Abstract
Background
Invasive pulmonary aspergillosis (IPA) increasingly affects non-neutropenic patients, posing diagnostic challenges due to nonspecific clinical features and limited sensitivity of conventional tests. Early identification is critical for improving outcomes. This study aimed to develop and validate a machine learning-based tool for early risk stratification of IPA in this population.
Methods
A systematic meta-analysis identified independent IPA risk factors in non-neutropenic patients. A retrospective cohort of non-neutropenic patients was analyzed. Least absolute shrinkage and selection operator (LASSO) regression selected predictive features, and multiple machine learning (ML) algorithms were evaluated. The optimal model (random forest) was externally validated on an independent cohort. SHapley Additive exPlanations (SHAP) values interpreted feature importance.
Results
A retrospective cohort of 524 hospitalized non-neutropenic patients (422 training, 102 validation) adhering to established diagnostic criteria was analyzed. Meta-analysis confirmed diabetes [odds ratio (OR) =1.43], respiratory disease (OR =1.76), corticosteroid exposure (OR =1.48), and smoking history (OR =1.64) as key risk factors. The random forest model incorporated nine predictors (including antibiotic use, viral pneumonia, intensive care unit (ICU) admission, low protein levels, and bacterial infections) and achieved high accuracy [area under the curve (AUC) =0.950, sensitivity =0.857, specificity =0.905]. External validation maintained robust discrimination (AUC =0.856; sensitivity and specificity =0.733). SHAP analysis revealed critical synergistic interactions, particularly between prolonged antibiotic use and viral pneumonia in elevating IPA risk.
Conclusions
This validated ML tool integrates meta-evidence with clinical data for early, accurate IPA risk stratification in immunocompetent patients. It effectively captures nonlinear predictor interactions and demonstrates strong generalizability, supporting clinical utility.
Keywords: Machine learning (ML), invasive pulmonary aspergillosis (IPA), immunocompetent hosts
Highlight box.
Key findings
• Meta-analysis (13 studies) identified four independent baseline risk factors for non-neutropenic invasive pulmonary aspergillosis (IPA). A nine-variable random-forest model delivered area under the curve (AUC) 0.95 (sensitivity 86%, specificity 91%) in 422 training cases and AUC 0.86 in an external 102-patient cohort, outperforming six alternative machine learning (ML) algorithms. SHapley Additive exPlanations (SHAP) visualisation revealed antibiotic-viral pneumonia interaction as the dominant driver of risk.
What is known and what is new?
• IPA incidence is rising in non-neutropenic, immunocompetent hosts, but validated prediction tools are lacking. Existing models are single-centre, logistic-regression-based and poorly generalisable.
• We integrate meta-evidence with ML, provide external validation and offer an interpretable, ready-to-use calculator.
What is the implication, and what should change now?
• This tool enables early, accurate risk stratification for IPA in immunocompetent patients, facilitating timely pre-emptive therapy and improving outcomes.
• The findings call for heightened clinical vigilance regarding IPA risk in patients with viral pneumonia, especially those receiving prolonged antibiotic courses, supporting more judicious antimicrobial stewardship.
• Implementation of this validated model in intensive care unit (ICU) and pulmonology settings could standardize and enhance proactive screening strategies for at-risk, non-neutropenic patients.
Introduction
Invasive pulmonary aspergillosis (IPA) is an opportunistic fungal infection caused by Aspergillus species, associated with a one-year mortality rate of 32% (1). While traditionally seen in immunocompromised patients, particularly those with hematologic malignancies or neutropenia (2), recent research has shown an increasing incidence among non-neutropenic individuals (3), including documented cases in otherwise immunocompetent hosts (4). Diagnosing IPA in non-neutropenic patients is particularly challenging due to nonspecific clinical and radiological indicators. Standard diagnostic techniques, such as fungal cultures and histopathology, exhibit limited sensitivity and specificity, often leading to delayed diagnosis and treatment initiation, which contributes to the high morbidity and mortality observed in these populations. This challenge is further compounded by the rising recognition of IPA following severe respiratory viral infections, such as influenza and coronavirus disease 2019 (COVID-19), where reported incidence can exceed 10% with mortality rates around 50% (5,6). Fungal cultures are positive in only 20% to 40% of cases, and obtaining histopathological confirmation is often not feasible in critically ill patients (7). Furthermore, conventional microbiological tests, such as the serum galactomannan assay, show low sensitivity and specificity in immunocompetent populations, exacerbating delays in diagnosis and adequate treatment (8). Delays or errors in antifungal therapy significantly deteriorate patient outcomes (9). Additionally, unreliable prognostic biomarkers complicate risk stratification and disease management (10). Addressing these challenges necessitates the development of enhanced diagnostic strategies and targeted therapeutic interventions to improve patient outcomes.
The mortality rate of IPA in non-neutropenic patients exceeds that of traditional high-risk groups, highlighting the urgent need for early identification (11). However, significant gaps persist in understanding risk factors for this population. While emerging diagnostic and therapeutic strategies are being explored, identifying prognostically relevant risk factors remains crucial for recognizing high-risk patients and guiding clinical decision-making to improve outcomes. Although some studies have examined risk factors of IPA development (12-14), few have specifically addressed non-neutropenic IPA and those that have been limited by small sample sizes (15). Furthermore, predictive models tailored to this population are lacking. Existing research is predominantly based on single-center retrospective analyses, revealing considerable heterogeneity in risk factors, including diabetes, respiratory diseases, and corticosteroid exposure (16-18). However, systematic integration and external validation of these findings remain lacking. Additionally, most studies rely on traditional statistical methods, such as logistic regression, which may fail to capture complex, nonlinear associations and interactions within multidimensional clinical data. This limitation reduces the predictive efficiency and generalizability of current models. Addressing these gaps requires the development of robust, validated predictive models incorporating advanced analytical techniques to enhance the early detection of IPA among non-neutropenic patients.
Machine learning (ML) is a branch of artificial intelligence that enables computer systems to learn from data and improve their performance without explicit programming. ML algorithms leverage large datasets to extract meaningful patterns and features, making them particularly useful in predictive modeling. Over the past decade, ML has been increasingly applied in clinical prediction models, facilitating early disease detection, severity assessment, classification, and prognostication. Notably, ML has demonstrated significant success in differentiating, treating, and evaluating COVID-19. Compared with conventional logistic regression, ML can effectively model complex linear and nonlinear relationships among variables in large datasets, resulting in superior predictive performance (19,20). Several studies have explored ML-based risk factors and prediction models for IPA (21-25). However, existing models often face limitations, including small sample sizes, single-center experiences, exclusion of non-neutropenic IPA cases, and reliance on features that are impractical for routine clinical application. Moreover, few models have been explicitly designed for the prediction of IPA among non-neutropenic patients.
