A Risk Prediction Score for Invasive Mold Disease in Patients with Hematological Malignancies

Background A risk score for invasive mold disease (IMD) in patients with hematological malignancies could facilitate patient screening and improve the targeted use of antifungal prophylaxis. Methods We retrospectively analyzed 1,709 hospital admissions of 840 patients with hematological malignancies (2005-2008) to collect data on 17 epidemiological and treatment-related risk factors for IMD. Multivariate regression was used to develop a weighted risk score based on independent risk factors associated with proven or probable IMD, which was prospectively validated during 1,746 hospital admissions of 855 patients from 2009-2012. Results Of the 17 candidate variables analyzed, 11 correlated with IMD by univariate analysis, but only 4 risk factors (neutropenia, lymphocytopenia or lymphocyte dysfunction in allogeneic hematopoietic stem cell transplant recipients, malignancy status, and prior IMD) were retained in the final multivariate model, resulting in a weighted risk score 0-13. A risk score of < 6 discriminated patients with low (< 1%) versus higher incidence rates (> 5%) of IMD, with a negative predictive value (NPV) of 0.99, (95% CI 0.98-0.99). During 2009-2012, patients with a calculated risk score at admission of < 6 had significantly lower 90-day incidence rates of IMD compared to patients with scores > 6 (0.9% vs. 10.6%, P <0.001). Conclusion An objective, weighted risk score for IMD can accurately discriminate patients with hematological malignancies at low risk for developing mold disease, and could possibly facilitate “screening-out” of low risk patients less likely to benefit from intensive diagnostic monitoring or mold-directed antifungal prophylaxis.


Introduction
Invasive mold diseases (IMDs) such as aspergillosis, and less commonly mucormycosis and fusariosis are a serious complication of myelosuppressive chemotherapy administered for hematological malignancies [1][2][3]. Patients undergoing allogeneic hematopoietic stem cell transplantation (HSCT) or remission-induction chemotherapy for acute myelogenous leukemia / myelodysplastic syndrome (AML/MDS) are at especially high risk, with 20-fold higher rates of aspergillosis compared to patients with underlying lymphoma or multiple myeloma [4]. Although diagnostic advances and new antifungal therapies have improved survival rates in patients with invasive aspergillosis [1], nearly one-third of patients still die with the infection, or have interruption of life-saving chemotherapy while the mold infection is being treated [1]. As a result, many hematologists routinely screen patients for incipient mold infection with the serum galactomannan test and high resolution computer tomography if the patient has fever, or administer mold-active antifungal prophylaxis for prolonged periods even though only a small proportion of patients (4-12%) may go on to develop a mold infection [5,6].
Risk stratification for IMD is a logical first step for identifying which patients would most likely benefit from more intensive monitoring or antifungal prophylaxis [7,8]. However, the development of an IMD risk prediction model in patients with hematological malignancies is complicated by the low overall disease prevalence, infrequently analyzed genetic risk factors related to host innate immunity, and dynamic clinical and environmental variables during their course of treatment [8,9]. Nevertheless, we hypothesized that an objective risk score for hematology patients based on easily documented demographic and clinical risk factors could have clinical utility if it accurately discriminates populations at low versus higher risk for developing IMD.
As a first step towards this goal, we retrospectively analyzed 17 candidate epidemiological and clinical risk factors for IMD in 840 patients during 2005-2008 to develop an objective risk score for proven or probable IMD. We then prospectively evaluated the performance of this risk score in 855 patients from 2009-2012. We found that a weighted risk score for IMD accurately discriminated a cohort of hematology patients at low ( < 1% incidence) versus higher (> 5% incidence) risk for mold infection, irrespective of the underlying malignancy, transplant status, and use of mold-active antifungal prophylaxis.

Ethics statement
The study was conducted in accordance with the Declaration of Helsinki, following review by the S'Orsola-Malpighi-University of Bologna ethics committee (http://www.aosp.bo.it/ content/comitato-etico). Full review was waived because of the non-interventional, observational nature of the study. As a standard protocol in our institute, all patients included in the study provided an informed consent the first day of hospitalization.

