Explore the scholarly record.

Abstract

Abstract Background Concurrent COVID-19 and influenza A(H1N1) epidemics since late 2019 have complicated early differentiation because of overlapping clinical and imaging features. Viral evolution, shifting population immunity, and emerging variants may have altered imaging phenotypes, underscoring the need for contemporary data. We aimed to compare clinical, laboratory, and CT characteristics between COVID-19 and H1N1 pneumonia, identify independent predictors for each diagnosis, and explore machine learning (ML) models for differentiation. Methods This retrospective study included 612 patients (442 H1N1, 170 COVID-19) diagnosed between January 2023 and December 2025. Clinical, laboratory, and chest CT features were comprehensively analyzed. Between-group differences were assessed by χ² test or Mann-Whitney U test. Multivariable logistic regression was performed to identify the independent predictive factors. Least absolute shrinkage and selection operator regression was used for feature selection. Five ML algorithms were trained (70% of cohort) and tested (30%). Model performance was assessed using area under the receiver operating characteristic curve (AUC). Results Parenchymal involvement was more frequent in COVID-19 than in H1N1 pneumonia (91.2% vs 74.7%; P <  0.001), with higher rates of ground‑glass opacities, interlobular septal thickening, and crazy-paving pattern. Airway involvement was more frequent in H1N1 pneumonia (48.2% vs 34.7%; P =  0.003), including higher rates of bronchial wall thickening, centrilobular nodules, and tree‑in‑bud nodules. On multivariable analysis, upper lobe involvement (OR = 2.19; 95% CI: 1.04, 4.65) and parenchymal involvement (OR = 3.44; 95% CI: 1.67, 7.10) were independent predictors of COVID-19. Tree-in-bud nodules (OR = 0.30; 95% CI: 0.12, 0.77), lymphadenopathy (OR = 0.29; 95% CI: 0.06, 0.81), and disorders of consciousness (OR = 0.20; 95% CI: 0.07, 0.61) were independently associated with H1N1 pneumonia. Among five ML models, the random forest model achieved the highest performance, with an AUC of 0.71 (95% CI: 0.58, 0.85) for differentiation. Conclusion Upper lobe involvement and parenchymal involvement are independent predictors of COVID-19, whereas tree-in-bud nodules, lymphadenopathy, and disorders of consciousness are independently associated with H1N1 pneumonia. These imaging and clinical features may facilitate early differential diagnosis.