92), and the specificity 0.92 (95% CI 0.89 to 0.96). The P values were 0.66 for the sensitivity and 0.01 for the specificity.

Open in table viewer

Table 5. Meta‐analysis of the pairs of sensitivity and specificity, sensitivity analysis

Meta‐analysis of the pairs of sensitivity and specificity, sensitivity analysis
	0.5 ODI (n = 12)	1.0 ODI (n = 11)
sensitivity	0.84 (95% CI 0.74 to 0.94)	0.80 (95% CI 0.69 to 0.92)
specificity	0.83 (95% CI 0.76 to 0.90)	0.92 (95% CI 0.89 to 0.96)

CI: confidence interval; ODI: optical density index

Investigations of heterogeneity

Incorporation bias

We examined the possibility of heterogeneity due to incorporation bias, see Table 6. We compared two groups of studies: one with the studies that incorporated serum and/or BAL galactomannan in their reference standard (or were unclear on this matter); versus a group with the studies that specifically excluded both galactomannan tests from their reference standard. In the latter group, the sensitivity and specificity were 0.98 (95% CI 0.95 to 1.00) and 0.72 (95% CI 0.53 to 0.90), respectively. In the former group the sensitivity was 0.74 (95% CI 0.63 to 0.86; P value for difference in sensitivity 0.01) and the specificity 0.91 (95% CI 0.86 to 0.95; P value for difference in specificity 0.02).

Open in table viewer

Table 6. Quantification of the incorporation bias

	with incorporation (n=12)	without incorporation (n = 4)
sensitivity	0.74 (95% CI 0.63 to 0.86)	0.98 (95% CI 0.95 to 1.00)
specificity	0.91 (95% CI 0.86 to 0.95)	0.72 (95% CI 0.53 to 0.90)

The effect of pre‐selection based on a positive serum galactomannan test

Few studies reported this quality item. We were therefore unable to investigate this possible source of heterogeneity.

The effect of the 2008 update of the reference standard

Four studies used the 2002 version, and 13 studies the 2008 version of the reference standard. Using this as a covariate had no effect on sensitivity or specificity.

The effect of a protocol that defines the exact and explicit indication of the index test

Whether or not there was a strict and exact indication for performing the index test had no significant effect on sensitivity or specificity.

Study design

We examined the possibility of heterogeneity due to study design, see Table 7. We compared the prospective with the retrospective (or on this matter unclear) studies. In the latter group the sensitivity and specificity were 0.94 (95% CI 0.86 to 1.00) and 0.78 (95% CI 0.65 to 0.91), respectively. In the former group the sensitivity was 0.74 (95% CI 0.60 to 0.88) and the specificity 0.91 (95% CI 0.86 to 0.96). The corresponding P values were 0.03 for the sensitivity and 0.03 for the specificity.

Open in table viewer

Table 7. Quantification of the effect of the study design

	prospective (n = 6)	retrospective (n = 10)
sensitivity	0.74 (95% CI 0.60 to 0.88)	0.94 (95% CI 0.86 to 1.00)
specificity	0.91 (95% CI 0.86 to 0.96)	0.78 (95% CI 0.65 to 0.91)

CI: confidence interval

The effect of sponsoring of the study by the manufacturer of the commercially available galactomannan test

Whether or not the manufacturer of the commercially available galactomannan test sponsored a study had no significant effect on sensitivity or specificity.

Other planned investigations

Department(s) where the participants were recruited

As 10 studies were performed in haematology departments, one in a paediatric department, one in a lung transplant centre, one in various transplant departments, and in four studies the departments were not clearly specified, we were able to examine the effect of dividing the departments in two groups: haematology departments versus other or unclear departments. This division had no significant effect in our meta‐analysis, when used as a covariate.

Additional investigations that proved not possible

As there was only one study that mainly included children a meta‐analysis on this item was not possible. As 10 studies included patients with a mix of host criteria, and the other studies were varied with respect to the study population, an analysis of the effect of host criteria (e.g. neutropenic patients versus the rest) was not possible. As there were no studies included in which prophylaxis was specifically withheld, an investigation into the effect of anti‐mould prophylaxis was not possible.

Sensitivity analysis

Methodological quality

We performed a sensitivity analysis per QUADAS‐2 domain to determine the influence of methodological quality, leaving out the studies with a high risk of bias.

In the Index Test domain there was only one study with a low risk of bias, and in the Reference Test domain only two. A sensitivity analysis was therefore not possible for these domains. In the Patient Selection and in the Flow and Timing domains the same six studies had a low risk of bias. The sensitivity analyses were identical and can be seen in Table 8. The sensitivity and specificity in the overall group were 0.83 (95% CI 0.70 to 0.92) and 0.88 (95% CI 0.80 to 0.92), respectively. In the low risk of bias group these were lower, respectively 0.75 and 0.86.

Open in table viewer

Table 8. Sensitivity analyses on methodological quality

Sensitivity analyses on methodological quality Patients and flow and timing domains
	overall (all studies, n = 16)	low risk of bias (n = 6)
sensitivity	0.83 (95% CI 0.70 to 0.92)	0.75 (95% CI 0.51 to 0.89)
specificity	0.88 (95% CI 0.80 to 0.92)	0.86 (95% CI 0.78 to 0.91)

CI: confidence interval

Effect of reference standard definition

Finally, we performed a sensitivity analysis on the definition of the reference standard by determining the sensitivity and specificity when following a number of pre‐specified different definitions, see Table 9. Unfortunately, we were unable to perform a bivariate analysis comparing 'proven' versus 'probable' / 'possible' / 'no' aspergillosis, as the variance was unstable and the model did not converge. We therefore performed two univariate analyses for this definition.

