Forecasting symptoms of pollen-related allergic rhinoconjunctivitis at the level of individual patients would be useful to improve disease control and plan pharmacological intervention. Information Technology nowadays facilitates a more efficient and easier monitoring of patients with chronic diseases. We aimed this study at testing the efficiency of a model to short-term forecast symptoms of pollen-AR at the "individual" patient level. We analysed the data prospectively acquired from a group of 21 Italian children affected by pollen-related allergic rhinoconjunctivitis and recorded their symptoms and medication "Average Combined Score" (ACS) on a daily basis during April-June 2010-2011 through an informatics platform (Allergymonitor™). The dataset used for prediction included 15 variables in four categories: (A) date, (B) meteo-climatic, (C) atmospheric concentration of 5 pollen taxa, and (D) intensity of the patient's IgE sensitization. A Partial Least Squares Discriminant Analysis approach was used in order to predict ACS values above a fixed threshold value (0.5). The best performing predicting model correctly classified 77.8% ± 10.3% and 75.5% ± 13.2% of the recorded days in the model and test years, respectively. In this model, 9/21 patients showed ≥ 80% correct classification of the recorded days in both years. A better performance was associated with a higher degree of patient's atopic sensitization and a time lag > 1. Symptom forecasts of seasonal allergic rhinitis are possible in highly polysensitised patients in areas with complex pollen exposure. However, only predictive models tailored to the individual patient's allergic susceptibility are accurate enough. Multicenter studies in large population samples adopting the same acquisition data system on smart phones are now needed to confirm this encouraging outcome.
Keywords: Allergic rhinitis; IgE, allergens; Partial Least Squares Discriminant Analysis; allergenic molecules; climate; electronic and technology; forecasting models; meteorology; patient-specific models; pollen; symptoms-andmedication score; time lag models.