Implementation of AESOP early-warning system for respiratory disease: a pilot and validation study using routinely collected data in Amazonas, Brazil.

Background: Respiratory infections cause major morbidity and mortality globally, highlighting the need for robust early-warning systems (EWS). The Alert-Early System of Outbreaks with Pandemic Potential (ÆSOP) was co-developed with surveillance stakeholders to detect outbreaks using administrative primary health care (PHC) data. We aimed to report ÆSOP's feasibility and performance for detecting influenza-like illness (ILI) outbreaks in a real-world context and to document lessons learned from translating research into routine health services. Methods: The pilot was conducted in Amazonas, Brazil, from April to July 2024, covering 62 municipalities. ÆSOP relies on anomaly detection models applied to weekly counts of ILI-related PHC encounters, and municipal-level warnings are issued when thresholds are exceeded. Local health authorities validated warnings through a structured questionnaire. Performance was evaluated by sensitivity, specificity, positive and negative predictive value (PPV and NPV). Findings: During the study period, 1.8 million PHC encounters were reported, of which 6.6% were ILI-related. ÆSOP issued 104 warnings across 41 municipalities. Sensitivity was 62.8% (95% CI 54.4%-70.5%), specificity 95.5% (95% CI 93.1%-97.1%), PPV 82.7%, (95% CI 74.3%-88.8%), and NPV 88.1% (95% CI 84.7%-90.9%). In 72% of confirmed outbreaks, ÆSOP provided the first signal to authorities, and in 56%, warnings supported responses including enhanced risk communication, equipment/test deployment, and strengthening of intergovernmental coordination. Interpretation: ÆSOP anticipated outbreaks in real-world usage and was validated by authorities as a decision-support tool. Its reliance on administrative data and open-source technologies enhances scalability and adaptability across health systems globally. Funding: Rockefeller Foundation (award 2023 PPI 007).