Pulmonary Aspergillosis in Patients with Suspected Ventilator-associated Pneumonia in UK ICUs

Abstract

Rationale: Aspergillus infection in patients with suspected ventilator-associated pneumonia remains uncharacterized because of the absence of a disease definition and limited access to sensitive diagnostic tests.

Objectives: To estimate the prevalence and outcomes of Aspergillus infection in adults with suspected ventilator-associated pneumonia.

Methods: Two prospective UK studies recruited 360 critically ill adults with new or worsening alveolar shadowing on chest X-ray and clinical/hematological parameters supporting suspected ventilator-associated pneumonia. Stored serum and BAL fluid were available from 194 nonneutropenic patients and underwent mycological testing. Patients were categorized as having probable Aspergillus infection using a definition comprising clinical, radiological, and mycological criteria. Mycological criteria included positive histology or microscopy, positive BAL fluid culture, galactomannan optical index of 1 or more in BAL fluid or 0.5 or more in serum.

Measurements and Main Results: Of 194 patients evaluated, 24 met the definition of probable Aspergillus infection, giving an estimated prevalence of 12.4% (95% confidence interval, 8.1–17.8). All 24 patients had positive galactomannan in serum (n = 4), BAL fluid (n = 16), or both (n = 4); three patients cultured Aspergillus sp. in BAL fluid. Patients with probable Aspergillus infection had a significantly longer median duration of critical care stay (25.5 vs. 15.5 d, P = 0.02). ICU mortality was numerically higher in this group, although this was not statistically significant (33.3% vs. 22.8%; P = 0.23).

Conclusions: The estimated prevalence for probable Aspergillus infection in this geographically dispersed multicenter UK cohort indicates that this condition should be considered when investigating patients with suspected ventilator-associated pneumonia, including patient groups not previously recognized to be at high risk of aspergillosis.

At a Glance Commentary

Scientific Knowledge on the Subject

Aspergillus infection is the most commonly missed infectious cause of death at autopsy in the ICU. There have been several studies estimating incidence and prevalence in critically ill patients, but they have used definitions of disease that are not tailored to this population and rely on the culture of Aspergillus species as a mycological criterion. Recent evidence has suggested that Aspergillus infection is likely to be underdiagnosed in this population, but to our knowledge, no study has assessed patients with suspected ventilator-associated pneumonia to estimate the burden of Aspergillus infection.

What This Study Adds to the Field

Clinically suspected ventilator-associated pneumonia is a commonly encountered clinical scenario and arises in a large patient population, globally. This multicenter study is the first robust estimate of the prevalence of Aspergillus infection in that patient group. We estimated its prevalence at 12.4%, which is higher than expected and builds on the growing body of evidence that Aspergillus infection may be underdiagnosed. This study points toward an opportunity to improve diagnostic pathways using nonculture tests for Aspergillus, such as galactomannan, to identify patients who may benefit from further investigation toward a diagnosis of aspergillosis.

Invasive pulmonary aspergillosis is a life-threatening disease classically affecting severely immunocompromised individuals, particularly those with prolonged neutropenia. Recently, nonneutropenic critically ill patients have been recognized as an at-risk population, although the nature of this risk remains uncertain (1). Reported estimates of incidence and prevalence vary substantially between studies of patients in ICUs. One single-center study estimated the rate of proven or probable invasive aspergillosis (IA) to be as high as 3.7% of all ICU patients without malignancy (2). A more recent large retrospective study reported a prevalence of 0.017% in all ICU admissions during a 3-year period, excluding those with severe immunosuppression, a well-established host factor (3). It has previously been estimated that between 821 and 9,665 cases of aspergillosis develop per year in UK ICU patients (4, 5). In a multicenter study of patients with influenza who were admitted to critical care, the prevalence of IA was 19%, with a 5% prevalence in patients with community-acquired pneumonia but not influenza (6).