Open in table viewer

Table 9. Effect of reference standard definition

	Definition	Studies	Sensitivity	95% CI		Specificity	95% CI
	Definition	Studies	Sensitivity	Lower limit	Upper limit	Specificity	Lower limit	Upper limit
1	Prov and Prob vs Pos and No	16	0.83	0.70	0.92	0.88	0.80	0.92
2	Prov vs Prob and Pos and No	11	0.77	0.33	0.96	0.73	0.63	0.81
3	Prov and Prob and Pos vs No	11	0.59	0.40	0.75	0.94	0.89	0.97
4	Possibles excluded	11	0.79	0.66	0.88	0.94	0.89	0.97

CI: confidence interval; Prob: probable; Prov: proven

Excluding the participants with 'possible' aspergillosis had only a small effect on sensitivity and increased specificity slightly. Defining participants with 'probable' aspergillosis as not having the target condition decreased sensitivity and specificity. Considering participants with 'probable' and 'possible' aspergillosis as having the target condition lowered sensitivity significantly and increased specificity.

Discussion

Summary of main results

We included 17 studies on 1634 participants in our review. Meta‐analysis of their results led to an estimated sensitivity of 0.88 (95% CI 0.75 to 1.00) and an estimated specificity of 0.81 (95% CI 0.71 to 0.91) at a cut‐off value of an ODI of 0.5. At a cut‐off value of an ODI of 1.0 these statistics were 0.78 (95% CI 0.61 to 0.95) and 0.93 (95% CI 0.87 to 0.98), respectively.

When the Platelia galactomannan test was introduced, there was no validated threshold value for use on bronchoalveolar lavage (BAL). This is reflected by the variable primary cut‐off levels used in the studies of our review. The effect of the cut‐off value on performance statistics, the threshold effect, can be seen in summary of findings Table. Because of the magnitude of this effect and the fact that clinical practice requires a cut‐off value, we chose not to calculate overall performance statistics but to calculate them per threshold value.

The difference in diagnostic performance of a cut‐off value of 0.5 versus 1.0 can be illustrated using a hypothetical group of 1000 potential invasive aspergillosis (IA) patients. As an example, we will assume a disease incidence of 12%. In such a population, 120 patients will have 'proven' or 'probable' IA, and 880 will not. If we use the test at a cut‐off value of 0.5, then based on the BAL galactomannan result we will miss 14 patient with IA (sensitivity 88%, 12% false‐negative rate). And there will be 167 patients incorrectly diagnosed with invasive aspergillosis (specificity of 81%, 19% false‐positive rate). If we use the test at a cut‐off value of 1.0, we will miss 26 patients with IA (sensitivity 78%, 22% false‐negative rate). There will be 62 patients incorrectly diagnosed with invasive aspergillosis (specificity of 93%, 7% false‐positive rate). Therefore, the optimal cut‐off value will depend on the local incidence of invasive aspergillosis, in an ideal situation specifically determined for the population at hand. And, it will depend on the relative importance of false‐positive and false‐negative results, as determined by the role of the index test within the clinical pathway.

Many quality features were poorly reported in the included studies, e.g. whether the clinicians were blinded, how uninterpretable or intermediate index and reference test results were dealt with, whether the results of the reference standard were known when the index test result was interpreted (and vice versa), the time interval between the execution of the index test and reference standard, and how the reference standard was exactly executed. Many studies did not confirm that all patients qualified as having a EORTC/MSG host criterion, or this seemed not to be the case in all participants. Many studies included patients with varying underlying pathologies and did not report extensively on subgroups. Studies often did not have a detailed clinical protocol specifying the indication and exact timing of the index test.

The sensitivity was more heterogeneous than specificity. Part of this heterogeneity can be explained by the inclusion of small studies and the on average low prevalence of IA. The cut‐off value had a significant effect on specificity but not on sensitivity. This could also be due to the low incidence of IA.

Comparison with other systematic reviews

A number of systematic reviews on BAL galactomannan have been published in recent years, see Table 10 (Guo 2010; Zou 2012; Heng 2015).

Open in table viewer

Table 10. Other systematic reviews

Systematic review	Number of studies	Number of studies also included in our review	Cut‐off ODI	Sensitivity (95% CI)	Specificity (95% CI)
Heng 2015	16	6	0.5	0.82 (0.70–0.91)	0.92 (0.85–0.96)
Heng 2015	16	6	1	0.75 (0.55–0.88)	0.95 (0.87–0.98)
Guo 2010	13	7	overall	0.90 (0.79‐0.96)	0.94 (0.90‐0.96)
			0.5	0.86 (0.70‐0.94)	0.89 (0.85‐0.92)
			1	0.85 (0.72‐0.93)	0.94 (0.89‐0.97)
Zou 2012	30	8	0.5	0.87 (0.79‐0.92)	0.89 (0.85‐0.92)
Zou 2012	30	8	1	0.86 (0.76‐0.92)	0.95 (0.91‐0.97)

CI: confidence interval; ODI: optical density index

Guo 2010 wrote the first review, aiming to produce an overall sensitivity and specificity. If more than one cut‐off value was reported in a study, the cut‐off value that offered the best test performance was used. This resulted in an overall sensitivity and specificity of 0.90 and 0.94. The authors included 12 studies (not counting Becker and colleagues twice), and did not focus on a specific population.

