Invasive aspergillosis (IA) is a life-threatening disease in hematology patients, especially in those with deep and prolonged neutropenia following myelotoxic chemotherapy and in those receiving immunosuppressive therapy for treatment of graft-versus-host disease after allogeneic hematopoietic cell transplantation [1
]. Early diagnosis remains difficult, especially when relying on conventional tools such as microscopic examination and fungal culture, resulting in delayed treatment and significant morbidity and mortality [2
Galactomannan (GM) is a cell wall polysaccharide of Aspergillus species (and some other fungi), which is released by growing hyphae during tissue invasion. Detection of circulating GM by a commercial sandwich-enzyme immunosorbent assay has become an important tool for the diagnosis of IA. This is underscored by the inclusion of GM as a microbiological criterion in the European Organization for Research and Treatment of Cancer/Mycoses Study Group (EORTC-MSG) consensus definitions for invasive fungal diseases [4
]. The diagnostic performance and clinical utility of GM measurements has been studied in different clinical settings and has been the subject of several meta-analyses [5
]. Nowadays, hematologists use this assay primarily as a diagnostic test (e.g., on broncho-alveolar lavage (BAL) fluid in patients with unexplained pulmonary lesions) or, in neutropenic patients, as a screening tool on serum samples.
Despite some recent advances in diagnosing the infection, the early assessment of response to antifungal therapy remains problematic. Historically, a composite endpoint of clinical, radiological, and microbiological outcomes has been used [9
]. However, clinical assessments are often subjective and based on non-specific signs and symptoms. Even the presumed objective assessments have significant limitations: on follow-up imaging studies, the initial fungal lesions often increase in size during the first (two) weeks; although this may indicate a poor treatment outcome, it more often reflects the expected course of the infection or it may be due to immune reconstitution following neutrophil recovery or tapering of the immunosuppression [10
]. In addition, serial microbiologic assessments are impractical if an invasive procedure such as repeated bronchoscopic lavage and/or biopsy is required. Therefore, a biomarker that can predict clinical outcome, particularly early after starting antifungal therapy (e.g., after 1 week), would be a valuable tool for clinicians to guide their antifungal treatment decisions.
Galactomannan could be such a biomarker. Indeed, animal studies have consistently demonstrated a strong correlation between baseline serum galactomannan (sGM) levels and sGM kinetics and biological outcome variables, such as quantitative cultures (fungal burden), organism-mediated pulmonary injury (measured by lung weight, infarct scores, and CT scan scores), survival, and treatment response [11
]. Therefore, a decrease in sGM correlates with a decreased fungal burden. However, demonstrating a similar strong correlation in patients appears more challenging. Besides, clearance of sGM occurs through several mechanisms, including elimination via neutrophils, hepatic uptake through macrophages (Kuppfer cells), and renal clearance [14
]. Furthermore, the class-specific mechanism of action of the different antifungal drugs could also alter the release of GM (e.g., echinocandins induce cell wall lysis), whereas all mold active agents interfere with the test performance [8
]. This demonstrates a complex interplay between sGM levels, fungal burden, antifungal therapy, and several host factors such as neutropenia and comorbidities. It is therefore essential to study the in vivo kinetics of sGM and its significance after initiation of antifungal therapy in a real-life patient population.
In this study, we assessed the correlation between sGM levels during the first week of antifungal treatment and clinical outcomes (survival and response) through day 42 (week 6) in patients with documented IA. We further created an easy-to-use prediction rule for week 6 mortality based on sGM values determined at baseline and after one week of anti-Aspergillus therapy.
We created a simple prediction rule for week 6 mortality in hematology patients with IA based on the baseline and week 1 sGM levels. It predicts that patients that have a sGM ODI > 1.4 and that fail to attain a negative sGM (ODI < 0.5) after one week will have a high mortality. Conversely, patients that have a low sGM (ODI ≤ 1.4) at diagnosis and are negative (ODI < 0.5) after one week will have a low mortality. This rule is easy to use at the bedside, without the need for multiple covariates or computer models, while still predicting a doubling in mortality after one week after treatment based on a single blood test. Visual analysis of Figure 2
shows where this model finds its basis: patients that survive by week 6 generally start out with a low sGM index at baseline and remain low by week 1. Patients that have died by week 6 started out with a high baseline sGM index on average and failed to decrease this index by week 1. This model appears to be robust and not the result of overfitting, as evidenced by the validation in an independent patient population, as well as from the multivariate regression. This is further exemplified by Table 3
, which shows no difference in baseline characteristics between the three groups, with the exception of EORTC/MSG classification. There were statistically less cases of proven IA in the low risk group, likely due to survivorship bias: the diagnosis of proven IA was often only made at autopsy. A second significant difference between the subgroups was in the number of patients that fulfills the 2019 revision of the EORTC/MSG criteria [22
]. Indeed, this study was designed and performed before these criteria were published. A post hoc analysis revealed that a significant number, though not all patients would still be classified as having probable IA. This number was significantly higher in the high risk subgroup (100%) as by definition, patients need to have a baseline sGM > 1.4 to be in this subgroup, fulfilling the new mycological criteria. This was not obligatory in the low and intermediate risk groups. In the subgroup of only patients fulfilling the new criteria, the risk stratification remained significant: survival was 53% in the high risk group, 66% in the intermediate group, and 93% in the low risk group (p
< 0.001, n
The robustness of our prediction rule is further supported by previous, independent studies that found similar cut-offs and correlations to the ones defined by our mathematical model [23
]. The outcome of our prediction rule could therefore help clinicians in deciding if treatment is successful early on, after only one week of treatment. If treatment failure is anticipated, possible causes should be explored, including drug failure (e.g., resistance to the antifungal agent or inadequate exposure) and/or clinical failure due to ongoing defects in the host immunity.
