valentineghanem-bit/disease-burden-forecasting-ghana

Forecasting Ghana's national health and demographic indicators to 2030, with district-level structural-vulnerability context # A Reproducible Workflow and Source-Provenance Audit for Forecasting Short National Health-Indicator Panels: A Worked Example from Ghana Using WHO Global Health Observatory and World Bank Data **Author:** Valentine Golden Ghanem | Ghana COCOBOD Cocoa Clinic, Accra, Ghana **ORCID:** 0009-0002-8332-0220 **Affiliation:** Ghana COCOBOD Cocoa Clinic, Accra, Ghana **Reporting standard:** STROBE (observational panel components) **Date:** 2026 **Status:** Analysis complete; manuscript in preparation for journal submission ## 1. Abstract Public-domain WHO Global Health Observatory (GHO) and World Bank exports are the default empirical basis for national health-indicator forecasting, but these files carry undocumented structural pitfalls that silently corrupt an analysis if uncorrected, and forecasting choices — method, model order, fitting scale — are typically made ad hoc and per-indicator with no shared, reproducible protocol. This project assembles a Ghana national indicator panel from nine public-domain WHO GHO/Global Health Estimates and World Bank World Development Indicators sources, documents and verifies every data-integrity correction required to reach an analysis-ready panel (six were required), and specifies a data-length-tiered forecasting protocol: fixed eligibility thresholds, small-sample-corrected (AICc) per-series ARIMA order selection, log-scale fitting for non-negative-bounded series, interval-reliability flags tied to series length, and a structural-break sensitivity test for the 2020 COVID-19 and 2022 Ghana currency-crisis candidate breaks. Of 25 indicators assembled, 21 met the 15-year minimum for formal forecasting. A uniformly-applied ARIMA(1,1,1) was the AICc-optimal order for only 1 of these 21 series; per-series order selection shifted the 2030 point forecast by up to ~24% for the most affected indicator. Separately, a naive raw-scale ARIMA-versus-exponential-smoothing disagreement for malaria (~24%) was closed mostly by log-scale fitting …