From Cases to Consumption: Evaluating Forecasting Model Portability for Public-Health Supply Chains
From Cases to Consumption: Evaluating Forecasting Model Portability for Public-Health Supply Chains
Conference: LOM Section Annual Conference (LOMSAC) 2026, Cardiff Business School.
Context
This work is motivated by a persistent disconnect in public health analytics between what is forecast and what is procured. Platforms such as DHIS2 and CHAP have substantially advanced the forecasting of disease incidence, often incorporating climate and environmental signals to provide early warning of outbreaks.
Yet, operational decisions do not act on cases alone. Health systems must translate predicted morbidity into product requirements such as vaccines, antimalarials, and diagnostics. At this point, forecasting performance often deteriorates. Consumption data are incomplete, irregular, and frequently censored by stockouts, reporting delays, and supply interruptions, making direct modelling of demand both unreliable and inconsistent across settings.
This presentation examines forecasting model portability across this divide. We ask whether models developed and validated for disease cases can be meaningfully transferred to forecast health commodity consumption, and under what conditions this transfer fails. Using routine surveillance and logistics data, we compare three strategies: direct reuse of case-based models, targeted fine-tuning using supply-chain covariates such as campaign timing or climate signals, and full retraining on consumption data.
The contribution is practical rather than purely methodological. We provide evidence on when portability is viable, when it introduces systematic bias, and how forecasting choices propagate into inventory decisions. By linking predictive performance to downstream ordering and stock availability, the work supports a shift from reactive replenishment to anticipatory logistics. This is particularly relevant for climate-sensitive diseases, where early warning only delivers value if commodities arrive before demand materialises.