An interesting effect can be seen in the intermediate risk group. Instinctively, an increasing sGM after 1 week despite therapy would suggest therapy failure and might appear worse than a baseline sGM ≤ 1.4 that fails to become negative after one week. However, it appears that a very high initial serum GM remains an important predictor of outcome, even if this then drops quickly, with an effect of roughly the same size as a rising serum GM after initial low GM.
Previous studies have shown a correlation between sGM ODI at baseline, kinetics, and outcome [23
]. However, we could identify only two studies that proposed a prediction rule based on sGM ODI kinetics at 1 week [24
]. Kovanda et al. found that an increase of sGM ODI by day 7 of >0.25 was associated with a 10-fold increase in risk of death, when compared to smaller increases, stable results, or declining sGM ODI [25
]. However, as this study only included patients that were sGM positive at baseline, this rule is not applicable in patients that have a negative sGM at baseline, but become positive by week 1 (which would be classified as intermediate risk in our model). Chai et al. found that in patients who were sGM positive at baseline, a decline in sGM index of >35% by week 1 was correlated with good outcome [24
]. However, this study only reported a correlation with week 12 clinical outcome, whereas we applied it to week 6 survival. Neither study used an independent cohort to validate their findings. We therefore applied these rules to our training cohort, but could not confirm their results: the 6 week mortality was not significantly different in the high risk group from the low risk group in the training cohort (p
= 0.200 for the rule by Chai et al., and p
= 0.170 for the rule by Kovanda et al.) or in the external validation cohort (p
= 0.570 for the rule by Chai et al., and p
= 0.053 for the rule by Kovanda et al.). One possible explanation for this discrepancy with the results by Kovanda et al. is that in that study sGM values were not tested per protocol at 1 week. This resulted in a large number of missing values, which were then interpolated from baseline and week 2 levels. However, it is not known what the actual kinetics are in this time interval, which could lead to overinterpretation of interpolated values.
Our prediction rule was mainly evaluated in neutropenic hematology patients. Therefore, generalization of our results to other at-risk populations, including solid organ transplant recipients, should be done cautiously, as there are differences in host response to the infecting mold, and due to differences in clearance of sGM by the immune system, which ultimately results in lower levels of sGM overall [5
]. Creating a general model of sGM kinetics remains difficult due to high variability in metabolism given the different elimination pathways [26
]. Furthermore, creating a perfect prediction rule based on a single biomarker remains a challenge, as other factors, such as the underlying disease, influence the outcome at week 6 in this highly comorbid population. As such, our model is not meant to eliminate all other factors that guide clinical decision making or replace clinical expertise from an experienced physician. In fact, we aim to offer the clinician an additional, robust and simple to use tool to be used in conjunction with all other data available. For example, imaging can be deceiving in the early course of therapy [10
]. As such, our risk stratification could help in cases where radiology is stable or worsening, yet the patient is improving clinically.
The generalizability of our study will also depend on the exact anti-fungal approach used, as our study was based on data from a population in which sGM screening was used in patients at the highest risk for IA, without antifungal prophylaxis. In case prophylaxis is used, we expect baseline GM to be lower in general. This would mean that if a patient attains sGM > 1.4 despite prophylaxis, mortality will likely be significantly higher. In such populations, the cut-offs would likely have to be lower. In populations where screening is not used, but rather a symptom-based approach, we would expect higher initial serum GM as screening for GM can detect IA before the onset of symptoms [27
]. However, to the best of our knowledge, there are no interventional trials comparing treatment outcomes in patients undergoing screening compared to patients managed using a symptom-driven approach.
A major advantage of our study is the fact that we had access to more than 9500 daily sGM values after initiation of therapy, did not rely on modeled or interpolated values, and included only probable and proven cases, in contrast to previous studies. Another strength of our model is the use of discrete cut-off points (as opposed to a continuous risk estimate), which eliminates any possible interference of inter-testing variability which can be seen at higher sGM values due to the influence of the nonlinear range inherent to photometry at higher densities [23
]. Furthermore, our prediction rule is not limited to patients who have a positive sGM at baseline, as our model was trained using data from patients who had positive GM in serum or BAL. This improves the generalizability of our model and allows its use within the complete hematologic population with IA. This is also supported by previous studies showing improved survival in patients with negative sGM at baseline [28