Heng 2015 focused specifically on haematological patients and included 16 studies. At a cut‐off level of 0.5, the sensitivity was found to be 0.82 and the specificity 0.92. At a cut‐off of 1.0 sensitivity dropped to 0.75 and specificity rose to 0.95.

Zou 2012. performed the systematic review with the largest number of studies, 30 in total (not counting Becker and colleagues twice), and included non‐English studies. The review did not focus on a specific population. A striking phenomenon in this review was the fact that increasing the cut‐off level from 0.5 to 1.0 increased the sensitivity from 0.78 to 0.94. This suggests large differences between the studies in the 0.5 and 1.0 group, overcompensating the threshold effect.

We excluded from our review many of the studies that were included in the other reviews. We excluded, for example, case‐control studies, studies that did not report all EORTC/MSG categories, studies that used a modified reference standard, and non‐English studies. Of the 16 studies Heng 2015 included, we excluded 10. Of the 30 Zou 2012 included, we excluded 22. We excluded six of the 13 studies Guo 2010 selected.

Table 10 suggests that the overall performance statistics reported in Guo 2010 may be an overestimation, as the overall performance statistics are better than the statistics at the 0.5 and 1.0 ODI cut‐off level. At the 0.5 ODI cut‐off level our review estimated the sensitivity to be lower and the specificity to be higher than the other reviews. The other reviews had comparable performance statistics. At the 1.0 ODI cut‐off level all reviews were similar.

Comparison with serum galactomannan and polymerase chain reactions (PCRs) on blood

Comparing our review with the Cochrane Review on serum galactomannan (Leeflang 2015), our review seems to show that the diagnostic accuracy of the Platelia test on BAL is slightly better than when performed on serum. When using an ODI of 0.5 as a cut‐off value, the sensitivity of the serum test was 0.78 and the specificity was 0.85. At a cut‐off value of 1.0, the sensitivity was 0.71 and the specificity was 0.90. At a cut‐off value of 1.5, the sensitivity was 0.63 and the specificity was 0.93. A biological explanation of better performance on BAL could be the sampling in the immediate vicinity of the infection.

It has been argued that airway invasion results in a rise of the amount of galactomannan in BAL fluid (Hope 2007). And that penetration of the endothelial cell layer by hyphae (angio‐invasive growth) could lead to more galactomannan in serum (Mennink‐Kersten 2004). Most likely, there is a degree of complementariness between BAL and serum galactomannan. The results of a number of studies supports this hypothesis (Bergeron 2012a; Fortún 2016; Hidalgo 2009).

The use of the galactomannan test in serum and BAL was validated in different studies. Both our review on serum galactomannan (Leeflang 2015) and our current review showed a large amount of heterogeneity. Therefore, the comparison of the reviews on BAL and serum galactomannan is hazardous and it is not possible to determine whether the serum or BAL galactomannan obviates the use of the other test. As stated, the tests probably complement each other. It would be highly interesting for future research to focus on the relative value of both tests.

The Cochrane Library also published a review on PCR tests on blood to diagnose invasive aspergillosis in immunocompromised patients (Cruciani 2015). The overall sensitivity and specificity of a single test (80.5% and 78.5%, respectively) are about comparable with the serum galactomannan test. For two consecutive positive test results these become 58.0% and 96.2%, respectively. An methodological challenge in research on PCRs to diagnose aspergillosis is the lack of a standardised protocol on how to perform the PCR. This has been improved on by for example the European Aspergillosis PCR Initiative (EAPCRI).

Strengths and weaknesses of the review

Strengths

The main strength of our review is the rigorous application of strict methodological criteria. This led to the exclusion of studies that were included in other reviews. For example, we excluded case‐control studies, which are often used because of the low incidence of IA. However, longitudinal cohort studies are preferable, as cases and controls are taken from the same population and the prevalence of the disease can be determined precisely. We also excluded studies that enrolled participants non‐consecutively or non‐randomly, studies that did not report all EORTC/MSG categories, and studies that modified the reference standard.

Weaknesses

Because of the rigorous application of methodological criteria relatively few studies were included in our review. Even these studies are characterised by a high risk of bias, and a large amount of heterogeneity in sensitivity and specificity as well as in the distribution of patients with 'proven', 'probable', 'possible' and non‐IA. Therefore, we were not able to investigate many of the predetermined sources of heterogeneity. Notwithstanding these weaknesses, our results show that the central role that is usually given to the galactomannan test in clinical practice is justified.

We excluded two non‐English studies from the assessment which studies were eligible for inclusion, as we are only fluent in English. In the planned update of this review in 2021, we will also include non‐English studies.

Applicability of findings to the review question

We decided not to assess the concern about applicability for the patient domain as we expected to find studies with a large amount of heterogeneity and a wide range of patient populations at risk for IA. We excluded no subgroups, to be able to determine generally applicable performance statistics. Instead, we chose to carefully document the types of participants and tried to analyse heterogeneity. Such analyses require large sample sizes, which may be difficult to obtain in studies on rare infections such as invasive aspergillosis. This emphasises the need for systematic reviews.

We expected the heterogeneity in the studies to derive from five categories:

the heterogenic manifestations of the disease,
the heterogenic characteristics of the patients,
heterogenic study protocols,
the imperfect nature of the reference standard; and
incorporation bias.

Heterogeneity in disease manifestations

As stated in the introduction, invasive aspergillosis (IA) consists of many subforms, of which invasive pulmonary aspergillosis (IPA) is the most common variant. IPA can manifest itself as an alveolar infection or a tracheobronchitis, the latter being seen more frequently in HIV/AIDS and patients with a lung transplant. The alveolar infections can be located centrally or more peripherally. This raises the question whether the BAL procedure is of influence on the diagnostic accuracy (see "Heterogeneity in study protocols" below). And whether the sensitivity of BAL might be lower in peripherally localised aspergillosis.

