valentineghanem-bit/maternal-reproductive-health-ghana-261-districts

Spatial inequities and ML risk stratification of maternal and reproductive health across 261 districts of Ghana: analysis pipeline, HI-EI dashboard, bespoke A0 poster. # Spatial Inequities in Maternal and Reproductive Health Outcomes and ML Risk Stratification Across 261 Districts of Ghana **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 · RECORD-Spatial · TRIPOD+AI **Date:** 2026 **Status:** Manuscript in preparation --- ## 1. Abstract This ecological, cross-sectional study quantifies spatial inequities in maternal and reproductive health service coverage across all 261 health districts of Ghana (2022 administrative boundaries). A 7-component composite maternal health index (DHS 2022 ANC4+, skilled birth attendance, facility delivery, and postnatal care, combined with inverted Census 2021 poverty, illiteracy, and uninsured rates) averaged 65.6 (SD 17.8) and was strongly spatially clustered (Global Moran's I = 0.8437, z = 21.36, p = 0.001; 999 permutations, KNN k = 4). Univariate and bivariate LISA and Getis-Ord Gi* localise a persistent northern disadvantage. A geographically weighted regression (GWR, pure-Python bisquare adaptive kernel, AICc bandwidth) substantially outperformed Global OLS (R² 0.9774 vs 0.898), evidencing spatially varying determinant effects. Gradient-boosting risk stratification with permutation importance and region-stratified leave-one-region-out cross-validation (LOROCV) identifies working-age population proportion, female population share, and the proportion of women without health insurance as the dominant district-level predictors of HIGH-risk classification. --- ## 2. Research Question & Aims - **Primary:** Quantify and map spatial inequities in a composite maternal and reproductive health service-coverage index across Ghana's 261 districts, and identify their district-level determinants. - **Secondary:** (a) Characterise spatial clustering of the composite index using Global/Local Moran's I, bivariate LISA, and Getis-Ord Gi*; (b) test whether de …