Trustworthy machine learning for asthma and COPD recognition and prediction

Artificial intelligence is increasingly supporting healthcare across screening, diagnosis, and prognosis, but its use in real clinical pathways requires methods that remain robust, transparent, and reliable when applied to new patients. This is particularly relevant in chronic respiratory diseases, where asthma and COPD are high impact conditions with heterogeneous presentations and variable clinical trajectories, and where earlier recognition and risk awareness can improve follow up planning and clinical decision making.

The work applies machine learning to real world clinical cohorts in respiratory and allergy care, with the primary goal of recognising asthma and COPD and supporting clinically meaningful risk prediction. The modelling is designed to generalise to future patients rather than only performing well on a single cohort, and to provide outputs that are suitable for translational research and potential clinical decision support.

A central element is trustworthiness. Evaluation goes beyond average metrics to assess reliability, identify failure modes, and handle uncertainty explicitly. This includes calibrated predictions, uncertainty aware decision strategies, and transparent reporting that enables clinical discussion and supports validation activities.