The Aspergillus species involved

The assay performance might vary with the Aspergillus species involved, as some might produce different quantities of galactomannan (Swanink 1997; Taghizadeh‐Armaki 2017). A lower sensitivity has been reported in A. fumigatus infection in comparison with other Aspergillus species (13% versus 49%; P = 0.001; Hachem 2009).

Heterogeneity in patient characteristics

The populations and the results of the studies in our review were very heterogeneous. Therefore, the interpretation and extrapolation of these results has to be performed with caution.

Preventative antifungal therapy

Preventative antifungal therapy might reduce fungal burden, leading to false‐negative results (Eigl 2015; Prattes 2014; Prattes 2015; Reinwald 2012a). Some clinical studies reported this (Luong 2010; Maertens 2009; Marr 2004; Park 2010), while others reported the opposite (Musher 2004), or found no effect (Frealle 2009; Nguyen 2011).

Racil 2011 reported a higher sensitivity in patients who just started antifungal therapy, and a lower sensitivity after two to three days of therapy, indicating a need to perform BAL within days of starting antifungal therapy. However, Frealle 2009 did not find this association.

Our data did not allow quantification of the effect of mould‐active antifungal agents.

Neutropenia

Neutropenia could cause an increase in fungal load, thereby decreasing the false‐negative rate. Conflicting results are reported. The sensitivity of BAL galactomannan varied from being no different (Maertens 2009) to significantly higher in neutropenic patients (Racil 2011). D'Haese 2012 reported that neutropenia influences the sensitivity of galactomannan in serum but not in BAL fluid.

We were unable to test whether there was an influence of neutropenia.

Colonisation

Aspergillus species can colonise the pulmonary tract and, as they are ubiquitous in the environment, their spores are inhaled on a daily basis. Colonisation can lead to false‐positive results. This becomes even more relevant in populations with an underlying condition that predisposes to colonisation, such as cystic fibrosis, bronchiectasis or after a lung transplant. In such populations the specificity can be expected to be lower. Indeed, the one study in our review on lung transplant patients had a markedly low specificity (Pasqualotto 2010).

Our data did not allow formal testing of this effect.

Cross‐reactive moulds

Some non‐Aspergillus moulds such as Histoplasma capsulatum (Min 2012; Narreddy 2008; Wheat 2007), Penicillium marneffei (Huang 2007; Rimek 1999), Fusarium oxysporum (Nucci 2014; Tortorano 2012) and Alternaria species (Kappe 1993) may cross‐react with galactomannan epitopes. These potential cross‐reactions could be important in the geographic areas in which these moulds are endemic. We were unable to test this hypothesis.

Cross‐reactive epitopes

Betalactam antibiotics such as amoxicillin‐clavulanate and piperacillin‐tazobactam, which are often given to these patients, have been reported to caused false‐positive results, as they are derived from natural compounds produced by the genus Penicillium (Bart‐Delabesse 2005; Boonsarngsuk 2010).

We had insufficient data to examine this possible relation.

Heterogeneity in study protocols

The study protocol has a major impact on test performance. For example, it can define a certain period of fever unresponsive to broad spectrum antibiotics as an indication for the index test. A shorter period of fever corresponds with an earlier disease phase that will be harder to detect. Shortening the period will therefore lower the estimated sensitivity of the index test. A second example is that the study protocol defines the study population and that disease prevalence influences diagnostic test accuracy. For instance, a population with a high disease prevalence may also include more severe cases of disease (Mulherin 2002).

Interpretation of the reference standard

The composite nature of the reference standard – that states that "[e]very reasonable attempt should be made to exclude an alternative etiology" for lower respiratory signs – raises the question how far one should go to exclude other aetiologies, and whether this diagnostic work‐up should be standardised in a study. For example, a patient with a pulmonary nodule and a negative BAL culture will be classified as having 'possible' aspergillosis. However, if a trans‐thoracic biopsy was performed, the patient would have been classified as having 'proven' aspergillosis if it was positive. The indication for trans‐thoracic biopsy as well as the extent of the microbiological work‐up on BAL in case of suspected IA differs among clinics. This introduces bias. Moreover, as the reference standard calls for an evaluation of the possibility of alternative aetiologies, subjectivity is introduced.

Multiple samples per patient

The number of measurements per patient can also introduce bias. Sensitivity and specificity are either calculated on a per‐patient basis or a per‐test basis. This means a patient is labelled as a case patient or as a control patient depending on whether they developed IA during their treatment course (per‐patient basis) or had it at the time of the index test (per‐test). In case of per‐patient analysis, a larger number of samples will result in a higher likelihood of a positive test due to chance (and usually, the better the performance statistics will be). Both per‐patient and per‐test analysis can lead to bias when different numbers of samples are obtained for each patient, or if the timing of samples is not systematically applied to each patient. The more they are clustered around the moment of infection, the better the performance statistics will become. To avoid these problems, some included studies used only one index test per patient. For example, the first one performed (Becker 2003 ‐‐ part I), or the one with the highest result (de Mol 2013) Effectively, the latter strategy still uses multiple tests per patient.

