10th U6+ Consortium of African Universities International Conference

Under-five mortality in the Lake Victoria basin counties of Siaya, Kisumu, Homa Bay, Migori, Busia and Kakamega remains among the highest in Kenya, driven by malaria, pneumonia, sepsis, and severe malnutrition, yet community health surveillance in this region remains largely reactive, paper-based, and linguistically inaccessible to the Swahili-speaking community health workers who deliver frontline care. Two decades of demographic surveillance data already exist for this region, but no analytical system converts this longitudinal evidence into household-level, real-time, causally grounded, and explainable guidance that a community health worker can act upon before a child deteriorates. This paper presents MtotoSalama, a quantum-enhanced causal artificial intelligence framework integrating Bayesian hierarchical survival modelling, triangulated causal inference, Swahili-English natural language processing, and quantum-classical resource optimisation within a single reproducible, DHIS2-interoperable architecture. A four-level hierarchical Weibull survival model, an XGBoost classifier, a Random Survival Forest, difference-in-differences, regression discontinuity, and inverse probability weighting designs, a retrieval-augmented Swahili clinical assistant, and a Quantum Approximate Optimisation Algorithm for community health worker allocation were implemented and validated end-to-end on a simulated cohort structurally calibrated to historical surveillance patterns, pending finalisation of live data access. The classifier achieved an AUROC of 0.81, the survival ensemble a concordance index of 0.77, and the Bayesian model converged with all Gelman-Rubin statistics below 1.01; a fairness audit flagged reduced precision among HIV-exposed infants and infants under one year, and the quantum allocator reached solutions within 3.2 percent of the classical optimum at small scale. The framework demonstrates technical feasibility for translating accumulated surveillance data into actionable, equity-audited intervention guidance. Deployment should proceed only after live-data replication and targeted recalibration of the flagged subgroups.