A novel predictive tool for pulmonary aspergillosis in critically ill patients with viral pneumonia: a multi-center, ambispective cohort study

Background Viral pneumonia-associated pulmonary aspergillosis (VAPA) is a lethal complication in critically ill patients. We aimed to develop a generalizable and methodologically robust prediction tool for early VAPA risk, overcoming key limitations of existing models. Methods In a multi-center, ambispective cohort of 4,966 patients with influenza or COVID-19 admitted to ICU (Jan 2023–Jun 2025), we developed and validated two prediction models for VAPA using LASSO-selected variables: a standard Cox proportional hazards model and a Fine-Gray subdistribution hazards model accounting for non-VAPA death as a competing risk. For both models, restricted cubic splines were used to address nonlinear associations between continuous predictors and VAPA. Performance was assessed at 7 and 14 days post-ICU admission. Subgroup analysis was employed to examine the virus-specific models’ performance. Results The incidence of VAPA was 8.8%. The Fine-Gray model demonstrated excellent discrimination and calibration in both derivation and validation cohorts. In contrast, the standard Cox model substantially overestimated VAPA risk, particularly among high-risk patients (by 20-30%). Key predictors included haematological malignancy, halo sign, use of priority antibiotic combinations, SOFA score, neutrophil-to-lymphocyte ratio (NLR), and pre-ICU corticosteroid dose; the latter three exhibited J-shaped associations with VAPA risk. A significant interaction was observed between viral subtype and pre-ICU corticosteroid dose ( P = 0.006), with a stronger association in COVID-19 patients (sHR=1.17, 95% CI: 1.09-1.26). Overall model performance remained consistent across viral subtypes. An interactive web-based tool was then deployed, which generates both individualized VAPA risk estimates and corresponding clinical management recommendations. Conclusions In critically ill patients with viral pneumonia, a competing-risk model accounting for non-VAPA death provides more accurate estimates of VAPA risk than traditional approaches. The model demonstrates robust generalizability across influenza and COVID-19 pneumonia and offers a practical tool for early, personalized VAPA prediction.