Aspergillosis is consistently found to be among commonly missed diagnoses in critically ill patients, demonstrated by the results of multiple autopsy studies (7–9). Multiple factors contribute to delayed and missed diagnosis in this patient group, which create difficulties with estimating the burden of disease. First, the European Organization for Research and Treatment of Cancer/Invasive Fungal Infections Cooperative Group and the National Institute of Allergy and Infectious Diseases Mycoses Study Group definitions of IA are intended for use in the research setting and in severely immunocompromised patients (10). Consequently, when patients do not present with the severe immunosuppression described in this definition, physicians are less likely to consider the diagnosis. Second, the definition of probable disease relies on a combination of host factors, which are generally not applicable to the ICU population (11, 12).

A definition of aspergillosis based on nonculture tests without the requirement of host factors has the potential to identify more patients with this disease and allow disease prevalence to be assessed more reliably. The recently published American Thoracic Society clinical practice guidelines for the diagnosis of fungal infection in critical care highlight the importance of noninvasive diagnostic tests in this population, with strong recommendations given to both serum and BAL fluid (BALF) galactomannan (GM) testing and the detection of Aspergillus DNA by PCR. However, the guidance is aimed at severely immunocompromised patients, advising caution in their application to other potentially susceptible groups (13).

The AspICU algorithm was developed to guide the investigation of patients with Aspergillus cultures from respiratory tract specimens (14). Although useful, it relies on a positive culture as a starting point. Furthermore, international clinical guidelines do not address the investigation of patients with ventilator-associated pneumonia (VAP) toward a diagnosis of aspergillosis in the absence of substantial immunosuppression (15, 16). As a result, this condition may be underdiagnosed.

A recent critical care–based study by Schauwvlieghe and colleagues on IA postinfluenza (6) used a modified definition of disease; this did not rely on host factors and included GM as a mycological criterion for defining the presence of aspergillosis. The aim of the present study was to apply this principle to estimate the prevalence of Aspergillus infection in nonneutropenic ventilated patients with suspected pneumonia in the United Kingdom; this represents a common scenario in which the diagnosis of aspergillosis may be considered. Some of the results of this study have been previously reported in the form of abstracts (17, 18).

Methods

Study Design and Patient Recruitment

Participants with suspected VAP who had been recruited to two prospective multicenter UK studies (the VAP-Rapid studies) between February 2012 and September 2016 were reevaluated for the diagnosis of aspergillosis (19, 20). All participants had undergone standardized BAL and serum sampling and were eligible for inclusion in the present study if sufficient volumes of stored serum and BALF were available for mycological testing.

Eligibility Criteria

Criteria for clinically suspected VAP required for recruitment to the two studies (19, 20) were fulfilled when a patient

• 1.Had been intubated and mechanically ventilated for at least 48 hours and
• 2.Had new or worsening alveolar infiltrates on chest X-ray or computed tomographic (CT) scan and
• 3.Had two or more of the following:
  • •Temperature >38°C or <35°C
  • •White cell count >11 × 10⁹ or <4 × 10⁹/L
  • •Purulent tracheal secretions.

Patients with the following characteristics were excluded from these studies:

• 1.Recent history of neutropenia (<0.5 × 10⁹ neutrophils/L for >10 d).
• 2.Features considered to predict poor tolerance of BAL.

BALF underwent conventional culture for bacteria and fungi as part of the original study protocols. A panel of mycological tests was subsequently performed on paired BALF and serum samples for the present study. These were as follows:

• 1.GM enzyme immunoassay on BALF and serum samples was performed using the Platelia Aspergillus enzyme immunoassay (Bio-Rad Laboratories), according to the manufacturer’s instructions.
• 2.PCR amplification for detection of Aspergillus DNA on BALF and serum samples according to a method outlined previously and in line with the recommendations of the European Aspergillus PCR initiative (21). Each sample was tested in duplicate. Given the likely higher fungal burden in BALF, a positive result was defined by both replicates generating a positive signal at any cycle during the PCR. In serum, in which fungal burden is limited, at least one replicate generating a positive signal at any cycle was considered a positive result.
• 3.(1→3) β-d-glucan (BDG) detection was performed on serum samples as per the manufacturer’s instructions (Fungitell assay; Associates of Cape Cod). A threshold for positivity of 80 pg/ml was applied, which is consistent with previous literature (22).

