RWE Health Analytics Platform

A modular analytics framework for working with Real World Evidence (RWE) and electronic health records (EHR).
The project is structured as a realistic healthcare data pipeline: from synthetic data generation to ingestion, cohort construction, exploratory analysis, and the foundation for downstream modeling (risk prediction, survival analysis, and causal inference).

---

Business / Clinical Objective

This project simulates a realistic Real-World Evidence (RWE) workflow using synthetic EHR data to demonstrate how patient-level healthcare data can be transformed into clinically meaningful insights and modelling-ready datasets. It is designed to reflect use cases relevant to HEOR, clinical research, digital health, and observational analytics

Use Cases / Why it matters

Cohort construction for observational studies

Risk stratification and outcome prediction

Survival analysis for time-to-event endpoints

Treatment pattern and medication utilization analysis

Causal inference and comparative effectiveness research

Synthetic data prototyping for privacy-preserving healthcare analytics

---

Potential Digital Health Applications

The analytical workflows implemented in this project are relevant for several digital health and precision medicine use cases, including:

• digital biomarker discovery • remote patient monitoring • clinical risk stratification • early disease detection • decision support systems

These approaches are particularly relevant for chronic disease management contexts such as diabetes, cardiovascular disease and metabolic disorders.

---

1. Overview

This repository provides a reproducible environment for analysing longitudinal, patient-level healthcare data.
It follows principles commonly used in pharmaceutical R&D, HEOR, and clinical analytics:

• Structured data ingestion and validation
• Cohort definition logic
• Domain-aware exploratory analysis
• Modular organisation under src/
• Reproducible workflows using notebooks

The project currently uses synthetic EHR-style data aligned with realistic clinical patterns, but the architecture supports real datasets with minimal adaptation.

---

2. Tech Stack

Core Languages & Tools

Machine Learning & Modeling

Causal Inference & Survival Analysis

Visualization

Apps, Data & Infra

---

3. Project Structure

rwe-health-analytics/
│
├── data/
│   ├── raw/                     # Synthetic EHR datasets
│   └── processed/               # Cleaned datasets (future)
│
├── notebooks/
│   └── 01_exploraroy.analysis.ipynb
│
├── src/
│   └── rwe_health_analytics/
│       ├── data/
│       │   ├── data_loader.py
│       │   └── data_generation/
│       │       └── synthetic_data_generator.py
│       ├── models/              # Survival / ML / causal (future)
│       ├── evaluation/          # Metrics and validation (future)
│       └── visualization/       # Plotting utilities (future)
│
├── tests/                       # Test suite
├── docs/                        # Methodology and technical notes
├── requirements.txt
└── setup.py

---

4. Data Flow

The system follows a structured, modular lifecycle:

1. Data Generation

Synthetic patient-level data are created using clinically-inspired rules and realistic statistical distributions.

2. Data Ingestion

HealthcareDataLoader reads and validates the raw tables, offering:

• loading by domain (patients, diagnoses, labs…)
• patient-level queries
• data quality checks
• basic cohort creation logic

3. Exploratory Analysis

Jupyter notebooks provide:

• demographic summaries
• comorbidity patterns
• laboratory distributions
• medication behaviour
• longitudinal event timelines

4. Modelling Layer (future)

The architecture is prepared for:

• survival analysis
• risk prediction
• clustering
• causal inference pipelines

Maintaining separation between data → logic → analysis.

---

5. Installation

git clone https://github.com/Finarosalina/rwe-health-analytics.git
cd rwe-health-analytics

Create environment:

python -m venv venv

Activate:

Windows

venv\Scripts\activate

macOS/Linux

source venv/bin/activate

Install dependencies:

pip install -r requirements.txt
pip install -e .

---

6. Synthetic Data Generation

Run:

python src/rwe_health_analytics/data/data_generation/synthetic_data_generator.py

This generates:

data/raw/
    demographics.csv
    diagnoses.csv
    lab_results.csv
    medications.csv
    outcomes.csv

Dataset Description

Dataset	Description
Demographics	Age, sex, race, BMI, smoking, insurance
Diagnoses	ICD-like codes, visit types, chronicity
Laboratories	Lab measures with normal/abnormal ranges
Medications	Prescriptions, dose, duration, specialty
Outcomes	Hospitalizations, ER visits, mortality

Data are generated using:

• normal, log-normal, Poisson, and multinomial distributions
• disease-driven physiological changes
• comorbidity-based risk modeling
• temporal sequencing for visits, prescriptions, labs and outcomes

This enables EDA, ML, causal inference and survival analysis without PHI/PII.

---

7. Exploratory Analysis

Launch notebooks:

jupyter notebook

Notebook included:

notebooks/01_exploratory_analysis.ipynb

The analysis covers:

• demographic distributions
• comorbidity burden
• diagnosis patterns
• medication dynamics
• laboratory value distributions
• outcome timelines

---

8. Extensibility

The project is structured as a Python package to support clean expansion.

Future modules may include:

• Cox PH models
• Random Survival Forests
• Gradient-boosted risk models
• Propensity score pipelines
• Streamlit dashboard
• Docker deployment

---

9. Testing

pytest
pytest --cov=src

---

10. License

MIT License.

---

11. Author

Maria Pais Fajín
GitHub: https://github.com/Finarosalina
LinkedIn: https://linkedin.com/in/maria-pais-fajin