Algorithmic Fairness in Clinical Machine Learning for Non-Communicable Disease Prediction in Sub-Saharan Africa: A Systematic Review, Empirical Analysis, and Research Agenda

Non-communicable diseases constitute the leading and most rapidly accelerating cause of preventable mortality across sub-Saharan Africa, yet the clinical artificial intelligence literature has overwhelmingly neglected both the African epidemiological context and the algorithmic fairness implications of deploying machine learning models in these settings. This paper presents the first systematic review and empirical analysis of algorithmic fairness in clinical machine learning for NCD prediction specifically oriented toward sub-Saharan Africa. Following a PRISMA-inspired protocol, 47 studies published between 2015 and 2026 were identified, screened, and analyzed. Key finding: only 8 of 47 studies (17.0%) included any form of fairness evaluation, only 3 (6.4%) employed formal fairness metrics, and only 12 (25.5%) were conducted on African patient data. The systematic review is enriched by original empirical evidence from the author's prior peer-reviewed studies on Type 2 diabetes prediction (AUC-ROC 0.8226, significant BMI subgroup recall disparities) and cardiovascular disease prediction (AUC-ROC 0.9270, 5.92 percentage point female recall deficit), forming the first multi-disease empirical fairness dataset for NCD prediction in West Africa. A five-pillar research agenda is proposed addressing the data representation crisis, the fairness measurement gap, the explainability requirement, the deployment infrastructure barrier, and the sex and demographic bias emergency in clinical AI. This is the third paper in the Nexora AI for African Health Research Program. Paper 1 (Diabetes):https://doi.org/10.5281/zenodo.20348652 Paper 2 (CVD):https://doi.org/10.5281/zenodo.20373324