This study aims to systematically identify key risk factors for the development of IPA in non-neutropenic patients through meta-analysis and subsequently develop a high-precision diagnostic model using ML techniques. Leveraging readily available clinical data, the study integrates variables identified through a systematic review and employs diverse ML algorithms to construct a pragmatic and clinically applicable model. The meta-analysis, regarded as the gold standard in evidence-based medicine, synthesizes multi-source research evidence and mitigates biases inherent in individual studies (26). Meanwhile, ML algorithms, such as random forests, excel in managing high-dimensional data and capturing complex feature interactions, demonstrating significant utility in disease risk prediction (27). By integrating these methodologies, this study provides a robust tool for IPA diagnosis in non-neutropenic patients, ultimately enhancing clinical decision-making, facilitating timely treatment initiation, and improving patient outcomes. We present this article in accordance with the TRIPOD and PRISMA reporting checklists (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2135/rc) (28).
Methods
Ethics
The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of the General Hospital of Ningxia Medical University (approval No. KYLL-2025-0899). Informed consent was waived for this retrospective study. To provide a comprehensive overview of the study protocol, a schematic framework is presented in Figure S1.
Meta-analysis
The study protocol was registered with PROSPERO (registration number: CRD42024585724). This analysis aimed to understand IPA in non-neutropenic individuals and perform further research by critically evaluating existing literature, particularly on immunocompetent populations. We searched extensively across multiple databases, including Wanfang Medical Network, VIP Database, CNKI, PubMed, Web of Science, Embase, and the Cochrane Database, to identify studies discussing IPA risk factors in non-neutropenic patients. The search terms included but were not limited to: “IPA”, “risk factors”, and “non-neutropenic”. Studies published from the inception of these databases until December 2024 were assessed. Titles, abstracts, and full texts were screened based on predefined inclusion criteria, and relevant data were extracted for pooled analysis. The risk of bias in each study was evaluated using the Newcastle-Ottawa scale.
Subjects
Between January 2019 and December 2024, non-neutropenic patients diagnosed with IPA, hospitalized at the General Hospital of Ningxia Medical University, were included in the study. Patients were included in the study if they were ≥18 years old and met the diagnostic criteria for IPA based on established guidelines (29).
Confirmed IPA was defined by at least one of the following: (I) histopathological or cytopathological evidence of hyphal invasion consistent with Aspergillus spp. in specimens obtained from sterile sites, such as lung tissue acquired through biopsy or puncture; (II) positive culture of Aspergillus spp. from sterile site specimens related to infection.
Probable or suspected IPA required the concurrent presence of all of the following: (I) clinical manifestations, including persistent fever unresponsive to antibiotics for more than three days, recurrence of remittent fever, pleuritic pain, dyspnea, hemoptysis, or progressive respiratory function deterioration; (II) high-risk host factors, such as complications from influenza or COVID-19 infection, moderate to severe chronic obstructive pulmonary disease (COPD), decompensated liver cirrhosis, uncontrolled human immunodeficiency virus (HIV) infection (CD4+ <200/mm3), or solid tumors; (III) radiological findings, including pulmonary infiltrates or cavitary lesions; and (IV) microbiological evidence, such as positive Aspergillus culture in bronchoalveolar lavage fluid (BALF) or a positive galactomannan test in serum [optical density index (ODI) >0.5] or BALF (ODI ≥1.0).
Patients were excluded if they had incomplete clinical data, had been hospitalized for less than 24 h, or exhibited neutropenia, defined as an absolute neutrophil count <0.5×109/L. Additionally, those with severe immunosuppression due to advanced HIV infection, hematopoietic stem cell transplantation, or solid organ transplantation requiring intensive immunosuppressive therapy were also excluded. Individuals with mental disorders or congenital developmental defects that could interfere with data collection and analysis were not included in the study.
Data collection
Eligible patients were identified through hospital medical records, with data extracted from electronic health records to ensure comprehensive documentation. All clinical and pathological data were systematically collected upon admission, encompassing six key dimensions: demographic characteristics (age and gender distribution), clinical manifestations (symptomatology and vital sign parameters), comorbidities (including diabetes, chronic respiratory diseases, cardiovascular diseases, and renal disorders), laboratory diagnostics (routine blood tests, acute-phase reactants such as C-reactive protein and procalcitonin, and microbiological test results), imaging assessments [radiological findings from chest computed tomography (CT) scans], and blood gas analysis [respiratory function metrics, particularly the oxygenation index, arterial oxygen partial pressure/fractional inspired oxygen (PaO2/FiO2) ratio].
Statistical analysis
Meta-analysis: effect estimates for each risk factor were pooled using a random-effects model, given the anticipated heterogeneity across studies. The odds ratio (OR) and corresponding 95% confidence interval (CI) were calculated for each risk factor. Heterogeneity among studies was assessed using the I2 statistic and Cochran’s Q test. Publication bias was evaluated using Egger’s test, with P<0.05 suggesting potential small-study effects. A sensitivity analysis was conducted to assess the robustness of the pooled estimates by iteratively removing individual studies and recalculating ORs. The stability of the results was further validated by comparing outcomes from both fixed-effects and random-effects models. Statistical analyses were performed using RevMan 5.4 software and Stata 14.0, and a significance threshold of P<0.05 was applied for all tests.
Descriptive and predictive model analysis: descriptive statistics were presented as the median (interquartile range) for measurement data not adhering to a normal distribution, and inter-group comparisons were made using non-parametric tests. Categorical variables were expressed as percentages, and differences between groups were assessed using the χ2 or Fisher’s exact test. The predictive model development involved a two-step variable selection strategy per the TRIPOD statement recommendations (30). This approach shows the importance of evaluating the biological rationale and clinical applicability of variables alongside domain knowledge, beyond automated algorithm screening, to enhance the model’s interpretability and its potential for clinical translation (30). The performance of the predictive model and its external validation were assessed using the area under the receiver operating characteristic (ROC) curve (AUC). Additionally, the calibration capability of both the optimal prediction model and the external validation model was evaluated using calibration curves. Statistical analyses were conducted using SPSS 23.0 and R (version 4.3.2).
