EHR Readmission and Care Quality Analytics

Python, SQL, and Power BI portfolio project using a synthetic EHR-style dataset. The project demonstrates cohort construction, encounter-level feature engineering, readmission outcome logic, data-quality checks, and dashboard-ready reporting tables.

Project Question

Among adult inpatient encounters in a synthetic EHR dataset, what operational and clinical factors are associated with 30-day readmission, and where are the main care-quality and data-quality gaps?

What This Project Demonstrates

• EHR-style data modeling across patients, encounters, diagnoses, observations, and medications.
• SQL cohort logic for index admissions, readmission windows, chronic-condition flags, and quality checks.
• Python/pandas data generation, cleaning, feature engineering, aggregation, and validation.
• SQLite validation that rebuilds cohort-level outputs from raw EHR-style tables.
• Power BI-ready reporting tables for readmission monitoring and care-quality review.
• Power BI theme, semantic model, grouped DAX measures, build instructions, and static reference previews for the report pages.
• Responsible interpretation using synthetic data only.

Additional report previews:

• Readmission Overview
• Care Quality and Data Quality
• Cohort Profile

Headline Results

Metric	Result
Patients	900
Total encounters	5,020
Eligible index admissions	1,102
30-day readmissions	80
Overall readmission rate	7.3%
Average length of stay	3.1 days
High-risk encounters	130

The derived high-risk tier had a 13.8% readmission rate, compared with 6.6% in the moderate-risk tier and 5.8% in the low-risk tier. Documentation completion was 78.7% for A1c among diabetes encounters and 82.3% for blood-pressure documentation among hypertension encounters.

Repository Structure

.
├── .github/
│   └── workflows/
│       └── tests.yml
├── assets/
│   └── dashboard-preview.png
├── data/
│   ├── processed/
│   └── raw/
├── docs/
│   ├── DATA_DICTIONARY.md
│   ├── DATA_SOURCE_AND_VALIDATION.md
│   ├── DATA_MODEL.md
│   ├── METHODS.md
│   ├── POWERBI_BUILD_INSTRUCTIONS.md
│   ├── POWERBI_VISUAL_BUILD_GUIDE.md
│   ├── POWERBI_INTERACTIVE_SPEC.md
│   └── POWERBI_GUIDE.md
├── pipeline/
│   ├── 01_schema.sql
│   ├── 02_cohort_logic.sql
│   ├── 03_quality_checks.sql
│   ├── 04_sqlite_cohort_validation.sql
│   ├── create_powerbi_previews.py
│   ├── build_sqlite_database.py
│   ├── README.md
│   └── build_ehr_readmission_dataset.py
├── reports/
│   └── ANALYST_BRIEF.md
├── powerbi/
│   ├── README.md
│   ├── EHR_Readmission_Quality_Analytics.pbip
│   ├── EHR_Readmission_Quality_Analytics.Report/
│   ├── EHR_Readmission_Quality_Analytics.SemanticModel/
│   ├── measures.dax
│   └── theme-ehr-readmission.json
├── requirements.txt
└── tests/
    └── test_pipeline_outputs.py

Run

pip install -r requirements.txt
python pipeline/build_ehr_readmission_dataset.py
python pipeline/build_sqlite_database.py
python pipeline/create_powerbi_previews.py
python -m unittest discover -s tests -v

The pipeline creates deterministic synthetic EHR records and dashboard-ready CSV tables in data/processed/. The SQLite validation script rebuilds the index-admission cohort from raw tables and exports sql_validation_summary.csv.

Key Outputs

• readmission_summary_by_service.csv
• risk_tier_summary.csv
• quality_measure_summary.csv
• feature_importance_proxy.csv
• data_quality_summary.csv
• operational_flag_summary.csv
• cohort_summary_by_age.csv
• cohort_summary_by_insurance.csv
• dashboard_kpis.csv
• sql_validation_summary.csv
• reports/ANALYST_BRIEF.md

Power BI

The Power BI project file is powerbi/EHR_Readmission_Quality_Analytics.pbip. It includes an embedded semantic model, grouped DAX measures, and three completed dashboard pages.

The data/processed/ tables remain the reproducible reporting layer. powerbi/measures.dax contains the recommended slicer-responsive measures, docs/POWERBI_VISUAL_BUILD_GUIDE.md provides the exact visual field map, and docs/POWERBI_INTERACTIVE_SPEC.md defines slicers, interactions, formatting, and page structure. The static preview in assets/dashboard-preview.png shows the intended layout.

For step-by-step Power BI report construction, use docs/POWERBI_VISUAL_BUILD_GUIDE.md and the theme stored in powerbi/theme-ehr-readmission.json.

Data Scope

All records are simulated. The project contains no real patient data, no PHI, no MIMIC-IV data, and no restricted clinical source data.

See docs/DATA_SOURCE_AND_VALIDATION.md for the data-source rationale, validation checks, and interpretation limits.

Interpretation Boundary

This project is a reproducible analytics demonstration. It should not be interpreted as clinical evidence, a validated predictive model, or a measure of real-world readmission performance.