Definition of Aspergillus Infection

To overcome previously encountered problems arising from the inaccessibility of tissue samples, low sensitivity of culture, poor applicability of traditional host factors, and absence of typical clinical and radiological findings, a definition of disease that is not reliant on these was adopted. Patients were categorized as having probable Aspergillus infection based on the principles used in a recent ICU study (6), comprising clinical, radiological, and mycological criteria. All three criteria were required for the definition of probable Aspergillus infection to be met.

Clinical Criteria

At least two of the following signs or symptoms had to be present:

• •Temperature <35°C or >38°C
• •White cell count <4 × 10⁹/L or >11 × 10⁹/L
• •Purulent tracheal secretions.

Radiological Criteria

Radiological criteria included any new or progressive infiltrate on pulmonary imaging by chest X-ray or CT scan of the lungs.

Mycological Criteria

One or more of the following had to be present:

• •Histopathology or direct microscopic evidence of dichotomous septate hyphae with positive culture for Aspergillus from tissue.
• •Positive Aspergillus culture from BALF.
• •GM optical density (OD) index in BALF of ≥1.
• •GM OD index in serum of ≥0.5.

Risk Factors and Clinical Outcomes

When available from the databases of the two studies that patients had been originally recruited to, possible risk factors for aspergillosis were noted, including age, Acute Physiology and Chronic Evaluation (APACHE) II score, need for renal replacement therapy, corticosteroid exposure, duration of ICU stay before enrollment, and concurrent diagnosis of influenza. BALF specimens from patients categorized as probable aspergillosis also underwent PCR for the detection of influenza A and B RNA. Presence of microbiologically confirmed bacterial VAP, length of ICU stay after enrollment, and ICU mortality were the outcome measures used.

Secondary Analyses

The prevalence of Aspergillus infection was determined using the same disease definition but applying a range of BALF GM OD index threshold values from 0.7 to 3.0. A further secondary analysis was undertaken in which more than one positive Aspergillus biomarker was required to meet the disease definition.

Statistical Analysis

Categorical variables were compared by χ² test, and continuous variables were compared with Student’s t test or the Mann-Whitney U test, as appropriate.

Statistical tests were performed using SPSS version 25 (IBM) with the significance level set at 5%.

Results

One hundred ninety-four patients with paired samples from 17 UK centers were eligible for inclusion (Figure 1). The characteristics of the cohort are summarized in Table 1. The median age of participants was 57 years (interquartile range [IQR], 44–69), and the mean APACHE II score on admission was 17.93, with men making up 70.6% of the cohort.

Figure 1.

Study flow diagram. VAP = ventilator-associated pneumonia.

Table 1.

