An Adaptive Dual-Loop Artificial Intelligence Framework for Integrated Disease Surveillance and Health Workforce Learning in Low-Resource Health Systems
Abstract
Background: In many low-resource health systems, disease surveillance and workforce training run on separate rails, so outbreaks are caught late, and frontline capacity stays thin. Most artificial intelligence (AI) tools pick one side or the other; almost none couple them. Methods: We built and tested an Adaptive Dual-Loop AI Framework. One loop, for epidemiological intelligence, handles multi-source ingestion, anomaly detection, and time series forecasting; the other, for workforce learning, runs a conversational AI tutor, contextual decision support, and interaction logging. Bidirectional feedback links them. We evaluated the system with a mixedmethods, implementation-science design in 24 facilities and 312 health workers across Nairobi (urban) and Kajiado (rural) counties, AQ1 Kenya, over 10 months, benchmarking the forecaster against ARIMA and rule-based baselines with rolling-window cross-validation on AQ2 Integrated Disease Surveillance and Response and DHIS2 data. Results: The dual-loop model reached an area under the curve of 0.94 (95% CI 0.91–0.96), against 0.81 for ARIMA and 0.72 for threshold rules. Mean lead-time gain was 3.6 days across cholera, malaria, and respiratory infections, and the outbreak-response cycle shortened from 12.9 to 6.2 days (52%). Competency climbed from 55.0% to 75.4% (p < 0.001), engagement tracked that gain (r = 0.76), and reporting completeness rose from 62% to 92%. Conclusion: Coupling prediction with embedded learning delivered gains neither half reached alone, a practical route to adaptive public-health intelligence where resources are scarce. Received: 7 May 2026 | Revised: 13 July 2026 | Accepted: 12 August 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The de-identified analytical datasets, model training and evaluation code, and figure-generation scripts are available on reasonable request to the corresponding author. Raw routine-surveillance data from DHIS2 and IDSR are governed by the Kenya Ministry of Health and available on request with MoH approval. Because these records contain sensitive, potentially reidentifiable data, their use is bound by the Kenya Data Protection Act (2019), the study’s ethical approval (KNH/UoN-ERC/A/0096-2024), and data-sharing agreements. Under managed access, the code, figure-generation scripts, environment specification, and a synthetic dataset that regenerates every reported figure and table are provided to bona fide requesters, so the pipeline can be verified even where the underlying records cannot be redistributed. Author Contribution Statement Kenneth Goga Riany: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration, Funding acquisition. Agnes Linus Muthoni: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration, Funding acquisition. Marcellah Eucabeth Onsomu: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration, Funding acquisition. Pauldy C.J. Otermans: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration, Funding acquisition. Dev Aditya: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration, Funding acquisition.