Bronchoalveolar lavage (BAL) procedure

Computerised tomography (CT) scans are ubiquitously available, and should preferably be used to guide the lavage. In that setting the probability of a misguided lavage, for example due to multifocal and multifactorial pathology, seems low. However, it is not known which fraction of the BAL is associated with the most accurate test result. Often the homogenised alveolar sample is used. Hsu 2010 tested the bronchial sample. Comparative studies found that sensitivity may be lower when testing bronchial samples (Racil 2011, Zou 2012). It could also depend on the localisation of the infection. The volume of aspirated fluid is another parameter that seems important as higher volumes have been associated with higher false‐negative rates (Racil 2011). Correction of the test result using a dilution marker might be a solution, as a universal standard seems unfeasible (Rennard 1986). False‐positive results could additionally result from contamination at processing, transport, or testing. This has been reported for Plasmalyte, which is used as lavage fluid and components of which are produced by A.niger (Hage 2007, Spriet 2016). Protracted storage has also been reported to lead to false‐negative results (Dufresne 2014, Johnson 2013, Johnson 2014; Kimpton 2014; Wheat 2014).

As the included studies did not use or did not report a study protocol, or had incomparable protocols, we were unable to quantify this bias. We did exclude studies with a varying number of tests per patient that reported performance statistics on a per‐test basis.

Study design

A prospective study is preferable to a retrospective study. It results in a cohort that is better defined in terms of patient characteristics. Moreover, the index test and reference standard are often performed in a more standardised fashion. This is of special importance when the timing of the index test is of influence. In our meta‐analysis, the study design had a large effect on the sensitivity and specificity, perhaps corresponding with an earlier BAL in the prospective studies.

The reference standard

Although the first three sources of heterogeneity could be solved by meta‐analysis of large studies that are methodologically well construed, two challenges remain that are currently unsolvable.

Misclassification bias

There is no reliable reference standard for IA. Aspergillus infection can be proven by the isolation and culture of the organism from a normally sterile site (De Pauw 2008). However, biopsies are associated with complications, and examination of tissue samples and BAL can be negative despite ultimately proven infection. This difficulty is reflected by the reference standard that contains grades of certainty, and the fact that it had to be revised. The reference standard contains many subjective items. It calls for the interpretation of tests by radiologists, pathologists, microbiologists and pulmonologists. This can be associated with intra‐ and inter‐observer variability. Misclassification will result in a wrong estimation of the prevalence and can underestimate a tests' sensitivity and specificity.

A non‐blinded study can lead to further underestimation as a clinician who is aware of a positive test result could start therapy. This could in turn prevent the reference standard from ever becoming positive. Because of the imperfect nature of the reference standard and the promising first study results, the index test has already become part of the current reference standard. It can be argued that blinding a clinician from the results is unethical. In 2013, a randomised controlled trial confirmed the clinical value of galactomannan testing on serum samples (Morrissey 2013).

A relatively minor problem with the reference standard is the fact that it focuses on invasive fungal disease, not aspergillosis. If microscopy or cytology is positive as a mycological criterion but all cultures are negative, it is not clear whether a patient has aspergillosis or not. If the patient also has a host and clinical criterion, he has 'probable' invasive mycosis. But whether he has 'probable' invasive aspergillosis is unclear.

Moment of infection

Not only is there no reliable way to determine whether a patient has the target disease, but due to the insidious and protracted course of the disease it is also not possible to determine the moment a patient became infected. This makes it hard to interpret a positive index test that was performed a number of days before the reference test was performed. This is called disease progression bias. Was it an early sign of the disease, a false‐positive test or a different episode of the same disease that was cured or spontaneously resolved?

The composite nature of the reference standard makes it nearly unavoidable that there will be varying time intervals between the index and reference test. Moreover, as the reference test consists of multiple tests, it is not a single point in time.

Although misclassification and disease progression bias could have played a role in the large amount of heterogeneity in our review, it is not possible to quantify the effect of these biases.

Incorporation bias

A second and related major difficulty is incorporation bias. As stated, the index test is part of the current reference standard. Following this definition in a diagnostic study leads to a form of circular reasoning, overestimating sensitivity and specificity. As an extreme example, if the index test is only performed in case of a CT scan showing a clinical criterion, a positive index test will result in a positive reference standard and the specificity will always be 100%. On the other hand, not incorporating the galactomannan test will underestimate specificity (and to a lesser extent sensitivity) as a negative reference standard does not exclude the disease.

In our review, we found a large difference between studies that did and did not avoid incorporation bias. Strangely enough, the sensitivity increased spectacularly in the group that avoided incorporation bias. This could be due to chance, because of the relatively low number of studies in this group (n = 4); or due to the fact that the studies were performed in populations in which false‐negative results are rare. The studies were three prospective studies, including only neutropenic haematological patients, and the only study on lung transplant patients.

Figure 1

Study flow diagram

Ir a la figura de la revisiónAbrir en una pestaña nueva

Figure 2

Risk of bias and applicability concerns graph: review authors' judgements about each domain presented as percentages across included studies

Ir a la figura de la revisiónAbrir en una pestaña nueva

Figure 3

Risk of bias and applicability concerns summary: review authors' judgements about each domain for each included study

Ir a la figura de la revisiónAbrir en una pestaña nueva

Figure 4

ROC plot; all studies reporting more than 1 cut‐off value.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Figure 5

Forest plot of tests: 1 ODI 0.5, 2 ODI 0.8, 3 ODI 0.85 and 0.87, 4 ODI 1.0, 5 ODI 1.5, 6 ODI 2.0, 7 ODI 2.5.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Figure 6

ROC Plot of all data in the review; all cut‐off values.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Figure 7

Plain language summary test consequence figure

Ir a la figura de la revisiónAbrir en una pestaña nueva

Figure 8

Summary estimates and confidence regions at cut‐off ODI 0.5 and 1.0.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Test 1

ODI 0.5.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Test 2

ODI 0.8.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Test 3

ODI 0.85 and 0.87.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Test 4

ODI 1.0.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Test 5

ODI 1.5.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Test 6

ODI 2.0.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Test 7

ODI 2.5.

