ABSTRACT
Background
Symptom monitoring can improve adherence to daily medication. However, controlled clinical trials on multi-modular allergy apps and their various functions have been difficult to implement.
 |
| Graphical Abstract |
Methods
We performed a stratified, controlled intervention study (May–August 2023) with grass pollen allergic participants (N = 167) in Augsburg, Germany.
Participants were divided into three groups, each receiving the same allergy app, but with increasing numbers of functions. Primary endpoint: rhinitis-related QoL; Secondary endpoints: symptom scores, relevant behavior, self-reported usefulness of the app, symptom forecast.Results
Rhinitis-related QoL was increased after the intervention, with no statistical inter-group differences. However, participants with access to the full app version, including a pollen forecast, took more medication and reported lower symptoms and social activity impairment than participants with access to a reduced-function app. Using an XGBoost multiclass classification model, we achieved promising results for predicting nasal (accuracy: 0.79; F1-score: 0.78) and ocular (accuracy: 0.82; F1-score: 0.76) symptom levels and derived feature importance using SHAP as a guidance for future approaches.
Conclusion
Our allergy app with its high-performance pollen forecast, symptom diary, and general allergy-related information provides a clinical benefit for allergy sufferers. Reliable symptom forecasts may be created given high-quality and high-resolution data.
No comments:
Post a Comment