saeeda-tech/naija-health-pipeline

**Write-up:** I Found a 28% Discrepancy in Child Mortality Data. Most of It Was My Own Mistake. # African Health Indicators Pipeline An automated data pipeline that collects, cleans and publishes public health indicators for six African countries from two independent international sources. **Status:** Weeks 1-4 and 6 complete. Pipeline runs weekly; write-up published. Dashboard outstanding. --- ## Why this exists Health indicators for African countries are published by several international bodies, in different formats, on different schedules, using different methodologies. Comparing them means manual downloads and one-off spreadsheets that go stale immediately. This pipeline automates the collection, standardises the output, and runs itself on a schedule so the numbers are never out of date. It also makes visible something that manual comparison hides: **WHO and DHS sometimes report different values for the same indicator, country and year.** Quantifying and explaining that divergence is a core aim of the project. ## Coverage **Countries:** Nigeria, Ghana, Kenya, Senegal, Egypt, South Africa **Sources:** | Source | What it provides | Access | |---|---|---| | WHO Global Health Observatory | Modelled annual estimates, 2000–present | Open OData API | | The DHS Program | Survey-based indicators, per survey round | Open REST API | **Indicators:** life expectancy, maternal mortality, under-five mortality, immunisation coverage, health expenditure per capita, antenatal care, facility delivery, skilled birth attendance, child stunting, contraceptive prevalence. ## Architecture ``` WHO GHO API ─┐ ├─→ ingest.py ─→ data/raw/ ─→ DuckDB ─→ dbt ─→ Streamlit DHS API ─────┘ (extract) (dated (load) (model) (dashboard) JSON) ▲ GitHub Actions (weekly schedule) ``` **Design decisions:** - **Raw data is never modified.** Extraction writes source JSON unchanged, into date-partitioned folders. Transformation logic can be corrected and re …