Distributional Post-Processing for Vaccine Demand Forecasting at the Operational Level in Low- and Middle-Income Countries

Udeshi Salgado, DL4SG, Cardiff University, UK
Lead Supervisor: Professor Bahman Rostami-Tabar
Co-supervisors: Dr Thanos E Goltsos, Dr Geraint Palmer, Dr Xun Wang

11 June 2026

Background

1 in 5 children worldwide still lack access to essential vaccines.
A key operational contributor is inefficiency in vaccine supply chains.
In low- and middle-income countries, these inefficiencies often involve:
- inaccurate demand forecasts
- inventory decisions made under limited uncertainty information
- wastage and stockouts

Vaccines

Vial

Dose

Administered

The Immunisation Supply Chain

Why Operational Administrative Level?

The Gap & The Question

In practice, the Forecasting, Supply Planning and Procurement (FSP) Tool relies on static demographic targets and produces point forecasts only, with no representation of uncertainty.

In academia, existing studies focus on aggregate planning levels and point forecast accuracy, with little attention to probabilistic forecasting at the operational level in LMICs.

Forecast distributions are not constrained to feasible values: negative doses and quantities exceeding the eligible population.

Research Question

How well do statistical, ML, DL, and foundation time-series methods forecast routine childhood vaccine doses probabilistically at the operational level in LMICs, and can FSP-informed distributional post-processing improve the feasibility and quality of forecast distributions?

Methodology

Distributional Post Processing

Key Results

Scaled CRPS

RMSSE