Study Population
This study was performed at a single regional hematology center in Italy (Institute of Hematology and Clinical Oncology "Lorenzo e Ariosto Seràgnoli", University of Bologna) during two periods. In the first study period (March 2005-December 2008), consecutive hospital admissions of patients with hematological malignancies were retrospectively analyzed for infections and IMD risk factors to develop a multivariate risk model for IMD. During the second study period (January 2009-December 2012), the performance of the risk score was prospectively analyzed in patients with a risk score calculated at the time of hospital admission, which was not reported to the treating hematologist. For each patient hospitalization, only the first infection episode was included in the analysis. Patients with hospitalizations shorter than 6 days were excluded from the analysis.
We collected data on 17 candidate predictors for IMD, which had been previously reported in the literature as risk factors for IMD in patients with hematological malignancies (Table 1). Additional data pertinent to each hospitalization and infection episode were collected from clinical records and registered on a standardized data collection form by the treating hematologist, while demographic data were extracted from an institutional centralized database. The accuracy of collected data was confirmed by a quality control procedure during data entry and with periodic reviews by 4 physicians (2 hematologists, 1 infectious diseases specialist, and 1 radiologist).

Study endpoint
The primary endpoint used for score development was documentation of proven or probable IMD within 90 days of hospitalization. Possible, probable or proven invasive aspergillosis (IA) and invasive mold disease (IMD) was defined according to the revised Mycoses Study Group and European Organization and Treatment of Cancer consensus criteria [10]. Serum galactomannan testing was routinely available at our institute after January 2007. Before this period, typical radiographic criteria as described by Cornely et al., were used to classify patients with proven or probable invasive aspergillosis [11]. In the case of non-Aspergillus molds not detected by galactomannan screening, diagnosis was always confirmed by histology or culture. Fluconazole (400 mg daily) was routinely administered to all patients undergoing allogeneic HSCT. Decisions regarding anti-mold antifungal prophylaxis in either non-transplant or transplant patients were at the discretion of the physicians caring for the patient.

Statistical Analysis
Demographic data were collected as either continuous data and compared by two-tailed unpaired t-test or Mann-Whitney test, or as categorical variables and compared by Chi-square test for patients with or without a probable or proven IMD. Variables with more than 5% missing data were excluded from analysis. Significant variables (P<0.05) were entered stepwise in a multivariate logistic regression model to evaluate the relationship between each variable and IMD risk using the Wald's statistic. Variables that maintained statistical significance by multivariate regression were then assigned a point value corresponding to the β-coefficient of that variable divided by the lowest β-coefficient of variables remaining in the regression model, and the resulting quotient was multiplied by two and rounded to the nearest whole number. Summation of the points resulted in a weighted risk score that was assigned to each patient episode retrospectively (2005)(2006)(2007)(2008), or prospectively (2009-2012) at the time of hospital admission.
The relationship of the calculated risk score and IMD risk was subsequently analyzed by receiver operator curves (ROC) to define an optimal cut-off score that discriminated low, versus high-risk patients. Our provisional cut-off was a risk score associated with 5% incidence of IMD, which has been proposed as the lower incidence limit of Aspergillus infection justifiable for antifungal prophylaxis in hematology patients [12]. All statistical analysis was performed using SPSS version 20 (IBM, Armonk, NY) and MedCalc 12.5 (Ostend, Belgium). of 2 separate hospitalizations to the database (range 1-12). The most common underlying malignancies were AML/MDS  [40,45] (31%), lymphoma (29%), and multiple myeloma/amyloidosis (25%) of which 63% were in partial or complete remission. Nearly 40% of the hospitalizations were for chemotherapy alone (i.e. no evidence of fever or infection on admission) with 46% of these admissions proceeding to HSCT (34% autologous, 12% allogeneic). Characteristics of the 1,709 cases are summarized in Table 2.
During the prospective score validation study period  Table 2). The increased use of anti-mold prophylaxis was associated with a corresponding decrease in empirical antifungal therapy for molds (8% vs. 14%, P<0.001). The most common anti-mold antifungal prophylaxis used during 2005-2008 was itraconazole capsules or solution (10%), which was largely replaced by posaconazole during 2009-2012 (11.4%) with some continued itraconazole use (8%). Voriconazole, lipid amphotericin B formulations, or aerosolized amphotericin B formulations were infrequently administered as prophylaxis during either study period (all less than 1%).