Results
Meta-analysis
The study selection process, detailed in Figure S2, included 13 studies (8,16-18,31-39). Data from these studies were aggregated for pooled estimates, and baseline characteristics, including bias assessment, are presented in Tables S1,S2. A random-effects model was applied, revealing independent risk factors for IPA development: diabetes (pooled OR =1.43, 95% CI: 1.24–1.64), respiratory disease (pooled OR =1.76, 95% CI: 1.41–2.19), corticosteroid exposure (pooled OR =1.48, 95% CI: 1.25–1.74), and smoking history (pooled OR =1.64, 95% CI: 1.34–2.01), with P<0.00001 for all (Figure S3). However, substantial heterogeneity (I2≥76%) and publication bias (Egger’s test, P<0.05) were observed across the studies. Sensitivity analysis confirmed the robustness of these findings, as OR values remained consistent between the random-effects and fixed-effects models (Table S3).
Publication bias was assessed using Begg’s test, which indicated significant bias for corticosteroid exposure (P=0.02), respiratory disease (P=0.003), and smoking history (P=0.050), whereas diabetes showed no significant bias (P=0.07). Egger’s test further confirmed significant publication bias for all examined risk factors (P<0.05; Table S4).
ML-model construction and verification
The study included 524 participants, with 422 in the training and 102 in the validation cohorts. The median age was 70 years, and 62.9% were male, with no significant sex differences (P=0.06). Key laboratory markers, including white blood cells, neutrophils, lymphocytes, eosinophils, procalcitonin, interleukin-6, C-reactive protein, and erythrocyte sedimentation rate, differed significantly between cohorts (P<0.001). Shortness of breath (64.9%) and cough (77.3%) were the most common symptoms, with higher prevalence in the derivation group (P=0.002; P＜0.001). Lung consolidation (60.3%) and pleural thickening (58.0%) were the most frequent imaging findings, whereas bilateral lung involvement was significantly higher in the derivation cohort (P<0.001). Diabetes (43.5%), hypertension (45.9%), and respiratory disease (45.9%) were the most common comorbidities, with kidney injury (P=0.001), septic shock (P<0.001), and hypertension (P=0.03) more prevalent in the derivation group. Mortality was 14.1%, with no significant difference (P=0.09), though SOFA (P<0.001) and Acute Physiologic Assessment and Chronic Health Evaluation (APACHE) II scores (P<0.001) indicated greater severity in the derivation cohort. These differences suggest variations in disease severity and inflammatory response, which may influence predictive modeling.
Variable selection adhered to TRIPOD guidelines, which emphasize balancing algorithmic screening with clinical relevance to enhance interpretability and translational potential. Least absolute shrinkage and selection operator (LASSO) regression [λ=0.028, 1 − standard error (SE); Figure 1] initially screened 22 variables. However, variables such as cough and pleural effusion, though statistically significant, were excluded due to their lack of clinical relevance. Ultimately, nine predictive variables were retained: diabetes, respiratory disease, low protein levels, corticosteroid exposure, antibiotic treatment, viral pneumonia, ICU admission, bacterial infections, and smoking history (Figure 1). Notably, the term “respiratory disease” primarily refers to COPD, whereas “viral pneumonia” predominantly pertains to H1N1 infections. Figure 2A shows the decision tree structure highlighting key split variables (such as antibiotic treatment, viral pneumonia, and ICU admission), while Figure 2B depicts the corresponding ROC curve, with an AUC of 0.915, indicating moderate discriminative capacity.
Figure 1.
LASSO regression analysis: (A) distribution of regression coefficients and (B) cross-validation curve. LASSO, least absolute shrinkage and selection operator.
Figure 2.
Decision tree analysis: (A) flowchart and (B) decision curve (ROC) analysis. AUC, area under the curve; ICU, intensive care unit; ROC, receiver operating characteristic.
The RF model (Figure 3A and Table 1) demonstrated the highest overall performance, achieving an AUC of 0.950, specificity of 0.905, and accuracy of 0.881, indicating strong predictive capability. The support vector machine (SVM) also performed well, with an AUC of 0.948, a sensitivity of 0.873, and an accuracy of 0.865. Logistic regression exhibited the highest sensitivity at 0.889 and a competitive AUC of 0.932, making it a robust predictive tool. Calibration analysis (Figure 3B) demonstrated that the RF model’s predicted probabilities were well-aligned with observed outcomes, and DCA (Figure 3C) indicated meaningful clinical net benefit, underscoring its potential utility in clinical practice. Other models, including K-nearest neighbors (KNN; Figure 4A), Naïve Bayes (Figure 4B), and Neural Networks (Figure 4C), displayed moderate predictive performances, with AUC values ranging from 0.930 to 0.941. The decision tree model, despite its good sensitivity and specificity (both 0.873), had a relatively lower AUC of 0.915 compared to the ensemble and kernel-based models. Overall, the RF model emerged as the most effective tool for IPA prediction. Feature importance analysis highlighted antibiotic treatment, viral pneumonia, and ICU admission as the most influential factors (Figure 4D). The calibration curve further confirms that the RF model’s predicted probabilities closely align with observed outcomes (Figure 5A).
Figure 3.
Performance evaluation of ML models for IPA prediction. (A) ROC curve analysis of multiple models, with RF (AUC =0.950) and SVM (AUC =0.948) demonstrating the highest predictive performance. Other models, including KNN (AUC =0.941), neural networks (AUC =0.939), logistic regression (AUC =0.932), Naïve Bayes (AUC =0.930), and decision tree (AUC =0.915), showed varying degrees of predictive capability. (B) Calibration curves showing agreement between predicted and observed outcomes. (C) DCA demonstrating the clinical net benefit across different threshold probabilities. AUC, area under the curve; DCA, decision curve analysis; IPA, invasive pulmonary aspergillosis; KNN, K-nearest neighbors; ML, machine learning; RF, random forest; ROC, receiver operating characteristic; SVM, support vector machine.
