DataFest 2026 — Stormont Vail Health

Analyzing patient journeys to understand how MyChart app engagement affects emergency utilization and care continuity.

---

Research Question

Does MyChart app activation improve patient journey quality — and does geographic distance compound or moderate this effect?

---

Key Findings

Group	ED Rate
App User	0.6%
Non-App User	1.3%
App User + Transport Hardship	2.5%
Non-App User + Transport Hardship	8.2%
Near + App User	0.9%
Near + Non-App User	2.3%

• MyChart activation is associated with a 54% reduction in ED visit rate (p < 0.0001)
• Non-app users with transport hardship show ED rates 13× higher than app users without transport issues
• Distance alone does not strongly predict ED utilization — app status and transportation hardship are stronger predictors
• The app benefit is strongest for patients living near SVH — precisely the group most likely to default to the ER for non-urgent care

---

Recommendation

SVH should prioritize MyChart outreach to patients who are:

1. Living within close proximity to SVH facilities
2. Flagged with transportation hardship in social determinants survey
3. Not yet activated on MyChart

This is a definable, reachable population that stands to benefit most from digital engagement.

---

Project Structure

datafest/
├── config.py                    # Centralized path configuration
├── requirements.txt             # Python dependencies
├── .gitignore
├── data/
│   ├── raw/                     # Original CSV files (not tracked by git)
│   ├── processed/               # Cleaned outputs used by visualization notebook
│   └── samples/                 # Development subsets
├── notebooks/
│   ├── 00_eda.ipynb             # Exploratory data analysis
│   ├── 01_mychart.ipynb         # Main analysis: app usage vs journey quality
│   ├── 02_distance.ipynb        # Supplementary: geographic distance analysis
│   └── 03_visualizations.ipynb  # All charts for final presentation
├── src/
│   ├── data_loader.py           # CSV loading with dtype optimization
│   ├── distance_utils.py        # Haversine distance + patient geocoding
│   ├── journey_builder.py       # Patient journey construction via DiagnosisValue
│   └── plot_utils.py            # Shared plotting style and save functions
└── outputs/
    ├── figures/                 # Saved charts (PNG + PDF)
    └── tables/                  # Saved result tables

---

Setup

# Clone the repo
git clone https://github.com/SYzhao666/datafest2026.git
cd datafest2026

# Create and activate virtual environment
python -m venv venv
venv\Scripts\activate        # Windows
source venv/bin/activate     # Mac/Linux

# Install dependencies
pip install -r requirements.txt

# Register Jupyter kernel
python -m ipykernel install --user --name=datafest

Place the raw CSV files in data/raw/ before running any notebooks.

---

Data

Source: Stormont Vail Health (SVH), Topeka, Kansas
Period: January 2022 – December 2025
Scale: 947,685 patients · 7,675,801 encounters

Key files used:

File	Rows	Description
encounters.csv	7,675,801	All patient-system interactions
patients.csv	947,685	Patient demographics + MyChart status
diagnosis.csv	1,531,262	ICD-10 diagnosis codes
departments.csv	11,597	Department locations and specialties
social_determinants.csv	3,977,901	SDOH survey responses
tigercensuscodes.csv	2,463	Census block centroids (Kansas)

Raw data is not tracked by git due to size and privacy constraints.

---

Notebook Order

Run notebooks in sequence:

Notebook	Purpose
00_eda.ipynb	Data structure, distributions, missing values
01_mychart.ipynb	Core app usage analysis + transport hardship cross-analysis
02_distance.ipynb	Geographic distance supplementary analysis
03_visualizations.ipynb	All 8 presentation charts

---

Technical Notes

• Encounters are sampled at 5% during development for speed; set sample_frac=1.0 for final runs
• Patient journeys are defined by (PatientDurableKey, DiagnosisValue) pairs — not PrimaryDiagnosisKey, which can change mid-journey due to federal coding updates
• Distance is calculated as straight-line (Haversine) from patient census block centroid to SVH main campus (39.0558, -95.6890)
• Patients with *Unspecified census blocks (~65%) are excluded from distance analysis

---

Dependencies

• pandas >= 2.0
• numpy
• matplotlib
• seaborn
• scipy
• jupyter
• ipykernel