A Framework for Context-Adaptive Artificial Intelligence Clinical Decision Support in Low-Resource Nigerian Hospital Settings

Artificial intelligence clinical decision support systems developed and validated in high-income country hospital environments embed implicit assumptions about data availability, infrastructure reliability, clinical workflow, and patient population characteristics that fail systematically when these systems are deployed in low-resource Nigerian hospital settings, creating a theoretical and engineering gap that limits beneficial AI adoption in Nigerian healthcare. This dissertation develops an original theoretical and engineering framework for context-adaptive AI clinical decision support specifically designed for low-resource Nigerian hospital deployment. The framework makes three original theoretical contributions. The first is a clinical context characterization theory specifying a multidimensional context space encompassing data quality and completeness dimensions, infrastructure reliability dimensions, workforce competency dimensions, and patient population characteristics, and formalizing the mathematical relationship between context space position and required AI system adaptation. The second is an adaptive model architecture theory specifying how AI model complexity, feature requirements, and output uncertainty quantification should adapt dynamically to context space position, including degradation pathways that maintain clinically useful predictions under data-poor conditions rather than failing entirely. The third is a human-AI interaction framework theory specifying how decision support presentation should adapt to clinician AI literacy, cognitive load, and available response time in Nigerian hospital conditions. Empirical validation is conducted through development of context-adaptive AI systems for sepsis early warning and chest X-ray triage across four hospitals spanning high and low resource levels in Lagos, Kano, and Borno states. Comparative analysis against non-adaptive counterpart systems confirms that context-adaptive systems maintain significantly higher clinical utility scores across the full resource level spectrum. Keywords: artificial intelligence, clinical decision support, context-adaptive, low-resource settings, Nigeria.