Characteristics and Outcomes of Suspected VAP Patient Cohort

Characteristics	Suspected VAP Cohort (n = 194)	With Probable Aspergillus Infection (n = 24)	Without Probable Aspergillus Infection (n = 170)	P Value
Age, median (IQR), yr	57 (44–69)	66.5 (49.8-72.5)	56 (43–69)	0.07
Sex, M, n (%)	137 (70.6)	15 (62.5)	122 (71.8)	0.35
APACHE II score on admission, mean (SD)	17.93 (7.5)	19.25 (7.5)	17.74 (7.7)	0.36
Medical reason for admission, n (%)	113 (58.2)	17 (70.9)	96 (56.5)	0.18
Surgical reason for admission, n (%)	81 (41.8)	7 (29.1)	74 (43.5)	0.18
Preenrollment length of stay, median (IQR), d	7 (4–11)	7.5 (7–12)	6 (4–10.75)	0.19
Steroids, n (%)	33 (16.9)	6 (25)	27 (15.8)	0.27
WCC on day of BAL, n	15.3	14.95	15.3	0.83
Renal replacement therapy, n (%)	16 (8.3)	3 (12.5)	13 (7.7)	0.42
Vasopressors, n (%)	61 (31.4)	8 (33.3)	53 (31.2)	0.83
Microbiologically confirmed VAP, n (%)	78 (40.2)	9 (37.5)	69 (40.6)	0.77
Length of stay in critical care, median (IQR), d	17 (11–31.5)	25.5 (17.25–32.8)	15 (10–30.5)	0.02
Length of stay in hospital, median (IQR), d	34 (17–62)	34 (24.5-61)	34 (14.25–62)	0.39
ICU mortality, n (%)	47 (24.1)	8 (33.3)	39 (22.8)	0.27

Definition of abbreviations: APACHE II = Acute Physiology and Chronic Evaluation II; IQR = interquartile range; VAP = ventilator-associated pneumonia; WCC = white cell count.

Prevalence Estimate for Probable Aspergillus Infection

All 194 patients met the clinical and radiological criteria because of the eligibility criteria for the original two studies. The mycological criteria were met by 24 patients, giving an estimated prevalence of Aspergillus infection of 12.4% (95% confidence interval [CI], 8.1–17.8). GM was positive in BALF alone in 16 patients, was positive in serum alone in 4 patients, and was positive in both serum and BALF samples in a further 4 patients (Table 2). Three of the 24 patients also cultured Aspergillus sp. in BALF.

Table 2.

Biomarker and Culture Results of Serum and BALF for Patients Who Met the Definition of Probable Aspergillus Infection

Patient	BALF GM	Serum GM	Serum BDG	Serum PCR	BALF PCR	BALF Culture >104 cfu/ml
1	14.67	1.22	+	+	+	Aspergillus sp.*
2	13.95	0.21	+	−	+	Acinetobacter sp.
3	11.59	0.50	+	−	+	A. fumigatus*
4	8.46	0.04	−	−	+	M. catarrhalis
5	8.04	0.11	+	+	+	A. fumigatus*
6	4.66	0.08	−	−	−	—
7	3.32	0.12	−	−	−	E. coli
8	3.16	0.07	−	−	−	K. pneumoniae
9	2.40	0.08	+	−	−	P. mirabilis
10	2.02	0.03	−	−	−	—
11	1.65	0.05	−	−	−	P. aeruginosa
12	1.52	0.05	−	−	−	S. aureus
13	1.47	0.17	+	−	−	—
14	1.47	0.08	−	−	−	—
15	1.44	0.08	−	−	−	—
16	1.41	0.05	−	+	−	—
17	1.36	0.18	−	−	−	—
18	1.24	0.52	+	−	+	—
19	1.05	0.21	−	+	+	—
20	1.03	0.51	+	+	−	—
21	0.14	2.65	+	−	−	E. cloacae
22	0.32	1.54	−	−	+	—
23	0.43	0.79	−	−	+	S. aureus, S. marcescens
24	0.92	0.50	+	−	−	—

Definition of abbreviations: A. fumigatus = Aspergillus fumigatus; BALF = BAL fluid; BDG = β-d-glucan; E. cloacae = Enterobacter cloacae; E. coli = Escherichia coli; GM = galactomannan; K. pneumoniae = Klebsiella pneumoniae; M. catarrhalis = Moraxella catarrhalis; P. aeruginosa = Pseudomonas aeruginosa; P. mirabilis = Proteus mirabilis; S. aureus = Staphylococcus aureus; S. marcescens = Serratia marcescens.

^*Any growth of Aspergillus sp. reported (no minimum cfu/ml applied).

