Beyond prediction speed: evaluating the data-to-action gap in AI-based outbreak surveillance in conflict-affected health systems
Artificial intelligence (AI) is increasingly being explored to strengthen infectious disease surveillance and support earlier outbreak warnings. However, improved predictive performance does not necessarily translate into effective public health action. In conflict-affected health systems, reporting delays, incomplete data, population displacement, and uneven surveillance coverage may undermine the reliability and representativeness of outbreak alerts. Even when an alert provides a useful signal, limited diagnostic capacity, insecurity, workforce shortages, and weak coordination may delay or prevent an effective response. This Mini Review examines potential breakdowns between surveillance-data collection and public health action by synthesizing research on AI-enabled early-warning systems alongside contextual evidence from conventional surveillance programmes in humanitarian settings. It proposes a fourstage evaluation framework comprising (1) data readiness, (2) alert performance, (3) verification and decision-making, and (4) public health action and equity. The framework's contribution is to connect data quality and technical performance with the operational processes and distributional consequences that determine whether alerts can inform timely and equitable responses. It explicitly distinguishes direct evidence on AI performance from contextual evidence concerning conventional surveillance systems. The framework is conceptual and has not yet been prospectively validated; it should not be interpreted as an established causal pathway. Given the limited direct evidence linking AI-generated outbreak alerts to improved health outcomes in conflict-affected populations, future research should prospectively evaluate the complete alert-to-action pathway. Key indicators include reporting delay, data completeness and coverage, sensitivity, positive predictive value, false-alert burden, warning lead time, verification time, response initiation, and the distribution of investigations and interventions across populations and geographic areas. Evaluating these dimensions together may help determine whether AIenabled surveillance provides operational value beyond earlier prediction.
Authors
- Ilyas Abdullahi Khalif (ORCID: https://orcid.org/0009-0002-7421-8638)
- Abdishakur Ahmed Mahad (ORCID: https://orcid.org/0009-0008-2928-8382)
- Mohamed Mahdi Hashi (ORCID: https://orcid.org/0009-0005-2113-6753)
- Abdimalik Dahir Ali (ORCID: https://orcid.org/0009-0003-0828-3877)
- Abdisamad Abdihakim Aadam (ORCID: https://orcid.org/0009-0000-5120-6787)
- Ayuub Ahmed Osman (ORCID: https://orcid.org/0009-0000-7635-7522)
- Maida Ali Daahiye
Institutions
- Somali National University (SO)
- Salaam University (SO)
- Benaadir Research, Consultancy and Evaluation Center (Somalia) (SO)
Publication Details
- Journal
- Frontiers in Public Health
- Published
- 2026-11-17
- DOI
- https://doi.org/10.3389/fpubh.2026.2026163
- Primary Topic
- Data-Driven Disease Surveillance
- Type
- article
- Field-Weighted Citation Impact
- 0.00