Forecasting improvements for better reproductive health and family planning operations in global health supply chains

Halgamuwe Hewage, Harsha 2026. Forecasting improvements for better reproductive health and family planning operations in global health supply chains. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Download (9MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (1MB)

Abstract

Public health supply chains in low- and middle-income countries depend on reliable demand forecasts to support procurement, replenishment, and distribution decisions. Yet forecasting in these settings remains difficult because demand is highly uncertain, routine data are often sparse and distorted, and forecasting methods must operate under substantial technical and operational constraints. This thesis examines how forecasting support in public health supply chains can be improved under three interrelated challenges: (i) representing uncertainty in local demand forecasting, (ii) estimating true demand when observed demand is censored by stockouts and service interruptions, and (iii) evaluating scalable forecasting architectures for large operational systems. The thesis addresses these challenges through three connected studies. The first develops a hybrid probabilistic forecasting framework that combines expert point forecasts with model-based probabilistic forecasts for contraceptive demand at healthcare-site level. The second develops a censored-demand forecasting framework, the Truncated Conformal Kalman Filter, to recover latent demand and assess its downstream inventory and public health implications in family planning supply chains. The third evaluates global and transfer-learning forecasting architectures, including time series foundation models, across large public health datasets using more than 25,000 facility-level time series, and assesses them in terms of both forecast performance and operational feasibility. The findings show that hybrid forecasting can improve local forecasting support when important contextual knowledge is not captured in routine data. Moreover, explicitly modelling censored demand can reduce under-ordering and improve downstream inventory performance, while supporting improved service outcomes. Finally, the results indicate that forecast quality should be assessed more broadly than statistical performance alone, and that model selection in resource-limited settings must also consider scalability, runtime, maintainability, and deployment feasibility. Overall, the thesis contributes a more operationally grounded understanding of forecast quality and provides methodological and empirical evidence to support forecasting in resource-constrained public health systems.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Business (Including Economics)
Subjects:	H Social Sciences > H Social Sciences (General)
Uncontrolled Keywords:	1. Public health supply chains 2. Family planning 3. Probabilistic forecasting 4. Censored demand 5. Artificial intelligence in forecasting 6. Time series foundation models
Funders:	ESRC Wales Doctoral Training Partnership (DTP) Studentship
Date of First Compliant Deposit:	28 October 2026
Last Modified:	29 Oct 2026 15:46
URI:	https://orca.cardiff.ac.uk/id/eprint/190490

Actions (repository staff only)

Edit Item