Table 1. Comparison of performance across ML models for predicting invasive pulmonary aspergillosis.
| Models | Sensitivity | Specificity | Accuracy | AUC |
|---|---|---|---|---|
| Decision tree | 0.873 | 0.873 | 0.873 | 0.915 |
| RF | 0.857 | 0.905 | 0.881 | 0.950 |
| SVM | 0.873 | 0.857 | 0.865 | 0.948 |
| KNN | 0.841 | 0.873 | 0.857 | 0.941 |
| Naïve Bayes | 0.841 | 0.825 | 0.833 | 0.930 |
| Neural network | 0.857 | 0.873 | 0.865 | 0.939 |
| Logistic | 0.889 | 0.857 | 0.873 | 0.932 |
AUC, area under the curve; KNN, K-nearest neighbors; ML, machine learning; RF, random forest; SVM, support vector machine.
Figure 4.
ML model evaluation: feature importance, parameter optimization, and performance visualization. This figure presents key analyses of ML models, including optimal K value selection for KNN (A), confusion matrix visualization for Naïve Bayes model (B), neural network architecture flowchart (C), and feature importance ranking for random forest model (D). ICU, intensive care unit; KNN, K-nearest neighbors; ML, machine learning.
Figure 5.
RF model performance evaluation. This figure illustrates the performance assessment of the RF model, including the calibration curve (A), the confusion matrix heatmap (B), and the variable importance visualization using a bubble chart (C), providing insights into the model’s classification accuracy and calibration reliability. MDA, Mean Decrease in Accuracy; MDG, Mean Decrease in Gini; RF, random forest.
Additionally, the RF model’s ability to identify high-risk patients and capture interactions among risk factors is visually represented through multiple analytical approaches. The confusion matrix heatmap (Figure 5B) illustrates the model’s classification performance, highlighting its accuracy in distinguishing risk groups. Additionally, the variable bubble chart (Figure 5C) provides an intuitive depiction of feature importance and their relative contributions, offering insights into key predictive factors. These visualizations enhance the model’s clinical interpretability, facilitating a more comprehensive understanding of risk stratification and supporting its potential for real-world clinical application.
External validation and model generalization
In the independent external validation cohort of 102 patients, comprising 51 IPA cases and 51 non-IPA cases, the RF model achieved an AUC of 0.856. The model demonstrated strong generalization capabilities with a sensitivity and specificity of 0.733 (Figure 6A,6B). Additionally, DCA (Figure 6C) confirmed that the model consistently provided a positive clinical net benefit within the external cohort, further supporting its potential utility in clinical decision-making. Despite significant differences in baseline characteristics such as white blood cell count, lymphocyte ratio, and SOFA score between the external validation and training cohorts (all P<0.001), the model consistently maintained high discrimination efficiency (Table 2). These results further validate the model’s performance on an independent dataset, reinforcing the rationale behind the variable selection process.
Figure 6.
Performance of the RF model in the external validation cohort. (A,B) ROC curves demonstrating the model’s discriminative performance in the independent validation cohort. (C) Decision curve analysis illustrating the positive net clinical benefit of the model across a range of threshold probabilities. AUC, area under the curve; RF, random forest; ROC, receiver operating characteristic.
Table 2. Baseline characteristics of study participants.
| Characteristic | Total (n=524) | Training (n=422) | Validation (n=102) | c2/t | P |
|---|---|---|---|---|---|
| Age, years | 70 [59.00–76.00] | 70 [59.00–76.00] | 67 [57.00–75.75] | −1.476 | 0.14 |
| Sex | 3.542 | 0.06 | |||
| Male | 330 (62.9) | 274 (64.9) | 56 (54.9) | ||
| Female | 194 (37.0) | 148 (35.1) | 46 (45.1) | ||
| PaO2/FiO2 ratio | 244 [194–287] | 244 [196–287] | 252 [178–287] | −0.326 | 0.75 |
| ICU admission | 206 (39.3) | 162 (38.4) | 44 (43.1) | 0.776 | 0.38 |
| Smoking history | 210 (40.3) | 163 (38.6) | 47 (46.1) | 1.900 | 0.17 |
| Corticosteroid exposure | 262 (50.0) | 210 (49.8) | 52 (50.9) | 0.049 | 0.83 |
| Therapy | |||||
| Antibiotic treatment | 256 (48.9) | 205 (48.6) | 51 (50.0) | 0.066 | 0.80 |
| Laboratory examination | |||||
| White blood cells (×109) | 8.09 [5.59–11.98] | 8.70 [5.80–12.59] | 6.49 [5.04–8.38] | −4.775 | <0.001 |
| Neutrophils (×109) | 6.26 [3.96–9.46] | 7.04 [4.40–10.62] | 4.39 [2.93–6.21] | −6.577 | <0.001 |
| Lymphocytes (×109) | 0.91 [0.50–1.48] | 0.81 [0.46–1.29] | 1.47 [0.96–1.79] | −6.530 | <0.001 |
| Eosinophils (×109) | 0.01 [0.00–0.09] | 0.00 [0.00–0.05] | 0.11 [0.03–0.21] | −8.030 | <0.001 |
| Procalcitonin (ng/mL) | 0.31 [0.08–1.82] | 0.45 [0.11–2.13] | 0.08 [0.06–0.21] | −8.478 | <0.001 |
| Interleukin-6 (pg/mL) | 8.00 [2.42–26.88] | 12 [3.76–35.93] | 1.42 [0.73–2.84] | −11.369 | <0.001 |
| C-reactive protein (mg/L) | 15.60 [5.00–45.50] | 20.25 [9.95–55.48] | 2.59 [1.34–8.39] | −9.939 | <0.001 |
| ESR (mm/h) | 12.00 [5.00–29.00] | 14.00 [5.00–32.00] | 7.00 [3.75–22.25] | −3.630 | <0.001 |
| Symptoms | |||||
| Fever | 275 (52.5) | 220 (52.1) | 55 (53.9) | 0.105 | 0.75 |
| Shortness of breath | 340 (64.9) | 287 (68.0) | 53 (51.9) | 9.286 | 0.002 |
| Chest pain | 74 (14.1) | 62 (14.7) | 12 (11.8) | 0.580 | 0.45 |
| Cough | 405 (77.3) | 312 (73.9) | 93 (91.2) | 13.914 | <0.001 |
| Chest tightness | 311 (59.4) | 259 (61.4) | 52 (50.9) | 3.678 | 0.055 |
| Hemoptysis | 52 (9.9) | 44 (14.7) | 8 (7.8) | 0.613 | 0.43 |
| Respite | 214 (40.8) | 169 (40.4) | 45 (44.1) | 0.563 | 0.45 |
| Chest imaging | |||||
| Pleural effusion | 228 (43.5) | 186 (44.1) | 42 (41.2) | 0.281 | 0.60 |
| Nodular shadow | 154 (29.4) | 120 (28.4) | 34 (33.3) | 0.949 | 0.33 |
| Patchy shadow | 277 (52.9) | 222 (52.6) | 55 (53.9) | 0.057 | 0.81 |
| Cavity | 71 (13.5) | 55 (13.0) | 16 (15.7) | 0.494 | 0.48 |
| Lung consolidation | 316 (60.3) | 251 (59.5) | 65 (63.7) | 0.619 | 0.43 |
| Bilateral lungs involvement | 117 (22.3) | 112 (26.5) | 5 (4.9) | 22.177 | <0.001 |
