Etiologic diagnosis of seasonal allergic rhinitis supported by artificial intelligence: The @IT-2020 project

Résumé fourni par la source

Background A precise etiological diagnosis of seasonal allergic rhinitis (SAR) is essential for a tailored prescription of its only curative treatment, allergen-specific immunotherapy (AIT). This is a challenging task in temperate climates, where most patients are polysensitized to multiple pollen with overlapping seasons. Objective The study aims to develop a modular, flexible and validated Clinical Decision Support System (CDSS) generated with Artificial Intelligence for the etiologic diagnosis of SAR. Methods In the context of the @IT-2020 Project, we developed a CDSS for SAR etiological diagnosis. We aimed to automate the CDSS by integrating Machine Learning (ML) algorithms. The CDSS adopts three progressive diagnostic modules: (a) clinical history and Skin Prick Tests (SPT), (b) plus molecular specific Immunoglobulin E (sIgE mol ) tests, (c) plus an electronic clinical and environmental Diary. To this end, three raters identified, following international guidelines and a Delphi-like procedure, the culprit pollen on 100 SAR patients in Rome (Italy). Results Three models best performing (AUROC >95%) have been then generated by ML training and tested on 2/3 and 1/3 patients, respectively. The validity of the three models was confirmed by: (A) contextual adaptability: as the models' performance was linked to the patients complexity; (B) interpretability: as both, clinical components and sensitization pattern, influenced the models' diagnostic decision (SHAP analysis); (C) geographical generalizability: given the models' equal high performance (AUROC >95%) in diagnosing 92 SAR patients in Tirana (Albania); (D) temporal generalizability: as a high performance was obtained by reducing the patients' monitoring period to 45 days; (E) Allergen Immunotherapy (AIT) prescription adaptability: as the models quite well reproduced the gold standard AIT prescriptions; (F) human-plus diagnostic capability: as the models outperformed 24 doctors in a competition evaluating 12 patients. Conclusion – In our Proof-of-Concept study, a modular Clinical Decision Support System, based on Machine Learning, replicated with enough reliability raters' diagnosis and AIT prescription in patients with Seasonal Allergic Rhinitis. Further studies are needed not only to target independent replication, but also to prospectively test whether an AI-driven CDSS may be useful to initiate personalized treatment and ultimately improve SAR disease control.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.