Opportunities, implementation models, and clinical impact of federated learning in African healthcare systems

Africa bears a disproportionate share of the global burden of disease while many health systems continue to operate with shortages of specialist staff, uneven diagnostic capacity, fragmented digital infrastructure, and limited interoperability across institutions [1-4]. At the same time, digital health, artificial intelligence (AI), and machine learning (ML) are increasingly being used across the continent for patient management, laboratory medicine, telemedicine, quality assurance, and disease monitoring, indicating both the need for and the feasibility of data-enabled innovation in resource-constrained settings [1-4]. However, the centralized data aggregation paradigm underlying much conventional ML is often poorly matched to African healthcare contexts. Clinical data are frequently siloed within hospitals, laboratories, ministries, and disease programs operating under heterogeneous technical and regulatory conditions. Cross-institutional and cross-border pooling may be limited by unreliable connectivity, cost of data transfer, fragmented governance arrangements, and concerns around data sovereignty and institutional control [4-6]. Federated learning (FL) has emerged as a promising alternative because it allows multiple institutions to collaboratively train a shared model while keeping raw data at the point of origin. In its typical form, only model parameters, gradients, or related updates are exchanged, which may reduce privacy risks and preserve local control over sensitive data [7,8]. For settings where centralized infrastructure is either infeasible or undesirable, FL offers a technically and politically attractive model for collaborative AI development. Outside Africa, landmark studies have shown that FL can yield clinically useful models with performance comparable to, and sometimes better than, single-site or pooled-data approaches. Examples include multinational work on COVID-19 outcome prediction and multi-institutional brain tumor segmentation [9,10]. Africa-specific implementations are now beginning to appear, including federated chest imaging for tuberculosis across multiple African countries, fetal plane classification on low-resource edge devices, HIV-related predictive modeling, and cross-border telemedicine systems using differential privacy [5,11-13]. Despite this momentum, important questions remain unresolved. Reported deployments differ substantially in architecture, bandwidth requirements, hardware availability, privacy guarantees, validation strategy, and operational maturity. Fairness and inclusion are also emerging concerns, because poorly representative multi-site training can reproduce or amplify existing health inequities [5,8,14]. In addition, the African regulatory landscape for health data and AI remains uneven, making governance and cross-border implementation especially important for interpretation and scale-up. To date, no Africa-focused systematic review has comprehensively synthesized the empirical evidence on FL implementation in healthcare settings across the continent. Existing reviews are usually global in scope or conceptual rather than specific to African deployment realities [7,8]. A focused synthesis is therefore needed to map where FL has been implemented, how it has been operationalized, how it performs relative to alternatives, what contextual barriers and enablers have been reported, and where critical evidence gaps remain. The present protocol was developed to address that gap using a PRISMA-P-compliant approach [15,16].