| Pleural thickening | 304 (58.0) | 248 (58.8) | 56 (54.9) | 0.504 | 0.48 |
| Disease | |||||
| Viral pneumonia | 234 (44.7) | 191 (45.3) | 43 (42.2) | 0.320 | 0.57 |
| Bacterial infections | 203 (38.7) | 158 (37.4) | 45 (44.1) | 1.543 | 0.21 |
| Kidney injury | 133 (25.4) | 120 (28.4) | 13 (12.7) | 10.679 | 0.001 |
| Hypoproteinemia | 245 (46.8) | 193 (45.7) | 52 (50.9) | 0.908 | 0.34 |
| Septic shock | 85 (16.2) | 82 (19.4) | 3 (2.9) | 16.436 | <0.001 |
| Organ failure | 178 (33.9) | 148 (35.1) | 30 (29.4) | 1.173 | 0.28 |
| Diabetes | 228 (43.5) | 183 (43.4) | 45 (44.1) | 0.019 | 0.89 |
| Hypertension | 241 (45.9) | 204 (48.3) | 37 (36.3) | 4.815 | 0.03 |
| Coronary heart disease | 120 (22.9) | 102 (24.2) | 18 (17.6) | 1.980 | 0.16 |
| Respiratory disease | 241 (45.9) | 191 (45.3) | 50 (49.0) | 0.467 | 0.49 |
| Other illnesses | 94 (17.9) | 81 (19.2) | 13 (12.7) | 2.321 | 0.13 |
| Outcome | |||||
| Death | 74 (14.1) | 65 (15.4) | 9 (8.8) | 2.932 | 0.09 |
| Clinical severity scores | |||||
| CURB-65 score | 2.00 [2.00–3.00] | 2.00 [2.00–3.00] | 2.00 [1.00–3.00] | −1.461 | 0.14 |
| SOFA score | 9.00 [7.00–10.00] | 9.00 [8.00–11.00] | 8.00 [7.00–9.00] | −7.055 | <0.001 |
| APACHE II score | 11.00 [10.00–13.00] | 12.00 [10.00–14.00] | 11.00 [9.00–12.00] | −3.975 | <0.001 |
Data are presented as n (%) or median [interquartile range]. APACHE II, Acute Physiologic Assessment and Chronic Health Evaluation II; CURB-65, Confusion, Uremia, Respiratory Rate, Blood Pressure, Age ≥65 years; ESR, erythrocyte sedimentation rate; ICU, intensive care unit; PaO2/FiO2, arterial oxygen partial pressure/fractional inspired oxygen; SOFA, Sequential Organ Failure Assessment.
Discussion
IPA is primarily associated with immune status, with neutropenia recognized as a significant risk factor. However, the incidence of IPA in non-neutropenic patients is on the rise. A significant U.S. study identified COPD, diabetes, and hematological malignancies as high-risk factors for invasive fungal infections (IFIs), highlighting the increasing prevalence of non-neutropenic IPA (40). Emerging risk factors include novel immunosuppressive therapies and co-infections with pathogens such as influenza and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) (41,42). The diagnosis of IPA in non-neutropenic patients is fraught with challenges due to nonspecific symptoms, which often lead to delayed treatment and result in higher mortality rates compared to neutropenic cases (11,43). This discrepancy is linked to differences in clinical management, as neutropenic patients typically undergo earlier screening and intervention. Early identification of at-risk individuals and timely treatment strategies are crucial to reduce mortality.
In this study, we systematically identified independent risk factors for IPA in non-neutropenic patients and developed a robust ML-based risk prediction model. A meta-analysis confirmed diabetes, respiratory disease, corticosteroid exposure, and smoking history as key risk factors. Using a training cohort, we applied LASSO regression to refine predictive variables, ultimately constructing an RF model with superior performance (AUC =0.950, sensitivity =0.857, specificity =0.905, accuracy =0.881). External validation further demonstrated strong generalization capabilities (AUC =0.856, sensitivity and specificity =0.733). Feature importance analysis highlighted antibiotic treatment, viral pneumonia, and ICU admission as the most influential predictors. Additionally, SHapley Additive exPlanations (SHAP) analysis revealed that antibiotic misuse post-influenza may exacerbate IPA susceptibility, emphasizing the need for judicious antibiotic use. The model’s calibration curve, confusion matrix heatmap, and variable bubble chart enhanced its interpretability, supporting its clinical applicability for early risk stratification.
COPD is the most prevalent comorbidity among non-neutropenic patients with IPA and serves as an independent risk factor for its development (44). Our findings align with this, as a respiratory disease in our study primarily involved COPD. Impaired ciliary function and chronic inflammation in COPD reduce fungal clearance, facilitating Aspergillus colonization and biofilm formation (45). Additionally, in vitro studies show that corticosteroids enhance Aspergillus fumigatus growth by 30–40% (46), increasing its invasive potential (47). Broad-spectrum antibiotic use further disrupts the microbiota balance, heightening the risk of secondary fungal infection. Feature importance analysis revealed a significant interaction between corticosteroid exposure and antibiotic treatment, highlighting the need for judicious antimicrobial use to prevent microbiota dysbiosis. Diabetes was also confirmed as a key risk factor, as hyperglycemia impairs neutrophil chemotaxis, phagocytosis, and reactive oxygen species production, increasing fungal susceptibility even in non-neutropenic patients (48). Long-term glycemic control and regular follow-up are crucial to mitigate this risk. Similarly, tobacco exposure impairs alveolar macrophage function, exacerbating IPA risk (49). Severe influenza, particularly H1N1, is strongly associated with IPA, with an incidence as high as 19% (50,51). Influenza-induced bronchial mucosal damage and impaired mucociliary clearance create a conducive environment for Aspergillus infection (52). Hypoproteinemia, a marker of malnutrition, may further increase fungal susceptibility even in immunocompetent individuals (53). Bacterial co-infections were significantly more frequent in non-neutropenic IPA patients (56% vs. 15%, P<0.05) (11), consistent with our findings. Glucocorticoids and broad-spectrum antibiotics, commonly used in bacterial infections, may inadvertently promote IPA development. Clinicians should, therefore, exercise caution in their use and closely monitor for fungal infections. ICU admission further elevates IPA risk due to prolonged hospitalization, invasive procedures, and environmental factors. Enhanced surveillance and preventive strategies in ICU settings are essential for early detection and intervention.
