ClimAIr tests privacy-preserving AI for environmental and respiratory health research

This report from the ClimAIr project marks an important early step toward studying how environmental conditions may link to non-communicable respiratory diseases across Europe. It focuses on developing and testing an artificial intelligence approach that can analyze complex patterns involving air pollution, weather, and pollen while preserving sensitive health information.

The work provides a secure basis for collaboration between healthcare-related organizations without requiring them to share individual patient records. It enables the project to prepare rigorous methods for future real-world research on environmental effects on respiratory health.

The report aims to establish and validate the first version of ClimAIr’s AI-based correlation analysis pipeline. It first focuses on allergic rhinitis, a respiratory condition whose symptoms can be affected by multiple environmental, biological, and behavioral factors. Rather than aiming for immediate clinical conclusions, this part was designed to confirm that the planned analytical methods can operate reliably in a distributed healthcare setting and support later, evidence-based investigations using harmonised clinical data.

The project first used a simulated network representing multiple healthcare centers. No real patient records were processed. It then combined privacy-preserving federated learning with complementary analytical methods. Federated learning lets participating organizations train and assess shared AI models while keeping sensitive records within their own systems. The analysis also accounts for both broad statistical relationships and more complex time-related patterns, helping researchers examine natural influences from different perspectives.

The main outcome is a functioning end-to-end framework for collaborative analysis that preserves data privacy. Testing showed that the framework can support alternative AI approaches, assess patterns across time, and produce aggregated results without centralizing individual-level health information. The work also demonstrated the value of combining different analytical views: simpler methods can identify broad associations, while AI models can help explore more complex and delayed patterns in environmental exposure. These results are qualitative demonstrations of technical feasibility only. Because the work used simulated information, it does not provide epidemiological evidence or establish actual health effects.

The report represents a key contribution to ClimAIr, as it establishes the methodological groundwork required for the next stages of research. It improves the project’s capacity to implement privacy‑preserving AI techniques on real‑world data when accessible, thereby advancing studies on respiratory diseases and the environmental factors that may affect their development.