The median BALF GM OD index was 1.84 (IQR, 1.43–5.51) in patients with aspergillosis and 0.12 (IQR, 0.07–0.25) in those without aspergillosis (P < 0.0001).

The median serum GM OD index was 0.15 (IQR, 0.08–0.51) in those with aspergillosis and 0.07 (IQR, 0.05–0.09) in those without aspergillosis (P = 0.001).

Prevalence Estimates Using Different BALF GM OD Index Threshold Values

A range of BALF GM OD index threshold values were used to assess the prevalence of aspergillosis (Table 3) in addition to the value of 1.0 used in the main analysis. Using a more stringent BALF GM OD index cutoff value of 1.5, six fewer patients would be classified as having probable Aspergillus infection, giving a prevalence of 9.3% (95% CI, 5.6–14.3%).

Table 3.

Prevalence of Aspergillus Infection at a Range of GM OD Index Threshold Values

BALF GM Threshold OD	Number of Patients with BALF Positive	Total Number of Aspergillosis Cases (BALF or Serum Positive)	Prevalence [% (95% CI)]
0.7	27	30	15.5 (10.7–21.3)
0.8	25	28	14.4 (9.8–20.2)
1.0	20	24	12.4 (8.1–17.8)
1.5	12	18	9.3 (5.6–14.3)
3.0	8	14	7.2 (4.0–11.8)

Definition of abbreviations: BALF = BAL fluid; CI = confidence interval; GM = galactomannan; OD = optical density.

Other Biomarkers

An additional biomarker (BDG in serum or PCR in either BALF or serum) was positive in 15 of the 24 patients with a positive GM in either serum or BALF (Table 2). If the disease definition required more than one positive biomarker, the estimated prevalence would be 7.8% (95% CI, 4.74–12.4%). Ten of the 24 patients with a positive GM test in either specimen type had a positive BDG, 10 had a positive Aspergillus PCR in BALF, and 5 had a positive Aspergillus PCR in serum. Of the three patients whose BALF cultured Aspergillus species, one was positive for all biomarkers; the other two patients were positive for four of the five biomarkers. All patients with a positive serum GM had at least one other positive Aspergillus biomarker.

Possible Risk Factors for Aspergillosis

There were no statistically significant differences in age or APACHE II score between the groups with and without probable Aspergillus infection, although the age of patients in the probable aspergillosis group was numerically higher (Table 1). The median duration of ICU stay before enrollment was numerically higher (7.5 vs. 6.0 d) in the probable aspergillosis group but was also not statistically significant (P = 0.19). No patients had documented influenza infection at the time of enrollment in the original studies. Of the 24 patients categorized as having probable aspergillosis in the main analysis, none had detectable influenza A or B RNA in BALF.

The use of steroids before bronchoscopy appeared to be higher in patients who had probable Aspergillus infection (25% vs. 15.8%), and renal replacement therapy also appeared to be used more frequently in this group (12.5% vs. 7.5%); neither of these differences were statistically significant. White cell count and the use of vasopressors were similar between groups.

Clinical Outcomes

The proportion of patients with bacteria cultured in significant numbers (>10⁴ cfu/ml) from BALF was similar in those with and without Aspergillus infection (Table 1).

The median length of critical care stay was significantly longer in those with probable Aspergillus infection compared with those who did not have Aspergillus infection (25.5 vs. 15 d; P = 0.02). ICU mortality was numerically higher in the group with probable Aspergillus infection (33.3% vs. 22.8%; P = 0.27); however, this did not reach significance with statistical testing. Only one patient classified as having probable aspergillosis received mold-active antifungal treatment during their ICU episode. Data regarding treatment after ICU discharge were not available.

Discussion

This study represents the first robust estimate of the prevalence of Aspergillus infection in a population of nonneutropenic, ventilated, critically ill patients with suspected pneumonia. The 12.4% prevalence estimate is higher than expected based on previous literature.