Recent studies highlight the potential of ML algorithms in predicting invasive pulmonary infections by integrating clinical and radiological evidence. For instance, Yan et al. developed an ML model that combines radiomic features with clinical factors, achieving superior diagnostic performance with an AUC of 0.844 compared to models based solely on clinical (AUC =0.696) or radiomic data (AUC =0.767), demonstrating its utility in differentiating IFIs (21). Similarly, Wang et al. applied deep learning to detect IPA using CT imaging, achieving an AUC of 0.95 (22). Cao et al. developed an ML-based predictive model for IFIs in ICU patients, with the BL-SMOTE RF model demonstrating the highest performance (AUC =0.88) (23). In our study, we constructed an RF model with superior predictive performance (AUC =0.950, sensitivity =0.857, specificity =0.905, accuracy =0.881) for IPA in non-neutropenic patients. The RF model effectively captures nonlinear interactions among multidimensional variables, significantly enhancing predictive efficiency in immunocompetent populations. Among ML algorithms, RF is particularly well-suited for high-dimensional clinical datasets, constructing multiple decision trees and aggregating results to model complex nonlinear interactions while maintaining computational efficiency. Its robustness against multicollinearity and capacity for variable importance assessment establish it as a leading predictive algorithm (27). By integrating systematic meta-analysis evidence into ML models, this study overcomes the limitations of single-center retrospective research, which traditionally relies on logistic regression. This approach enhances the generalizability and predictive power of IPA risk assessment in non-neutropenic patients, facilitating earlier diagnosis and improved clinical decision-making.
However, there are several limitations in this study. First, the geographic scope of the meta-analysis was limited, as most included studies were conducted in China. While this reflects a substantial body of research in this population, future systematic reviews should aim for broader international inclusion to enhance the global generalizability of identified risk factors. Second, although multiple variables were identified through meta-analysis, some were not incorporated into the final model, potentially limiting its applicability. Additionally, uncorrected differences in comorbidities may introduce confounding biases. Nevertheless, the model demonstrated strong discriminative performance (AUC =0.856) in the external validation cohort, suggesting that its core predictors are robust across diverse populations. Future research should expand sample sizes and integrate multidimensional biomarkers, including host immune indicators and fungal-specific antigens, to enhance model sensitivity and specificity. Moreover, advancements in multi-omics offer deeper insights into IPA pathogenesis and biomarker discovery, refining diagnostic precision and facilitating the development of targeted interventional strategies.
Conclusions
This study successfully developed and validated an ML-based predictive model for IPA in non-neutropenic patients, integrating meta-analysis and clinical evidence with advanced ML algorithms. The RF model exhibited high predictive accuracy during internal validation and demonstrated strong generalization in an external cohort. Key predictors identified included diabetes, respiratory disease, low protein levels, corticosteroid exposure, antibiotic treatment, viral pneumonia, ICU admission, bacterial infections, and smoking history. By leveraging high-dimensional clinical data, the RF model effectively captured nonlinear interactions, surpassing the capabilities of traditional models. Future research should focus on expanding sample sizes and incorporating multi-omics biomarkers to further enhance IPA risk prediction and improve patient outcomes.
Supplementary
The article’s supplementary files as
Acknowledgments
None.
Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of the General Hospital of Ningxia Medical University (approval No. KYLL-2025-0899). Informed consent was waived for this retrospective study.
Footnotes
Reporting Checklist: The authors have completed the TRIPOD and PRISMA reporting checklists. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2135/rc
Funding: This work was supported by the National Natural Science Foundation of China (No. 82360022, to J.Z.), and the Ningxia Key Research and Development Project (Nos. 2022BEG03102 and 2024BEH04027, to J.Z.).
Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2135/coif). The authors have no conflicts of interest to declare.
Data Sharing Statement
Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2135/dss
References
- 1.Henao-Martínez AF, Corbisiero MF, Salter I, et al. Invasive pulmonary aspergillosis real-world outcomes: Clinical features and risk factors associated with increased mortality. Med Mycol 2023;61:myad074. 10.1093/mmy/myad074 [DOI] [PubMed] [Google Scholar]
- 2.Ledoux MP, Guffroy B, Nivoix Y, et al. Invasive Pulmonary Aspergillosis. Semin Respir Crit Care Med 2020;41:80-98. 10.1055/s-0039-3401990 [DOI] [PubMed] [Google Scholar]
- 3.Latgé JP, Chamilos G. Aspergillus fumigatus and Aspergillosis in 2019. Clin Microbiol Rev 2019;33:e00140-18 . 10.1128/CMR.00140-18 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.AlShaheen H, Abuzied Y, Aldalbahi H, et al. Aspergilloma in an Immunocompetent Host: A Case Report. Cureus 2023;15:e40727. 10.7759/cureus.40727 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Cadena J, Thompson GR, 3rd, Patterson TF. Aspergillosis: Epidemiology, Diagnosis, and Treatment. Infect Dis Clin North Am 2021;35:415-34. 10.1016/j.idc.2021.03.008 [DOI] [PubMed] [Google Scholar]
