Sarcouncil Journal of Engineering and Computer Sciences

Abstract: The promise of Artificial Intelligence transforming global health hinges on the availability of representative data. However, a stark imbalance exists: while some populations generate vast amounts of digitized health information ("data floods"), others remain largely invisible in digital health ecosystems ("data deserts"). This study provides a comprehensive analysis of this global disparity through the development of the Health Data Density Index (HDDI), a novel measurement framework that quantifies health data availability across five dimensions: health system digitization, population health surveillance capacity, clinical research infrastructure, genomic and biobanking resources, and digital connectivity. Using expert panel consensus through modified Delphi techniques and data from international health organizations including WHO Global Health Observatory, World Bank Health Statistics, and ITU Telecommunications databases, the HDDI reveals profound global disparities. North America, Western Europe, and parts of East Asia demonstrate consistently high data density scores, while sub-Saharan Africa, parts of South Asia, and Central Asia exhibit the lowest levels globally. The study introduces the Health Data Poverty Trap Model, a formalized theoretical framework explaining how regions become locked in either high-equilibrium or low-equilibrium states through five interconnected feedback mechanisms. Quantitative analysis across four clinical domains demonstrates systematic algorithmic performance degradation: diabetic retinopathy screening algorithms achieve 94.2% sensitivity in data flood regions but only 58.4% in data deserts, translating to 261 additional missed diagnoses per 100,000 patients. Sepsis prediction models similarly deteriorate from AUROC 0.91 in high-density regions to 0.66 in data deserts. These algorithmic deterioration gradients follow mathematical functions with domain-specific parameters, revealing that AI deployment below critical HDDI thresholds may cause more harm than benefit. The profound data imbalance poses fundamental threats to health AI equity, as models trained predominantly on data from "flooded" regions risk poor generalizability and may perpetuate or exacerbate health disparities when applied globally. Breaking these poverty trap dynamics requires coordinated interventions exceeding critical investment thresholds across multiple system components simultaneously. The findings underscore the urgent need for global health AI policies that prioritize data equity—investing in data infrastructure and responsible data generation in underserved regions—as a prerequisite for developing fair, equitable, and truly global AI-driven health solutions.