We acknowledge uncertainty in this estimate arising from the definition of probable aspergillosis; however, we address this uncertainty by presenting a range of prevalence data using more stringent disease definitions. We considered 1) the lower bound of the CI for the main analysis, 2) the effect of using a more stringent cutoff value of 1.5 for BALF GM OD, and 3) requiring more than one biomarker to be positive for the disease definition to be met. In these scenarios, the range of prevalence estimates obtained is 7.8–9.3%. Although lower than the prevalence estimated in our main analysis, this more conservative range of estimates may indicate that a genuine burden of illness exists in these patients. This builds on the work by Schauwvlieghe and colleagues, in which 5% of nonimmunocompromized critically ill patients with community-acquired pneumonia had IA (6). We provide further evidence that, in the ICU population, there is potential for underdiagnosis in clinical practice.

One of the reasons for potential underdiagnosis is the lack of a validated definition of aspergillosis in this population. Such a definition is difficult to produce because of insufficient clinical evidence. The AspICU algorithm, primarily intended to guide management of patients with a positive lower respiratory tract culture for Aspergillus (14), does not enhance the detection of disease because of the low sensitivity of culture (23). GM detection has been suggested as an alternative entry route to better capture patients with aspergillosis (24).

It is acknowledged that the disease definition we used is not perfect; for example, there is debate about the most appropriate threshold value to use for GM positivity. D’Haese and colleagues (25) demonstrated that a threshold value of 1.0 for the OD index in BALF offers a 93.8% specificity for IA. Zhou and colleagues (26) demonstrated 0.7 as an optimal threshold OD index in nonneutropenic patients. Because there is uncertainty regarding the optimal threshold value for BALF GM in this heterogeneous population of nonneutropenic critically ill patients, a secondary analysis used a range of BALF GM OD index threshold values. This demonstrated that even when more stringent BALF GM threshold values, offering higher specificity, are applied (e.g., 1.5 OD, with 95% specificity) the estimated prevalence remained higher than would be expected at 9.3%. Using an extremely stringent OD threshold of 3.0 for BALF GM and excluding serum GM from the disease definition would give a prevalence estimate of 4.1%. This might be considered the “minimum” estimate because a threshold of 3.0 for the BALF GM index has been reported to have specificity of 100%; however, with sensitivity that may be as low as 56%, this approach is likely to yield an underestimate (25). It is clear that the true prevalence of aspergillosis in this population remains subject to uncertainty given the variation seen across a range of plausible BALF GM threshold values.

Compounding this uncertainty is the potential effect of disease prevalence on the false-positive rate even in a test with such high specificity as GM. Consequently, our disease definition may overestimate prevalence, and, similarly, the use of GM alone may lead to overdiagnosis and overtreatment in clinical practice. Collectively, these issues underscore the importance of not relying on a single diagnostic test as a trigger for starting antifungal therapy in the context of suspected VAP. However, we propose that a high level of GM in BALF has the potential to identify patients who should undergo more extensive evaluation toward the diagnosis of aspergillosis, even in patient groups that are not classically regarded as being at high risk. Such further evaluation is likely to be multimodal, because no single alternative test (direct microscopy, culture, serum GM or β-d-glucan, Aspergillus PCR, or CT scan) is more sensitive or specific than BALF GM.

Other biomarkers (serum β-d-glucan and serum and BALF PCR for Aspergillus DNA) did not form part of the disease definition in this study; however, they provided useful corroboration of our results because 15 of the 24 probable cases had other positive biomarkers. These could be useful in a panel of rapid, non–culture-based tests but require further assessment of diagnostic accuracy in the ICU population to establish their utility.

