AI-derived Histopathological Signatures Are Associated With Response to Mepolizumab in Chronic Rhinosinusitis With Nasal Polyps
Jens Tidemandsen, Simon Høj, Thomas Hl Jensen, Martin Stampe, Anne Sophie Homøe, Marie Høxbro, Kasper Aanæs, Sarah Charabi, Vibeke BackerBackground
Chronic rhinosinusitis with nasal polyps (CRSwNP) is a heterogeneous inflammatory disease with variable response to biologic therapy. Tissue-based predictors of treatment response are limited, and conventional eosinophil-focused histopathology has shown inconsistent predictive value.
Objective
To assess whether artificial intelligence (AI)-derived baseline histopathology of nasal polyps contains information associated with clinical response to mepolizumab in CRSwNP.
Methods
This hypothesis-generating substudy included patients with severe CRSwNP enrolled in a randomized controlled trial. Baseline biopsies were analysed using an AI-driven spatial histopathology pipeline extracting 84 morphological features per cell. Features were evaluated in eosinophil-only, noneosinophil, and all-cell configurations. Treatment response at 6 and 12 months was defined using EUFOREA criteria. Linear discriminant analysis and partial least squares regression were applied.
Results
Fifty-eight patients were included. At 12 months, the eosinophil-only model showed limited discrimination (AUC ≈ 0.43), the noneosinophil model moderate discrimination (AUC ≈ 0.62), and the all-cell model the highest performance (AUC ≈ 0.75). At 6 months, all models showed limited discrimination (AUC ≈ 0.49). For continuous outcomes, noneosinophil models explained the greatest variance (R 2 up to ≈ 0.34).
Conclusion
AI-derived baseline histopathology contains information associated with clinical response to mepolizumab in CRSwNP. Models incorporating the broader tissue microenvironment outperformed eosinophil-focused approaches, supporting the potential of computational pathology for pretreatment stratification of biologic therapy in CRSwNP.