- 6.Shi C, Shan Q, Xia J, et al. Incidence, risk factors and mortality of invasive pulmonary aspergillosis in patients with influenza: A systematic review and meta-analysis. Mycoses. 2022;65:152-163. 10.1111/myc.13410 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Yusuf E, Vonk A, van den Akker JPC, et al. Frequency of Positive Aspergillus Tests in COVID-19 Patients in Comparison to Other Patients with Pulmonary Infections Admitted to the Intensive Care Unit. J Clin Microbiol 2021;59:e02278-20. 10.1128/JCM.02278-20 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Jenks JD, Mehta SR, Taplitz R, et al. Point-of-care diagnosis of invasive aspergillosis in non-neutropenic patients: Aspergillus Galactomannan Lateral Flow Assay versus Aspergillus-specific Lateral Flow Device test in bronchoalveolar lavage. Mycoses 2019;62:230-6. 10.1111/myc.12881 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Barchiesi F, Santinelli A, Biscotti T, et al. Delay of antifungal therapy influences the outcome of invasive aspergillosis in experimental models of infection. J Antimicrob Chemother 2016;71:2230-3. 10.1093/jac/dkw111 [DOI] [PubMed] [Google Scholar]
- 10.Patterson TF. Clinical utility and development of biomarkers in invasive aspergillosis. Trans Am Clin Climatol Assoc 2011;122:174-83. [PMC free article] [PubMed] [Google Scholar]
- 11.Cornillet A, Camus C, Nimubona S, et al. Comparison of epidemiological, clinical, and biological features of invasive aspergillosis in neutropenic and nonneutropenic patients: a 6-year survey. Clin Infect Dis 2006;43:577-84. 10.1086/505870 [DOI] [PubMed] [Google Scholar]
- 12.Gu Y, Ye X, Liu Y, et al. A risk-predictive model for invasive pulmonary aspergillosis in patients with acute exacerbation of chronic obstructive pulmonary disease. Respir Res 2021;22:176. 10.1186/s12931-021-01771-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Friol A, Dumas G, Pène F, et al. A multivariable prediction model for invasive pulmonary aspergillosis in immunocompromised patients with acute respiratory failure (IPA-GRRR-OH score). Intensive Care Med 2025;51:72-81. 10.1007/s00134-024-07767-z [DOI] [PubMed] [Google Scholar]
- 14.Li Y, Ren X, Wang Q, et al. A Predictive Model for Pulmonary Aspergillosis in ICU Patients: A Multicenter Retrospective Cohort Study. Infect Drug Resist 2025;18:441-54. 10.2147/IDR.S493019 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Liu F, Chen W, Qi H, et al. Clinical characteristics and prognosis of patients treated as invasive pulmonary aspergillosis outside of severe immunosuppression. Sci Rep 2025;15:3379. 10.1038/s41598-025-87605-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Russo A, Falcone M, Vena A, et al. Invasive pulmonary aspergillosis in non-neutropenic patients: analysis of a 14-month prospective clinical experience. J Chemother 2011;23:290-4. 10.1179/joc.2011.23.5.290 [DOI] [PubMed] [Google Scholar]
- 17.Lugosi M, Alberti C, Zahar JR, et al. Aspergillus in the lower respiratory tract of immunocompetent critically ill patients. J Infect 2014;69:284-92. 10.1016/j.jinf.2014.04.010 [DOI] [PubMed] [Google Scholar]
- 18.He Q, Cao J, Zhang M, et al. IL-17 in plasma and bronchoalveolar lavage fluid in non-neutropenic patients with invasive pulmonary aspergillosis. Front Cell Infect Microbiol 2024;14:1402888. 10.3389/fcimb.2024.1402888 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Handelman GS, Kok HK, Chandra RV, et al. eDoctor: machine learning and the future of medicine. J Intern Med. 2018;284:603-619. 10.1111/joim.12822 [DOI] [PubMed] [Google Scholar]
- 20.Rashid A, Al-Obeidat F, Hafez W, et al. ADVANCING THE UNDERSTANDING OF CLINICAL SEPSIS USING GENE EXPRESSION-DRIVEN MACHINE LEARNING TO IMPROVE PATIENT OUTCOMES. Shock 2024;61:4-18. 10.1097/SHK.0000000000002227 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Yan C, Hao P, Wu G, et al. Machine learning-based combined nomogram for predicting the risk of pulmonary invasive fungal infection in severely immunocompromised patients. Ann Transl Med 2022;10:514. 10.21037/atm-21-4980 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Wang W, Li M, Fan P, et al. Prototype early diagnostic model for invasive pulmonary aspergillosis based on deep learning and big data training. Mycoses 2023;66:118-27. 10.1111/myc.13540 [DOI] [PubMed] [Google Scholar]
- 23.Cao Y, Li Y, Wang M, et al. Interpretable machine learning for predicting risk of invasive fungal infection in critically ill patients in the intensive care unit: a retrospective cohort study based on MIMIC-IV database. Shock 2024;61:817-27. 10.1097/SHK.0000000000002312 [DOI] [PubMed] [Google Scholar]
- 24.Zhang K, Zhao G, Liu Y, et al. Clinic, CT radiomics, and deep learning combined model for the prediction of invasive pulmonary aspergillosis. BMC Med Imaging 2024;24:264. 10.1186/s12880-024-01442-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Liu J, Zhang J, Wang H, et al. Machine Learning Methods Based on Chest CT for Predicting the Risk of COVID-19-Associated Pulmonary Aspergillosis. Acad Radiol 2025;32:3725-38. 10.1016/j.acra.2025.01.027 [DOI] [PubMed] [Google Scholar]
- 26.Chalmers TC. Meta-analysis in clinical medicine. Trans Am Clin Climatol Assoc 1988;99:144-50. [PMC free article] [PubMed] [Google Scholar]
- 27.Wang F, Wang Y, Ji X, et al. Effective Macrosomia Prediction Using Random Forest Algorithm. Int J Environ Res Public Health 2022;19:3245. 10.3390/ijerph19063245 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372: n71. 10.1136/bmj.n71 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Bassetti M, Giacobbe DR, Agvald-Ohman C, et al. Invasive Fungal Diseases in Adult Patients in Intensive Care Unit (FUNDICU): 2024 consensus definitions from ESGCIP, EFISG, ESICM, ECMM, MSGERC, ISAC, and ISHAM. Intensive Care Med 2024;50:502-15. 10.1007/s00134-024-07341-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Collins GS, Reitsma JB, Altman DG, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ 2015;350:g7594. 10.1136/bmj.g7594 [DOI] [PubMed] [Google Scholar]