This study suggests that the detection of GM in BALF appeared to be more sensitive than in serum for the detection of probable Aspergillus infection in this population. Twenty (88.3%) of the 24 patients with probable infection had a positive BALF result, whereas only 8 (33.3%) had a positive serum result. This is in keeping with previous studies, in which BALF GM sensitivity has been demonstrated to be significantly higher than culture or serum GM in this patient population (24, 26). In nonneutropenic patients, several studies have found that serum GM lacks sufficient sensitivity when used alone (26–28). Our data support this view because not all patients with the highest values for BALF GM had a positive serum result. Although serum GM lacks sensitivity, all eight patients in our dataset who had a positive serum GM had at least one other positive biomarker. This suggests that, although not a sensitive test, it may still have an important role in a diagnostic panel because it seems likely to represent established disease.

We note with interest that patients categorized as having probable Aspergillus infection had a significantly greater length of stay in critical care after enrollment. This group also had mortality that was numerically higher (though not statistically significant). Whether these findings represent an association with increased risk of aspergillosis or are a consequence of other factors cannot be ascertained within the present study design.

The same is true of risk factors that were noted more generally. The risk factor data are incomplete, and hence subject to bias, because the patients had been originally recruited to studies that focused on bacterial VAP; as a result, data collected were not ideally tailored toward risk factors for aspergillosis. The use of steroids and renal replacement therapy, which are previously identified risk factors for aspergillosis (29), were not significantly higher in the probable aspergillosis group; however, insufficient details of the intensity, timing, and duration of exposure limit inference from these data. Moreover, because the epidemiology of aspergillosis in this patient group has not been well characterized, our understanding of risk factors is incomplete; this, together with the missing details referred to, may lead to futile risk factor analysis in the present study.

Because of recent attention that influenza has received as a possible risk factor, it is interesting that none of our probable aspergillosis cases had detectable influenza RNA in BALF. Nonetheless, this finding does not necessarily exclude the possibility of antecedent influenza infection.

Only one of the patients had treatment active against aspergillosis during their ICU admission. An incomplete picture of whether other patients received such treatment subsequently during their hospital stay makes this finding difficult to interpret; for that reason, we cannot infer that it materially undermines the definition we used. Likewise, the coexistence of a significant bacterial pathogen in BALF from several patients classified as having probable Aspergillus infection is difficult to contextualize. There is no evidence to indicate whether coinfection is likely in such patients, and it is not possible to determine whether either organism had a dominant role in the pathogenesis of such patients’ pneumonia.

Limitations of our study design included the sample size, which was limited by the size of the VAP-Rapid studies, and the retrospective analysis of prospectively collected data, which may introduce bias from incomplete information. More complete data on host factors, such as chronic obstructive pulmonary disease, liver cirrhosis, and immunosuppressant drugs, as well as details of clinical response to antibiotic treatment may have strengthened the analysis. Incomplete risk factor, antifungal treatment, and long-term outcome data were such that we could not triangulate these findings to provide additional support for the disease definition used. Because this placed additional reliance on GM to classify patients, we considered the effect of applying different GM index thresholds and requiring more than one positive Aspergillus biomarker in the disease definition. Sample storage and retrospective testing may have influenced the performance of nonculture tests. Furthermore, as discussed earlier, the risk of error from diagnostic misclassification arises from the lack of a robust definition of aspergillosis in this patient group.

Strengths of the study included the number of sites participants were recruited from and their geographic spread. In addition, the study addresses the critically ill cohort, which is a patient group of interest with broad applicability in health care that has been understudied to date in the context of aspergillosis. Standardized testing was performed on all patient samples, leading to a complete laboratory data set, strengthening the analysis. Another strength is that the definition used overcomes the weaknesses of previous definitions with its lack of reliance on host factors or nonapplicable clinical and radiological findings.

In conclusion, we present a range of estimates for the burden of aspergillosis in this multicenter UK study of nonneutropenic critically ill patients that highlights the potential for underdiagnosis in clinical practice. We suggest that these data demonstrate the need for increased awareness among clinicians. More widespread use of GM when patients with suspected VAP undergo BAL may provide a means to identify patients who might benefit from extensive clinical investigation to seek a diagnosis of aspergillosis.