Data Management Strategy for AI Deployment in Ethiopian Healthcare System

Abstract

This paper reviewed data management practices in healthcare system in Ethiopia. Through the lens of strategic approach for good data management practices, the extent to which healthcare data is leveraged for AI applications was assessed. As the main body for collecting, archiving, managing, and sharing health data in Ethiopia, the national data management center for health (NDMC) at the Ethiopian public health Institute (EPHI) is considered for this study. To guide the study, the FAIR (Findability, Accessibility, Interoperability, and Reusability) principles; the key elements of data strategy by Inner City Fund (ICF) International Inc: data governance, data architecture, needs assessment, future state vision and roadmap, organizational analytics maturity assessment; and data from Papers with Code (data in Meta Artificial Intelligence (AI) Research), more specifically, State-of-the-Art (SOTA) benchmarks, AI models, use cases, libraries, and frameworks, were used to assess the extent to which NDMC data activities support de-ployment of AI applications. From the analysis of primary and secondary sources, as well as based on the principles and frameworks, the following key findings were reported: NDMC by and large pursues data strategy as outlined by ICF International, except two key elements that could not be detected from the review and an in-person discussion and these are: future state vision and roadmap, organizational analytics maturity assessment. When the FAIR principle is used as a guide, the data and metadata at NDMC do not strictly adhere to community standards in bio-medical sciences that promote findability, accessibility, interoperability, and reusability. In regard to the SOTA benchmarks, NDMC, data analytics and reporting being one of its core charges, mainly relies on descriptive and inferential statistical methods to generate data reports and no AI, Machine Learning (ML), or Deep Learning (DL) applications were identified that could offer more predictive and prescriptive insights.