Aspergillus conidia are ubiquitous airborne spores that are continuously released into the environment. These spores can readily penetrate the alveoli, making the lungs the primary site of infection and leading to invasive pulmonary aspergillosis (IPA) (Bulpa et al., 2007). Historically, IPA was regarded as a disease largely confined to individuals with neutropenia or other forms of immunocompromise. However, recent epidemiological studies have shown a steady increase in incidence among non-neutropenic patients (Latgé and Chamilos, 2019). Early diagnosis of IPA remains particularly challenging. In non-neutropenic populations, the absence of specific clinical manifestations, coupled with the limited sensitivity and specificity of conventional microbiological assays, such as the serum galactomannan test, frequently results in delayed diagnosis and suboptimal treatment (Jenks et al., 2019).

Conventional diagnostic approaches for Aspergillus detection include fungal culture and direct microscopic examination of bronchoalveolar lavage fluid (BALF) or sputum, as well as histopathological evaluation of lung biopsy specimens (Ascioglu et al., 2002). However, these methods have notable limitations. Culture, although considered a diagnostic reference standard, is time-consuming, requires viable organisms, and has a low positive yield because of the fastidious growth of Aspergillus spp. and the influence of prior antifungal therapy. Moreover, the turnaround time for histopathology is long, and invasive sampling carries substantial procedural risk. Although direct microscopy of BALF or sputum provides a rapid and simple alternative, its diagnostic performance is suboptimal, with substantial variability in both sensitivity and specificity (Denning et al., 2011). In addition, metagenomic next-generation sequencing (mNGS) enables the unbiased detection of diverse pathogens, including rare ones (Miller et al., 2019; Ramachandran and Wilson, 2020). However, its clinical adoption remains limited by high costs, host DNA interference, dual DNA/RNA workflows, and the absence of standardized protocols or large-scale validation, leading to subjective interpretation of results (Zheng et al., 2021).

Recently, highly multiplexed polymerase chain reaction–based targeted next-generation sequencing (tNGS) has emerged in China as a promising complement or alternative to mNGS for infectious disease diagnostics. This approach utilizes curated panels that encompass clinically relevant pathogens, thereby enhancing analytical sensitivity and enabling detection even in specimens with low microbial loads (Mamanova et al., 2010; Dennis et al., 2022). Targeted amplification also mitigates interference from host background sequences, improving data efficiency and reducing costs. However, quantitative sequencing reads generated by tNGS should not be interpreted as independent markers of pathogenicity; rather, they must be integrated with clinical findings and ancillary diagnostic evidence (Gu et al., 2019). Interpretation based solely on absolute read counts remains controversial due to substantial inter-platform variability (Lee et al., 2014; Khamina et al., 2022). Notably, tNGS (Deeplex) performance in cerebrospinal fluid has been shown to correlate strongly with bacterial load; the probability of Mycobacterium tuberculosis detection decreased markedly with increasing Xpert cycle threshold (Ct) values, with stepwise differences in read depth across Ct strata (Tram et al., 2024). This finding indicates that tNGS read depth and Xpert Ct have comparable quantitative and clinical significance. Accordingly, integrating normalized Aspergillus read counts from BALF-tNGS, stratified into cohort-based low, medium, and high tiers, into diagnostic models may improve discrimination of IPA from airway colonization in non-neutropenic patients.

Therefore, this multicenter retrospective study aimed to develop and validate an integrated diagnostic model that combines BALF-tNGS data with key host clinical factors to differentiate IPA from airway colonization in non-neutropenic patients. By incorporating normalized Aspergillus read counts as a semi-quantitative indicator of fungal burden, the study sought to enhance diagnostic accuracy, improve clinical decision-making, and provide a practical framework for interpreting tNGS results in real-world settings.

Growing evidence suggests that conditions commonly found in non-neutropenic hosts, such as COPD, diabetes mellitus, and hematologic malignancies, are independent risk factors for invasive fungal infections, as demonstrated in a seminal U.S. study (Menzin et al., 2009). In contrast to neutropenic patients, non-neutropenic individuals often exhibit non-specific or indolent manifestations, leading to under-recognition, diagnostic delay, and higher attributable mortality (Cornillet et al., 2006). In this context, the prompt and accurate identification of IPA is essential for optimizing antifungal therapy and improving outcomes (Bassetti and Bouza, 2017). In the present multicenter cohort, we integrated normalized Aspergillus reads from BALF-tNGS with host clinical factors to develop a diagnostic model capable of distinguishing IPA from colonization. Seven independent predictors were identified: Aspergillus load, diabetes, corticosteroid exposure, bacterial co-infection, ICU admission, presence of a nodular shadow on imaging, and positive BALF culture. The model demonstrated outstanding discrimination (AUC = 0.966), with sensitivity of 91.8%, specificity of 95.2%, PPV of 96.1%, and NPV of 90.0%. Calibration was excellent, and DCA confirmed substantial net clinical benefit across a wide range of threshold probabilities. Collectively, by coupling fungal molecular load with key host determinants, the integrated model substantially outperformed conventional microbiologic or radiologic criteria, establishing a framework for evidence-based interpretation of tNGS data in the diagnosis of IPA among non-neutropenic patients.

