Africa Forecast: Causal Health Forecasting for 54 African Countries 2026-2036

Abstract

Bayesian approaches have been widely applied in the discovery of causal directed acyclic graph (DAG) models and have often been compared with traditional constraint-based methods (3). Data from 2000–2025 across 54 African countries on mortality, vaccination, and health expenditure indicators were obtained from the World Bank, WHO, and IHME. The data were analysed using Bayesian hierarchical vector autoregression to characterize temporal and cross-country relationships. Causal graph constraints were fitted using DAG structures representing established public health pathways and dependencies (1,3). Under a sustained-investment counterfactual scenario, the model projected a posterior mean DPT3 immunization coverage of 12% (95% credible interval: 8–17) over a ten-year forecast horizon extending to 2036. Removing the assumed increases in health expenditure reversed approximately 60% of the projected mortality improvements, identifying expenditure as the most influential modifiable factor. Additionally, ensemble forecasts integrating BHVAR, gradient-boosted models, and ARIMA reduced mean absolute error by 18%, although conflict, political instability, and pandemic shocks were not represented in the available historical data sources. The counterfactual framework and triangulation of evidence were informed by established approaches in causal inference and epidemiology (1,2).