akbempah1/medication-adherence-prediction

Machine learning framework for predicting medication non-adherence among chronic disease patients in Ghana using EMR data # Medication Adherence Prediction Using Machine Learning A machine learning framework for predicting medication non-adherence among chronic disease patients in Ghana using structured electronic medical record (EMR) data. ## Overview This project develops an interpretable, context-specific ML model to identify patients at risk of medication non-adherence. The work addresses a critical gap in healthcare AI: most adherence prediction models are developed in high-income settings and fail to generalize to low- and middle-income countries (LMICs) where healthcare financing, access patterns, and socioeconomic factors differ substantially. ### Key Results | Metric | Score | |--------|-------| | Accuracy | 89.5% | | AUC-ROC | 0.934 | | F1-Score | 0.93 | | Recall (Non-adherent) | 77% | | Recall (Adherent) | 94% | ### Top Predictive Features (SHAP Analysis) 1. **Insurance Status (NHIS)** — Strongest predictor of adherence 2. **Age × Medication Complexity** — Interaction effect capturing polypharmacy burden in older patients 3. **Chronic Comorbidity Status** — Presence of conditions like diabetes, heart failure 4. **Age** — Independent effect on adherence behavior 5. **Total Medication Count** — Regimen complexity 6. **Estimated Medication Cost** — Financial burden proxy 7. **Cost Burden × Insurance** — Interaction capturing uninsured cost sensitivity ## Methodology ### Data Source - Structured EMR data from Presbyterian Hospital, Agogo, Ghana - 1,367 adult patients with chronic conditions (hypertension, diabetes, cardiovascular disease) - 6-month observation window for adherence measurement ### Adherence Measurement - **Proportion of Days Covered (PDC)** calculated from prescription refill patterns - Adherent: PDC ≥ 80% | Non-adherent: PDC =1.5.0 numpy>=1.23.0 scikit-learn>=1.2.0 xgboost>=1.7.0 shap>=0.41.0 matplotlib>=3.6.0 seaborn>=0.12.0 streamlit>=1.20.0 ``` ## Usage ### Running the Analysis ```python # Load and preprocess data import pandas as pd from src.pr …