A key observation from this study was that the highest prevalence of IPA occurred in the medium–Aspergillus-load group (75.4%), significantly exceeding that observed in both the low-load (36.7%) and high-load (38.3%) strata. Patients in the medium-load group also exhibited greater disease severity, characterized by higher SOFA and APACHE II scores, as well as more frequent corticosteroid exposure (78.0%), diabetes mellitus (63.6%), and bacterial co-infection (81.4%) compared to those in the other groups. To further delineate the quantitative relationship between fungal molecular burden and IPA risk, both restricted cubic spline (RCS) and generalized additive model (GAM) analyses were performed, treating normalized Aspergillus reads as a continuous variable.

The RCS model revealed a significant overall association between read count and the probability of IPA (overall P = 0.033), whereas the test for nonlinearity was not significant (P for nonlinearity=0.412), indicating an approximately linear dose–response relationship. Consistently, the GAM smoothing curve showed a steadily increasing IPA probability with higher read counts (P = 0.0014), although confidence intervals widened at the extremes of the distribution. These findings suggest that, at the population level, higher fungal molecular signals generally correspond to an increased risk of IPA.

However, when the continuous variable was categorized into discrete strata (low, medium, and high) for multivariable modeling, a contrasting pattern emerged. After adjustment for host factors, including diabetes, corticosteroid exposure, bacterial co-infection, and ICU admission, a higher categorized fungal load was paradoxically associated with a lower odds ratio for IPA (OR = 0.21, 95% CI 0.09–0.51). This association remained robust in sensitivity analyses using Firth penalized regression, which reduced the magnitude of extreme odds ratios and mitigated the risk of overfitting. Rather than indicating a protective effect of higher fungal burden, this finding underscores the complex and context-dependent relationship between molecular fungal signals and invasive disease. Specifically, categorized fungal load captures only part of the pathogenic process and is strongly modulated by host susceptibility and clinical context. The divergent effects observed between continuous and categorized analyses reinforce the notion that tNGS-derived fungal read counts should not be interpreted in isolation but must be integrated with host factors to accurately distinguish invasive infection from colonization.

This seemingly inconsistent pattern highlights the complex, nonlinear relationship between fungal burden and invasive disease. Specifically, an elevated molecular load does not necessarily reflect tissue invasion (Fang et al., 2023); instead, it may represent an acute but predominantly noninvasive state, with residual fungal nucleic acids persisting after pathogen clearance (Żarczyńska et al., 2023). In contrast, moderate fungal loads may correspond to an active yet balanced state of invasion, reflecting a dynamic equilibrium between the host and pathogen. This interpretation is supported by the lower rate of IPA diagnosis in the high-load group, which may reflect limited viable fungal presence despite abundant nucleic acid signals (Sturny-Leclère et al., 2025). This phenomenon is analogous to that observed in tuberculosis, where cell-free mycobacterial DNA may persist long after viable bacilli have been eliminated (Hernández-Pando et al., 2000). Collectively, these results reinforce that fungal molecular abundance cannot be directly equated with bacterial burden and that the interpretation of tNGS-derived fungal load must be integrated with the host’s clinical context to distinguish true invasive infection from colonization. This principle forms the core rationale of the composite diagnostic model proposed in this study.

The present study further delineates the host determinants that modulate susceptibility to IPA in non-neutropenic patients. Diabetes mellitus, corticosteroid exposure, bacterial co-infection, and ICU admission emerged as the most influential risk factors in our multivariable model, each contributing distinct yet synergistic pathways of immune compromise. Diabetes was identified as a powerful predictor (OR = 25.6), consistent with the well-established association between hyperglycemia and impaired innate immune function. A sustained hyperglycemic milieu disrupts metabolic homeostasis, attenuating neutrophil chemotaxis, phagocytosis, and reactive oxygen species generation, thereby predisposing individuals to fungal invasion even in the absence of neutropenia (Thind et al., 2023). Corticosteroid therapy, another major determinant (OR = 24.2), exerts a dual influence by suppressing macrophage activity and enhancing fungal virulence through glucocorticoid-responsive gene expression in Aspergillus fumigatus (Peghin et al., 2014). Bacterial co-infection was also a frequent finding (69.3% overall) and amplified the risk of IPA nearly tenfold (OR = 10.3). Beyond diagnostic masking, bacterial infections and their empirical treatment with broad-spectrum antibiotics may perturb the airway microbiome and local immune tone, facilitating fungal proliferation and invasion. Similarly, ICU admission (OR = 6.1) represents a composite risk state encompassing critical illness, mechanical ventilation, and invasive procedures, compounded by environmental factors such as elevated humidity that promote Aspergillus spore persistence (Koulenti et al., 2023). Collectively, these findings highlight that non-neutropenic IPA arises not from a single dominant factor but from the cumulative disruption of multiple host defense barriers. Rigorous glycemic control, judicious corticosteroid use, and targeted fungal surveillance, particularly in ICU settings, are therefore imperative for early recognition and prevention in this expanding at-risk population.

