Authors: Mohseni, M1., Rai, TS1., Wang, H2., Zheng, H2., Watterson, S1.

Affiliations: 1 Personalised Medicine Centre, School of Medicine, Ulster University, Derry/Londonderry, BT48 7JL. 2 School of Computing, Ulster University, Belfast, BT15 1AP

Background/ Introduction: Cardiovascular diseases (CVD) remain the leading cause of death in Europe, accounting for up to half of all annual mortalities and exhibiting profound East-West regional inequalities. Effective public health intervention requires timely, granular mortality surveillance across diverse geographic regions. However, exploiting massive, harmonized regional databases typically demands specialized coding and statistical expertise, creating labour-intensive bottlenecks in traditional epidemiological workflows.

Material & Methods: We developed the “Eurostat Mortality Explorer,” an advanced multi-agent AI system designed to automate public health analysis. Powered by a Large Language Model (LLM) utilizing an iterative Observation-Thought-Action reasoning loop (ReAct paradigm), the framework integrates a Retrieval-Augmented Generation (RAG) module to semantically map natural language queries directly to precise ICD-10 codes and NUTS geographical regions. To validate the system, the AI autonomously analysed the Eurostat health database (dataset hlth_cd_asdr2), processing 398,544 records across 452 NUTS2 regions in 37 countries spanning 2011–2022. Further local system validation workflows were assessed against localized regional registries referencing in my poster.

Results: The autonomous framework successfully executed data processing, statistical modelling, and geographic visualization in under 5 minutes, representing a >95% reduction in analysis time. The agent identified an overall regional mean CVD death rate of 392.67 per 100,000. It independently detected a statistically significant pre-pandemic declining trend of -4.39 deaths per 100,000 per year ($p < 0.001$, a 9.14% overall reduction from 2011–2021), followed by automated detection of a significant pandemic-era trend reversal (+2.82% increase during 2020–2021). Furthermore, the agent exposed severe geographic disparities, mapping a 7.83-fold mortality gradient between Western regions (under 200 per 100,000) and high-risk Eastern European regions (exceeding 1,000 per 100,000).

Conclusion: Agentic AI systems can autonomously conduct rigorous, publication-quality epidemiological surveillance. By lowering technical barriers and enabling real-time, natural language interrogation of massive datasets, this framework establishes a new paradigm for democratized public health intelligence and rapid health surveillance.