Implementing a Privacy-Preserving Learning System for Paediatric Asthma Management

Introduction: Digital monitoring of asthma in children is becoming a central element of childcare, but the security of data and parental consent combined with algorithm transparency means that digital monitoring is not widely used. Privacy-preserving learning systems provide a chance to utilise clinical data in a real-world environment without breaching the confidentiality of patients. This paper examines the application of a federated, privacy-safe learning system that is aimed at providing better asthma control in children with high predictability without compromising the protection of their data.

Methods: In three paediatric clinics, an experimental federated learning (FL) system was implemented. It allowed locally training machine learning models on device-level spirometry, symptom diaries, and medication-use data. To reduce the risk of re-identification, the methods of differential privacy and secure aggregation were combined. Clinician surveys, parental feedback forms, and system-level measures were used to evaluate implementation feasibility, model performance, user acceptability, and system usability over 12 weeks.

Results: The FL system was found to be a model with an accuracy of 82 percent in predicting the likelihood of early exacerbation, similar to centralised models, with a greater level of data protection. Clinicians said that they had more confidence in data-driven decision support (78%), and parents had high confidence in privacy protection (84%). The uptime of the system was 96 and training cycles were performed within the usual clinically acceptable intervals allowed. There were cases of no data-leakage or privacy violations.

Conclusions: A privacy-friendly learning system is achievable, acceptable, and efficient in the management of asthma in children. This model strikes a balance between the requirement to have a strong clinical decision support system and high-level data confidentiality, and it represents a scalable way of incorporating open and safe digital tools into the work of paediatric care.