tNGS offers several technical and clinical advantages that make it particularly suitable for diagnosing respiratory fungal infections. It employs multiplex PCR amplification of predefined pathogen targets, yielding high analytical sensitivity and specificity within 24 h (Li et al., 2022). A key methodological innovation of the present study is the use of normalized RPM to stratify fungal burden into cohort-based categories. This approach circumvents the interplatform variability inherent in absolute read counts and establishes a reproducible, semi-quantitative framework for comparing tNGS results across laboratories. By integrating this standardized fungal-load metric with host clinical variables, we developed a model that achieved an AUC of 0.966 and provided individualized risk estimates through a visual nomogram. The nomogram enables clinicians to translate multivariable probabilities into actionable decisions, facilitating early initiation or de-escalation of antifungal therapy based on patient-specific risk profiles. Collectively, these advances extend the clinical interpretability of tNGS data, bridging the gap between molecular signal quantification and bedside decision-making. A nomogram score exceeding the validated threshold should prompt strong consideration of antifungal therapy, whereas a lower score may support treatment de-escalation in the appropriate clinical context, in conjunction with ongoing clinical reassessment.

Despite its robust multicenter design and rigorous analytical validation, this study has several limitations. First, the retrospective design may have introduced inherent selection and information biases, and the timing of BALF collection and antifungal exposure was not standardized across centers. Prior antifungal therapy may have confounded the observed relationship between fungal load and IPA risk, underscoring the need for prospective studies with standardized pre-sampling treatment windows. Second, although internal bootstrapping confirmed model robustness, external validation using prospectively enrolled, geographically diverse cohorts and alternative sequencing platforms is necessary to establish generalizability. Third, the diagnostic classification of IPA was based on consensus definitions rather than histopathologic confirmation in all cases, which may have led to the misclassification of borderline presentations. Fourth, the quantitative interpretation of tNGS reads remains partly assay-dependent; despite normalization by RPM, residual interplatform variability and batch effects cannot be entirely excluded. The absence of data on fungal viability and precise timing of antifungal therapy relative to BALF sampling further limits interpretation; prior antifungal exposure may reduce detectable fungal DNA, resulting in load misclassification and attenuation of its association with IPA risk. In addition, the percentile-based fungal-load categorization, while pragmatic, requires validation in external cohorts. Fifth, several predictors incorporated into the model, most notably ICU admission, corticosteroid exposure, and bacterial co-infection, are clinically correlated. Although stepwise selection and bootstrap resampling were applied to enhance model stability, and formal collinearity diagnostics demonstrated no evidence of problematic multicollinearity (all variance inflation factors < 3.5), residual clinical interdependence among predictors cannot be fully excluded. Accordingly, the large odds ratios observed for diabetes mellitus and corticosteroid exposure should be interpreted with caution, as they may partly reflect cohort-specific risk profiles and may not be directly transferable to other populations. Although the Hosmer–Lemeshow test suggested good overall calibration, this test has limited sensitivity to detect overfitting. Indeed, bootstrap-derived calibration slopes indicated moderate optimism in the original model, which improved after application of Firth penalized regression, highlighting the importance of cautious interpretation of absolute risk estimates. Finally, because this study was conducted in tertiary hospitals in China, the findings may not be directly generalizable to other healthcare systems or patient populations with differing pathogen spectra and antifungal exposure patterns. Future prospective multicenter investigations incorporating serial sampling, quantitative fungal viability assessments, and cross-platform standardization are warranted to refine and externally validate the proposed diagnostic model.

This multicenter study established and internally validated an integrated diagnostic model that combines BALF-tNGS data with key host clinical factors to distinguish IPA from airway colonization in non-neutropenic patients. The model demonstrated excellent discrimination and clinical utility, outperforming conventional microbiologic and radiologic criteria. Normalized Aspergillus reads derived from tNGS correlated with IPA risk but required contextual interpretation alongside host determinants such as diabetes, corticosteroid exposure, and bacterial co-infection. By providing a standardized, semi-quantitative framework for interpreting Aspergillus molecular signals, this approach bridges the gap between molecular diagnostics and bedside decision-making. Prospective multicenter validation is warranted to confirm its applicability across diverse populations and sequencing platforms.