Using crowd-sourced allergic rhinitis symptom data to improve grass pollen forecasts and predict individual symptoms

Area of Science:

• Environmental Health
• Epidemiology
• Computational Biology

Background:

• Seasonal allergic rhinitis (AR), or hay fever, is a widespread respiratory condition influenced by environmental factors like pollen.
• Previous research established links between community AR symptoms, collected via mobile apps in Australian cities, and environmental variables.
• This study extends prior work by developing predictive models for AR risk and ambient pollen concentrations.

Purpose of the Study:

• To assess the accuracy of models forecasting individual AR risk for the next day.
• To evaluate models that nowcast ambient grass pollen concentrations using crowd-sourced AR symptom data.
• To determine if crowd-sourced symptom data can improve pollen forecasting and personalized AR risk prediction.

Main Methods:

• Developed models to forecast individual AR risk and nowcast ambient grass pollen using crowd-sourced symptom data.
• Grass pollen forecasts were categorized (low/moderate/high) based on daily symptom scores.
• Individual AR risk models utilized forward variable selection (environmental, demographic, behavioral, health data) and proportional-odds logistic regression.

Main Results:

• AR symptom-based grass pollen concentration estimates were more accurate than benchmark forecasting methods.
• Next-day AR symptom predictions were correct in 36% of cases and within one scale point in 82% of cases.
• Both prediction outcomes significantly outperformed chance, demonstrating the utility of crowd-sourced data.

Conclusions:

• Large-scale, crowd-sourced AR symptom data effectively predicts daily average grass pollen concentrations.
• The findings support the use of crowd-sourced data for personalized AR risk forecasting.
• This approach offers a valuable tool for managing seasonal allergic rhinitis.