- 31.Russo A, Giuliano S, Vena A, et al. Predictors of mortality in non-neutropenic patients with invasive pulmonary aspergillosis: does galactomannan have a role? Diagn Microbiol Infect Dis 2014;80:83-6. 10.1016/j.diagmicrobio.2014.05.015 [DOI] [PubMed] [Google Scholar]
- 32.Özger S, Hızel K, Kalkancı A, et al. Evaluation of risk factors for invasive pulmonary aspergillosis and detection of diagnostic values of galactomannan and PCR methods in bronchoalveolar lavage samples from non-neutropenic intensive care unit patients. Mikrobiyol Bul 2015;49:565-75. 10.5578/mb.9906 [DOI] [PubMed] [Google Scholar]
- 33.Lin PC, Lai QQ, Zhou Y, et al. The diagnostic performance of galactomannan detection for invasive pulmonary aspergillosis in non-neutropenic hosts. Chinese Journal of Tuberculosis and Respiratory Diseases 2016;39:929-33. 10.3760/cma.j.issn.1001-0939.2016.12.005 [DOI] [PubMed] [Google Scholar]
- 34.Yu Y, Zhu C, Shen H, et al. Galactomannan detection in bronchoalveolar lavage fluid corrected by urea dilution for the diagnosis of invasive pulmonary aspergillosis among nonneutropenic patients. J Thorac Dis 2019;11:465-76. 10.21037/jtd.2019.01.07 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Inoue K, Muramatsu K, Nishimura T, et al. Association between early diagnosis of and inpatient mortality from invasive pulmonary aspergillosis among patients without immunocompromised host factors: a nationwide observational study. Int J Infect Dis 2022;122:279-84. 10.1016/j.ijid.2022.05.048 [DOI] [PubMed] [Google Scholar]
- 36.Lu Y, Liu L, Li H, et al. The clinical value of Aspergillus-specific IgG antibody test in the diagnosis of nonneutropenic invasive pulmonary aspergillosis. Clin Microbiol Infect 2023;29:797.e1-7. 10.1016/j.cmi.2023.02.002 [DOI] [PubMed] [Google Scholar]
- 37.Zhu N, Zhou D, Xiong W, et al. Performance of mNGS in bronchoalveolar lavage fluid for the diagnosis of invasive pulmonary aspergillosis in non-neutropenic patients. Front Cell Infect Microbiol 2023;13:1271853. 10.3389/fcimb.2023.1271853 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Sheng L, Jiang W, Yao Y, et al. Value Evaluation of Quantitative Aspergillus fumigatus-Specific IgG Antibody Test in the Diagnosis of Non-neutropenic Invasive Pulmonary Aspergillosis. Infect Drug Resist 2024;17:2043-52. 10.2147/IDR.S460513 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Ma F, Sun A, Xu YP, et al. Diagnosis and prognostic analysis of invasive pulmonary aspergillosis in non-neutropenic patients with chronic obstructive pulmonary disease. J Clin Pulm Med 2024;29:649-53. [Google Scholar]
- 40.Menzin J, Meyers JL, Friedman M, et al. Mortality, length of hospitalization, and costs associated with invasive fungal infections in high-risk patients. Am J Health Syst Pharm 2009;66:1711-7. 10.2146/ajhp080325 [DOI] [PubMed] [Google Scholar]
- 41.Cadena J, Thompson GR, 3rd, Patterson TF. Aspergillosis: Epidemiology, Diagnosis, and Treatment. Infect Dis Clin North Am 2021;35:415-34. 10.1016/j.idc.2021.03.008 [DOI] [PubMed] [Google Scholar]
- 42.Shi C, Shan Q, Xia J, et al. Incidence, risk factors and mortality of invasive pulmonary aspergillosis in patients with influenza: A systematic review and meta-analysis. Mycoses 2022;65:152-63. 10.1111/myc.13410 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Russo A, Tiseo G, Falcone M, et al. Pulmonary Aspergillosis: An Evolving Challenge for Diagnosis and Treatment. Infect Dis Ther 2020;9:511-24. 10.1007/s40121-020-00315-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Bassetti M, Carnelutti A, Righi E. Issues in the management of invasive pulmonary aspergillosis in non-neutropenic patients in the intensive care unit: A role for isavuconazole. IDCases 2018;12:7-9. 10.1016/j.idcr.2018.02.011 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Ramage G, Rajendran R, Gutierrez-Correa M, et al. Aspergillus biofilms: clinical and industrial significance. FEMS Microbiol Lett 2011;324:89-97. 10.1111/j.1574-6968.2011.02381.x [DOI] [PubMed] [Google Scholar]
- 46.Ng TT, Robson GD, Denning DW. Hydrocortisone-enhanced growth of Aspergillus spp.: implications for pathogenesis. Microbiology (Reading) 1994;140 (Pt 9):2475-9. 10.1099/13500872-140-9-2475 [DOI] [PubMed] [Google Scholar]
- 47.Peghin M, Ruiz-Camps I, Garcia-Vidal C, et al. Unusual forms of subacute invasive pulmonary aspergillosis in patients with solid tumors. J Infect 2014;69:387-95. 10.1016/j.jinf.2014.03.018 [DOI] [PubMed] [Google Scholar]
- 48.Thind MK, Uhlig HH, Glogauer M, et al. A metabolic perspective of the neutrophil life cycle: new avenues in immunometabolism. Front Immunol 2023;14:1334205. 10.3389/fimmu.2023.1334205 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.Wu Y, Xu H, Li L, et al. Susceptibility to Aspergillus Infections in Rats with Chronic Obstructive Pulmonary Disease via Deficiency Function of Alveolar Macrophages and Impaired Activation of TLR2. Inflammation 2016;39:1310-8. 10.1007/s10753-016-0363-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50.Schauwvlieghe AFAD, Rijnders BJA, Philips N, et al. Invasive aspergillosis in patients admitted to the intensive care unit with severe influenza: a retrospective cohort study. Lancet Respir Med 2018;6:782-92. 10.1016/S2213-2600(18)30274-1 [DOI] [PubMed] [Google Scholar]
- 51.Duan Y, Ou X, Chen Y, et al. Severe Influenza With Invasive Pulmonary Aspergillosis in Immunocompetent Hosts: A Retrospective Cohort Study. Front Med (Lausanne) 2020;7:602732. 10.3389/fmed.2020.602732 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52.Shieh WJ, Blau DM, Denison AM, et al. 2009 pandemic influenza A (H1N1): pathology and pathogenesis of 100 fatal cases in the United States. Am J Pathol 2010;177:166-75. 10.2353/ajpath.2010.100115 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 53.Ader F, Nseir S, Guery B, et al. [Acute invasive pulmonary aspergillosis in chronic lung disease--a review]. Rev Mal Respir 2006;23:6s11-6s20. [PubMed]