valentineghanem-bit/hiv-spatial-epidemiology-ghana

Bivariate LISA spatial lag regression and Random Forest analysis of HIV prevalence across 261 Ghana districts # Regional HIV Spatial Patterns Mapped Across Ghanaian District Boundaries **Author:** Valentine Golden Ghanem, MSc Public Health, MSc Data Science **Affiliation:** Ghana COCOBOD Cocoa Clinic, Accra, Ghana **ORCID:** 0009-0002-8332-0220 **Reporting standard:** STROBE; ecological cross-sectional spatial analysis **Target journal:** BMC Public Health **Status:** Manuscript in preparation as a BMC Public Health Research article **Repository:** `hiv-spatial-epidemiology-ghana` ## 1. Abstract This repository contains the reproducible analysis for a Ghana HIV spatial epidemiology study using Ghana DHS regional HIV estimates, DHS behavioural indicators, 2021 Census socioeconomic covariates, and GSS district boundaries. The central finding is a strong south-north HIV prevalence gradient when regional DHS estimates are mapped across district boundaries. The manuscript is deliberately conservative: HIV prevalence, VCT uptake, and wife-beating acceptance are regional estimates, not direct district measurements. District maps are therefore communication geometry, not proof of district-resolved HIV differences. ## 2. Research Question & Aims The study asks how HIV prevalence varies geographically in Ghana when public regional DHS estimates are linked to district boundaries, and how far the observed pattern is explained by behavioural, socioeconomic, and spatial structure. The aims are to quantify HIV spatial autocorrelation, compare spatial lag and spatial error regression, test GWR feasibility under regional predictor granularity, compare LASSO and Random Forest/SHAP sensitivity models, and define what public aggregate DHS/Census data can support without overclaiming district inference. ## 3. Methods Summary | Method | Tool | Purpose | |---|---|---| | Global Moran's I | PySAL/esda | National spatial autocorrelation | | LISA and bivariate LISA | PySAL/esda | Local clustering and local covariate association | | Getis-Ord Gi* | PySAL/esda | Hotspot and coldspot cross-che …