Daniellamuli/Phase-5-HIV-Care-Gap-AI

HIV Care Gap AI: County risk mapping, dropout risk factor identification, and 2030 scenario projections for Kenya's HIV programme. Built with KMeans clustering, Logistic Regression (odds ratios), and scenario-based forecasting. Deployed Streamlit dashboard: hivcaregapai.streamlit.app # HIV Care Gap AI · Kenya ### County Risk Mapping · Individual Dropout Risk Factors · 2030 Scenario Forecasting **Using Machine Learning to Identify Who Kenya is Leaving Behind** **Team:** Daniella Muli · Eve Michelle · Naomi Opiyo · Pheonverah Achieng' · Lorenah Mbogo · Dennis Kamuri --- ## The Problem Kenya's HIV response reversed course in 2024. New infections rose **19% in a single year** from 16,752 to **19,991** breaking a decade of hard-won progress. Just **10 counties account for 60% of all new infections**. Treatment Interruption (IIT), defined as missing an ART visit by 28 or more days, is the primary driver of viral rebound and onward transmission. Yet the Ministry of Health had no tool to answer three fundamental questions at county level: | Question | Why it matters | |----------|----------------| | **WHERE** is the health system losing patients, by county? | Resource allocation requires knowing which counties are worst, not just nationally | | **WHO** is most likely to disengage from care, by profile? | Community health worker outreach must be targeted to be effective | | **WHAT HAPPENS** by 2030 if we act or if we don't? | Donors and policymakers need quantified projections to justify investment | **HIV Care Gap AI** builds that tool three integrated models giving the MOH precision intelligence to close the gap. --- ## The Three-Model Architecture | Model | Question | Algorithm | Primary Output | |-------|----------|-----------|----------------| | **Model 1: County Care Gap Map** | *Where* to intervene? | Care Gap Index (CGI) + KMeans clustering (k=4) | 47 counties tiered 🔴 Critical / 🟠 High / 🟡 Moderate / 🟢 Low | | **Model 2: Dropout Risk Factors** | *Who* to prioritise? | Logistic Regression + 500-sample bootstrap | Odds ratios with 95% CI for 17 demographic features | | **Model 3: 2030 Scenario Projection** | *What happens next?* | Tier-based scenario projection (BAU vs Bridged Gap) | Dual-scenario IIT/VLS trajectories + 233,186+ …