Ir a la figura de la revisiónAbrir en una pestaña nueva

Summary of findings Summary of findings table

What is the diagnostic accuracy of the Platelia© Aspergillus ELISA for invasive aspergillosis in immunocompromised patients?
Patients/population: immunocompromised patients, mostly neutropenic haematology patients. Prior testing: varied. Index test: the Platelia© ELISA, a sandwich ELISA for galactomannan, an Aspergillus antigen. Importance: currently regarded as the most accurate test to diagnose invasive aspergillosis. Reference standard: the EORTC/MSG criteria. Studies: cohort studies reporting all EORTC‐MSG categories and of which the 2 × 2 or 4 × 2 table contained not more than one index test per participant.
Cut‐off value	No. of participants	Summary estimates (95% CI)	Implications	Quality and Comments
0.5 ODI (12 studies)	229 with proven or probable IA	Sensitivity: 0.88 (95% CI 0.75 to 1.00)	At a prevalence of 12%, 120 out of 1000 patients will develop proven or probable IA. Of these, 14 will be missed. Of the 880 patients with possible or no IA, 167 will have a false‐positive test result and may be unnecessarily treated. Of all 273 patients with a positive test, 106 will indeed have proven or probable IA; of all 727 negative tests, 14 will have IA after all (2%).	All studies except one had a high risk of bias in two or more domains. For all studies the concerns regarding applicability were low.
	894 with possible or no IA	Specificity: 0.81 (95% CI 0.71 to 0.91)
1.0 ODI (11 studies)	177 with proven or probable IA	Sensitivity: 0.78 (95% CI 0.61 to 0.95)	At a prevalence of 12%, 120 out of 1000 patients will develop proven or probable IA. Of these, 26 will be missed. Of the 880 patients with possible or no IA, 62 will have a false‐positive test result and may be unnecessarily treated. Of all 156 patients with a positive test, 94 will indeed have proven or probable IA; of all 844 negative tests, 26 will have IA after all (3%).
	594 with possible or no IA	Specificity: 0.93 (95% CI 0.87 to 0.98)
Note, the populations and results were very heterogeneous. Therefore, interpretation and extrapolation of these results has to be performed with caution.
CI: confidence interval;IA: invasive aspergillosis; ODI: optical density index

Summary of findings Summary of findings table

Ir a la tabla de la revisión

Table 1. EORTC/MSG criteria, 2008 version

	Host criterion	Clinical criterion	Microbiological criterion
No IA	Not required	Not required	Not required
Possible IA	Required	Required	Not required
Probable IA	Required	Required	Required
IA: invasive aspergillosis

Table 1. EORTC/MSG criteria, 2008 version

Ir a la tabla de la revisión

Table 2. EORTC/MSG criteria, 2002 version

	Host criterion	Clinical criterion (1 major or 2 minor)	Microbiological criterion
No IA	Not required	Not required	Not required
Possible IA	Required	Either one of these is required
Probable IA	Required	Required	Required
IA: invasive aspergillosis

Table 2. EORTC/MSG criteria, 2002 version

Ir a la tabla de la revisión

Table 3. Full EORTC/MSG criteria, 2008 version

Possible / probable invasive aspergillosis
Host factors	Recent history of neutropenia (< 0.5 x 10 9 neutrophils/L [< 500 neutrophils/mm³] for > 10 days) temporally related to the onset of fungal disease. Receipt of an allogeneic stem cell transplant Prolonged use of corticosteroids (excluding among patients with allergic bronchopulmonary aspergillosis) at a mean minimum dose of 0.3 mg/kg/day of prednisone equivalent for > 3 weeks Treatment with other recognized T cell immunosuppressants, such as cyclosporine, TNF‐a blockers, specific monoclonal antibodies (such as alemtuzumab), or nucleoside analogues during the past 90 days Inherited severe immunodeficiency (such as chronic granulomatous disease or severe combined immunodeficiency
Clinical criteria	Lower respiratory tract fungal disease (every reasonable attempt should be made to exclude an alternative etiology), with the presence of 1 of the following 3 signs on CT Dense, well‐circumscribed lesions(s) with or without a halo sign Air‐crescent sign Cavity Tracheobronchitis: tracheobronchial ulceration, nodule, pseudomembrane, plaque, or eschar seen on bronchoscopic analysis Sinonasal infection: Imaging showing sinusitis plus at least 1 of the following 3 signs: Acute localized pain (including pain radiating to the eye) Nasal ulcer with black eschar Extension from the paranasal sinus across bony barriers, including into the orbit CNS infection: 1 of the following 2 signs: Focal lesions on imaging Meningeal enhancement on MRI
Mycological criteria	Direct test (cytology, direct microscopy, or culture) showing mold in sputum, bronchoalveolar lavage fluid, bronchial brush, or sinus aspirate samples, indicated by 1 of the following: Presence of fungal elements indicating a mold Recovery by culture of a mold (e.g., Aspergillus, Fusarium, Zygomycetes, orScedosporium species) Indirect tests (detection of antigen or cell‐wall constituents) Aspergillosis: Galactomannan antigen detected in plasma, serum, bronchoalveolar lavage fluid, or CSF Invasive fungal disease other than cryptococcosis and zygomycoses: b‐d‐glucan detected in serum
Proven aspergillosis
Criteria	Histopathologic, cytopathologic, or direct microscopic examination of a specimen obtained by needle aspiration or biopsy in which hyphae or melanized yeast‐like forms are seen accompanied by evidence of associated tissue damage; or, Recovery of a mold or “black yeast” by culture of a specimen obtained by a sterile procedure from a normally sterile and clinically or radiologically abnormal site consistent with an infectious disease process, excluding bronchoalveolar lavage fluid, a cranial sinus cavity specimen, and urine.

