Healthcare Analysis

A small toolkit for exploring hospital discharge data and building KPI-focused dashboards. The repository currently centers on pandas-based transformations with supporting notebooks for department-level, patient-level, and “powerplay” dashboards.

Repository structure

• patient_dashboard.py — core data utilities used across the dashboards.
• patient_dashboard.ipynb, department_wise_dashboard.ipynb, powerplay_wise_dahboard.ipynb, outpatient_details_script.ipynb, patient_discharge_details_script.ipynb, patient_wise_dashboard.ipynb — exploratory notebooks for specific KPIs and reporting slices.
• tests/ — pytest coverage for the PatientDashboard helper.
• requirements.txt — minimal runtime and test dependencies.

Getting started

1. Create a Python 3.10+ virtual environment.
2. Install dependencies:

   pip install -r requirements.txt

3. Run the unit tests to confirm the environment:

   pytest

4. Open the notebooks in your preferred environment (Jupyter, VS Code, or similar) to explore or extend the dashboards.

Core helper usage

patient_dashboard.py exposes a PatientDashboard class with a few focused utilities:

• Reading data: read_csv(path) loads comma-delimited CSV files into DataFrames.
• Expense aggregation: add_total_expense_column(df, expense_columns=None, expense_prefix="expense_") sums numeric expense columns and adds an added_amt column. The method validates expected columns and safely coerces non-numeric entries.
• Patient rankings: fetch_patient_phone(df, order) returns the top or bottom 10 patients by added_amt, including phone and department context.
• Age-enriched details: fetch_patient_by_age(discharge_df, details_df) joins age information from a patient-details table, then sorts by age to highlight senior patients with their location and total spend.
• Exporting: to_csv(df, path) persists transformed data.

Current strengths (quick code review)

• Strict column validation: Expense aggregation explicitly checks for missing columns and clear error messaging.
• Defensive numeric handling: Non-numeric expense values are coerced with errors="coerce" to avoid runtime failures.
• Focused tests: Pytest covers prefix-based expense aggregation and missing-column errors.

Improvement opportunities

• Avoid in-place mutation: Several helpers mutate incoming DataFrames (e.g., adding added_amt and age). Returning copies or documenting mutation would reduce side effects when functions are reused.
• Type hints and contracts: Expanding type hints (e.g., accepted order values) and adding docstring examples would clarify expected inputs.
• Data validation: Schema validation (pydantic models or pandera checks) would ensure dashboard inputs are well-formed before aggregation.
• Packaging: Converting the utilities into a small package/CLI with entry points for batch runs would improve reproducibility beyond notebooks.
• Notebook hygiene: Adding lightweight smoke tests that load each notebook’s key cells (via nbmake or papermill) would guard against regression in dashboard logic.

Suggested roadmap

1. Data contract & validation
   • Define a minimal schema for discharge and patient detail tables (required columns, types, nullability).
   • Introduce validation with pandera or pydantic to fail fast before computations.
2. API hardening
   • Make helper methods return copies by default and document mutating behavior.
   • Add error handling for missing/duplicate patient identifiers when merging age and expense data.
3. Automation & testing
   • Expand pytest coverage to include fetch_patient_phone, fetch_patient_by_age, and to_csv.
   • Add notebook execution checks (e.g., pytest --nbmake) to ensure dashboards run end-to-end.
4. Packaging & distribution
   • Convert patient_dashboard.py into an installable package with a CLI (e.g., healthcare-dashboard build --input discharge.csv).
   • Provide sample datasets and expected outputs to accelerate onboarding.
5. Documentation
   • Document expected input CSV headers and sample records.
   • Add how-to guides for generating department/patient/powerplay dashboard extracts.

Contributors ✨

• Rishikesh R Pote
• Amrut Kinage
• Quaid Johar