ABC Pharma – Pharmacy Patient Adherence Risk Prediction System

Business Context

Retail pharmacies like Walgreens serve millions of patients managing chronic diseases such as diabetes. One of the biggest operational challenges is medication non-adherence, which leads to:

• poor health outcomes
• increased healthcare costs
• lost pharmacy revenue

This project simulates how a retail pharmacy analytics team can use healthcare data and machine learning to:

• identify patients at risk of medication non-adherence
• forecast refill gaps
• quantify revenue at risk
• prioritize pharmacy interventions

A production-style end-to-end analytics system for analyzing diabetes medication adherence, predicting non-adherence risk, and generating actionable insights for pharmacy operations and patient engagement.

---

🧭 Overview

This project simulates how a large pharmacy organization like ABC Pharma analyzes diabetes medication behavior using:

• Public datasets (CMS Part D Drug Spending, Prescriber Diabetes)
• Synthetic enterprise-grade patient-level data
• A full adherence analytics pipeline

The system includes:

• Data processing
• Adherence metric computation (MPR, PDC, refill gaps)
• Feature engineering
• Predictive modeling (Logistic Regression, RandomForest)
• Visualization
• Modular architecture

---

🎯Business Impact

This analytics system enables pharmacy organizations to:

• identify high-risk patients before medication gaps occur
• prioritize pharmacist outreach programs
• improve medication adherence rates
• reduce revenue loss from missed refills
• optimize patient engagement strategies

Potential measurable outcomes:

• 10–15% improvement in adherence
• improved patient outcomes
• millions in recovered pharmacy revenue

🏗️ Project Structure

src/
  data_process.py        # Load data, filter diabetes drugs, compute MPR/PDC/gaps
  analysis.py            # Feature engineering + ML models
  visualization.py       # Plots for adherence + model curves
  main.py                # Orchestration pipeline

reports/
  tables/                # adherence_metrics_diabetes.csv, feature_table_diabetes.csv
  figures/               # PDC/MPR/gap distributions, ROC/PR curves

data/
  raw/synthetic/         # Synthetic enterprise data

Data Sources
    ↓
CMS Medicare Part D Data
Synthetic Patient Dataset
Pharmacy Store Data

    ↓
Data Engineering Layer
Python ETL Pipelines
Feature Engineering

    ↓
Analytics Layer
Adherence Metrics
PDC / MPR
Refill Gap Detection

    ↓
Machine Learning
Logistic Regression
Random Forest

    ↓
Business Insights
Patient Risk Scores
Revenue-at-Risk Analysis
Dashboard Reporting

---

📊 Outputs

1. Tables (reports/tables/)

File	Description
adherence_metrics_diabetes.csv	MPR, PDC, refill gaps, non-adherence label
feature_table_diabetes.csv	Full ML feature table

---

2. Figures (reports/figures/)

pdc_distribution

mpr_distribution

max_gap_distribution

ROC_LOGIT

PR_LOGIT

---

3. Console Model Results

Model evaluation outputs include:

• AUC
• PR-AUC
• Classification Report
• Positive Rate
• Feature Table Shape

---

⚙️ Tech Stack

• Python
• Pandas
• NumPy
• Scikit-learn
• Matplotlib
• Synthetic healthcare data generation

---

📌 Future Improvements

• Add XGBoost / LightGBM models
• Deploy a Streamlit dashboard
• Integrate real CMS Medicare Part D claims data
• Implement patient segmentation analytics

---

📜 License

This project is for educational and learning purposes.