Table 3. Full EORTC/MSG criteria, 2008 version

Ir a la tabla de la revisión

Table 4. Full EORTC/MSG criteria, 2002 version

Possible / Probable invasive aspergillosis
Host factors	Neutropenia (< 500 neutrophils/mm³ for > 10 days) Persistent fever for > 96 h refractory to appropriate broad‐spectrum antibacterial treatment in high‐risk patients Body temperature either > 38°C or < 36°C and any of the following predisposing conditions: prolonged neutropenia (> 10 days) in previous 60 days, recent or current use of significant immunosuppressive agents in previous 30 days, proven or probable invasive fungal infection during previous episode of neutropenia, or coexistence of symptomatic AIDS Signs and symptoms indicating graft‐versus‐host disease, particularly severe (grade ≥ 2) or chronic extensive disease Prolonged (> 3 weeks) use of corticosteroids in previous 60 days
Clinical criteria	Must be related to site of microbiological criteria and temporally related to current episode Lower respiratory tract infection Major: Any of the following new infiltrates on CT imaging: halo sign, air‐crescent sign, or cavity within area of consolidation, in absence of infection by organisms that may lead to similar radiological findings including cavitation, such as Mycobacterium, Legionella, and Nocardia species. Minor: Symptoms of lower respiratory tract infection (cough, chest pain, haemoptysis, dyspnoea); physical finding of pleural rub; any new infiltrate not fulfilling major criterion; pleural effusion Sinonasal infection Major: Suggestive radiological evidence of invasive infection in sinuses (i.e., erosion of sinus walls or extension of infection to neighbouring structures, extensive skull base destruction) Minor: Upper respiratory symptoms (e.g. nasal discharge, stuffiness); nose ulceration or eschar of nasal mucosa or epistaxis; periorbital swelling; maxillary tenderness; black necrotic lesions or perforation of hard palate CNS infection Major: Radiological evidence suggesting CNS infection (e.g. mastoiditis or other parameningeal foci, extradural empyema, intraparenchymal brain or spinal cord mass lesion) Minor: Focal neurological symptoms and signs (including focal seizures, hemiparesis, and cranial nerve palsies); mental changes; meningeal irritation findings; abnormalities in CSF biochemistry and cell count (provided that CSF is negative for other pathogens by culture or microscopy and negative for malignant cell
Microbiological criteria	Positive result of culture for mold (including Aspergillus, Fusarium, or Scedosporium species or a Zygomycetes) or Cryptococcus neoformans or an endemic fungal pathogen from sputum or bronchoalveolar lavage fluid samples Positive result of culture or findings of cytologic/direct microscopic evaluation for mold from sinus aspirate specimen Positive findings of cytologic/direct microscopic evaluation for mold or Cryptococcus species from sputum or bronchoalveolar lavage fluid samples Positive result for Aspergillus antigen in specimens of bronchoalveolar lavage fluid, CSF, or ≥ 2 blood samples Positive result for cryptococcal antigen in blood sample Positive findings of cytologic or direct microscopic examination for fungal elements in sterile body fluid samples (e.g. Cryptococcus species in CSF) Positive result for Histoplasma capsulatum antigen in blood, urine, or CSF specimens Two positive results of culture of urine samples for yeasts in absence of urinary catheter Candida casts in urine in absence of urinary catheter Positive result of blood culture for Candida species
Proven aspergillosis
Criterion	Histopathologic or cytopathologic examination showing hyphae from needle aspiration or biopsy specimen with evidence of associated tissue damage (either microscopically or unequivocally by imaging); or positive culture result for a sample obtained by sterile procedure from normally sterile and clinically or radiologically abnormal site consistent with infection, excluding urine and mucous membranes.

Table 4. Full EORTC/MSG criteria, 2002 version

Ir a la tabla de la revisión

Table 5. Meta‐analysis of the pairs of sensitivity and specificity, sensitivity analysis

Meta‐analysis of the pairs of sensitivity and specificity, sensitivity analysis
	0.5 ODI (n = 12)	1.0 ODI (n = 11)
sensitivity	0.84 (95% CI 0.74 to 0.94)	0.80 (95% CI 0.69 to 0.92)
specificity	0.83 (95% CI 0.76 to 0.90)	0.92 (95% CI 0.89 to 0.96)
CI: confidence interval; ODI: optical density index

Table 5. Meta‐analysis of the pairs of sensitivity and specificity, sensitivity analysis

Ir a la tabla de la revisión

Table 6. Quantification of the incorporation bias

	with incorporation (n=12)	without incorporation (n = 4)
sensitivity	0.74 (95% CI 0.63 to 0.86)	0.98 (95% CI 0.95 to 1.00)
specificity	0.91 (95% CI 0.86 to 0.95)	0.72 (95% CI 0.53 to 0.90)

Table 6. Quantification of the incorporation bias

Ir a la tabla de la revisión

Table 7. Quantification of the effect of the study design

	prospective (n = 6)	retrospective (n = 10)
sensitivity	0.74 (95% CI 0.60 to 0.88)	0.94 (95% CI 0.86 to 1.00)
specificity	0.91 (95% CI 0.86 to 0.96)	0.78 (95% CI 0.65 to 0.91)
CI: confidence interval