Risk Factors Associated with Proven or Probable IMD
Among the 17 candidate variables evaluated in the retrospective cohort, 11 were associated with IMD by univariate analysis (Table 3). These included patient occupational risk factors, the status of the underlying hematologic malignancy, variables related to the severity of underlying immunosuppression, a prior history of IMD, as well as the admission to a non-HEPA air-filtered room. However, in multivariate regression, only 4 of the 11 variables were independently associated with IMD risk: 1) Prolonged neutropenia, 2) lymphocytopenia or functional lymphocytopenia in allogeneic HSCT patients; 3) prior history of IMD, and 4) underlying malignancy that was not in partial or complete remission (Table 4). Points assigned on the basis of the weighted odds ratios for these 4 independent variables resulted in a risk score from 0-13 for each patient (mean 3.3, 95% CI 3.1-3.4) ( Table 4). Risk scores were well calibrated with observed rates of IMD ( Figure 1). When risk scores and the rates of true-positive and false-positive IMD rates were analyzed by ROC curves ( Figure  2), a score of less than 6 was found to be optimal cut-off for discriminating low-risk patients with an area under the ROC curve   (Figure 1). A score of less than 6 was also confirmed as the optimal cut-off for  . When the risk score performance was analyzed in different subgroups of hematological malignancy patients with varying IMD prevalence (1.5% to 10.6%) and rates of anti-mold prophylaxis use (7.2% to 57%), we found that a score of < 6 consistently identified a cohort of patients at low risk for IMD with NPVs ranging from 0.96-0.99 ( Table 5).

Impact of Posaconazole Prophylaxis
Posaconazole prophylaxis reduces the incidence of IMD in high-risk hematology patients and was associated with a mortality benefit in AML/MDS patients receiving remissioninduction chemotherapy [13,14]. We examined rates of proven or probable IMD among patients who had received posaconazole with risk scores of < 6 versus > 6 ( Figure 3a). Posaconazole prophylaxis was not associated with any discernable benefit in terms of reducing the incidence of IMD in patients with risk scores of < 6. However, among higher-risk patients with scores > 6, posaconazole prophylaxis was associated with a 7.8% risk reduction in IMD (P=0.01). We also found that among 131 individual patients with AML/MDS undergoing remission-induction chemotherapy during 2009-2012, patients who received posaconazole prophylaxis had a significantly lower risk of crude mortality within 6 weeks of hospitalization [15], versus patients who did not receive

Discussion
Physicians must weigh multiple factors when considering a patient's risk for developing IMD [16,17]. Prognostic models or risk scores can complement this clinical assessment by providing an objective summation of multiple risk factors, thereby clarifying which patients should be targeted for more aggressive intervention [18]. To our knowledge, our single-  institution study represents the first attempt to develop and validate an unconditional risk model for IMD in a heterogeneous population of patients with hematological malignancies. Our data demonstrate that an objective weighted risk score could reliably discriminate patients who had a very low probability of developing IMD within 90 days of hospitalization, and thus may be candidates for more conservative management with respect to higher-risk patients. An ideal risk score for IMD in patients would have both a high negative predictive value (NPV) and high positive predictive value (PPV). Yet, development of such a risk score that could be applied for routine screening of a heterogeneous population of hematology patients is challenging, given the overall low prevalence of IMD and fluctuating risk factors for infection [8]. Alternatively, a risk score could be developed in a more homogenous high-risk population of with a higher prevalence of IMD ( > 5%), such as allogeneic HSCT or AML/MDS patients undergoing remission-induction chemotherapy. Risk scores targeting populations who have already been shown to have proven to benefit from antifungal prophylaxis or intensive monitoring, however, may have less clinical utility for routine patient care [18] or considered too restrictive by treating physicians [19].
A limitation of our study is that our risk score was devised from observational data in a single center, and could not control for "real life" confounding factors such as use of antifungal prophylaxis. Notably, the discriminative performance of the risk score in our institution was similar among various subgroups of hematology malignancy patients with varying risk for IMD and usage patterns of antifungal prophylaxis. However, the performance of our risk score will undoubtedly vary in other hospitals depending on the type of patients treated and the baseline incidence of IMD. Additionally, clinical risk factors for IMD such as graft versus host disease and corticosteroids, which were not retained in our final risk model, would likely be more important if the score was developed specifically in allogeneic HSCT patients. Therefore multicenter validation and center-specific adjustments would likely be required if the risk score was applied to the clinical management of IMD in other hospitals.
In conclusion, we found than an objective, weighted riskscore for IMD could reliably discriminate the large majority of patients with hematological malignancies who were at low-risk for developing IMD. The discriminative performance of the score was consistent across various hematology patient subtypes with varying underlying baseline risk for IMD and exposure to antifungal prophylaxis. The continued refinement and multicenter validation of IMD risk scores could complement the clinical assessment of patients with hematological malignancies, and possibly improve the targeted use of