Table 7. Quantification of the effect of the study design

Ir a la tabla de la revisión

Table 8. Sensitivity analyses on methodological quality

Sensitivity analyses on methodological quality Patients and flow and timing domains
	overall (all studies, n = 16)	low risk of bias (n = 6)
sensitivity	0.83 (95% CI 0.70 to 0.92)	0.75 (95% CI 0.51 to 0.89)
specificity	0.88 (95% CI 0.80 to 0.92)	0.86 (95% CI 0.78 to 0.91)
CI: confidence interval

Table 8. Sensitivity analyses on methodological quality

Ir a la tabla de la revisión

Table 9. Effect of reference standard definition

	Definition	Studies	Sensitivity	95% CI		Specificity	95% CI
	Definition	Studies	Sensitivity	Lower limit	Upper limit	Specificity	Lower limit	Upper limit
1	Prov and Prob vs Pos and No	16	0.83	0.70	0.92	0.88	0.80	0.92
2	Prov vs Prob and Pos and No	11	0.77	0.33	0.96	0.73	0.63	0.81
3	Prov and Prob and Pos vs No	11	0.59	0.40	0.75	0.94	0.89	0.97
4	Possibles excluded	11	0.79	0.66	0.88	0.94	0.89	0.97
CI: confidence interval; Prob: probable; Prov: proven

Table 9. Effect of reference standard definition

Ir a la tabla de la revisión

Table 10. Other systematic reviews

Systematic review	Number of studies	Number of studies also included in our review	Cut‐off ODI	Sensitivity (95% CI)	Specificity (95% CI)
Heng 2015	16	6	0.5	0.82 (0.70–0.91)	0.92 (0.85–0.96)
Heng 2015	16	6	1	0.75 (0.55–0.88)	0.95 (0.87–0.98)
Guo 2010	13	7	overall	0.90 (0.79‐0.96)	0.94 (0.90‐0.96)
			0.5	0.86 (0.70‐0.94)	0.89 (0.85‐0.92)
			1	0.85 (0.72‐0.93)	0.94 (0.89‐0.97)
Zou 2012	30	8	0.5	0.87 (0.79‐0.92)	0.89 (0.85‐0.92)
Zou 2012	30	8	1	0.86 (0.76‐0.92)	0.95 (0.91‐0.97)
CI: confidence interval; ODI: optical density index

Table 10. Other systematic reviews

Ir a la tabla de la revisión

Table Tests. Data tables by test

Test	No. of studies	No. of participants
1 ODI 0.5 Show forest plot	12	1123

2 ODI 0.8 Show forest plot	4	414

3 ODI 0.85 and 0.87 Show forest plot	4	286

4 ODI 1.0 Show forest plot	11	911

5 ODI 1.5 Show forest plot	3	242

6 ODI 2.0 Show forest plot	3	242

7 ODI 2.5 Show forest plot	3	242

Table Tests. Data tables by test

Ir a la tabla de la revisión

	Idioma de la Revisión Cochrane Escoja su idioma de preferencia para las revisiones Cochrane y otros contenidos. Las secciones sin una traducción aparecerán en inglés.

	Idioma de la web Escoja su idioma de preferencia para la web de la Biblioteca Cochrane.

Idioma de la Revisión Cochrane

Idioma de la web

Resumen

Antecedentes

Objetivos

Métodos de búsqueda

Criterios de selección

Obtención y análisis de los datos

Resultados principales

Conclusiones de los autores

Resumen en términos sencillos

Medición de galactomananos en el líquido de lavado pulmonar para el diagnóstico de la aspergilosis invasiva en pacientes con deterioro del sistema inmunitario

Resumen visual

Authors' conclusions

Implications for practice

Implications for research

Summary of findings

Background

Target condition being diagnosed

Index test(s)

Clinical pathway

Indication

Clinical consequence

Risk factors

Alternative test(s)

Rationale

Objectives

Secondary objectives

Methods

Criteria for considering studies for this review

Types of studies

Participants

Index tests

Target conditions

Reference standards

Search methods for identification of studies

Electronic searches

Searching other resources

Data collection and analysis

Selection of studies

Data extraction and management

Assessment of methodological quality

Statistical analysis and data synthesis

Investigations of heterogeneity

a. Effect of cut‐off value

b. Effect of incorporation bias (incorporation of the index test in the reference standard)

c. Effect of pre‐selection based on a positive serum galactomannan test

d. Effect of the reference standard

e. Effect of a protocol that defines the exact and explicit indication of the index test

f. Study design

g. Effect of sponsoring

h. Clinical subgroups

Sensitivity analyses

a. Methodological quality

b. Effect of positive and negative reference standard definition

Results

Results of the search

Methodological quality of included studies

Findings

Raw data

Meta‐analysis

Investigations of heterogeneity

Incorporation bias

The effect of pre‐selection based on a positive serum galactomannan test

The effect of the 2008 update of the reference standard

The effect of a protocol that defines the exact and explicit indication of the index test

Study design

The effect of sponsoring of the study by the manufacturer of the commercially available galactomannan test

Other planned investigations

Department(s) where the participants were recruited

Additional investigations that proved not possible

Sensitivity analysis

Methodological quality

Effect of reference standard definition

Discussion

Summary of main results

Comparison with other systematic reviews

Comparison with serum galactomannan and polymerase chain reactions (PCRs) on blood

Strengths and weaknesses of the review

Strengths