Healthcare Analytics

Analyzing patient healthcare data to find patterns in admissions, billing, and treatment outcomes.

About

I built this project to explore real-world healthcare data and practice my data analysis skills. The goal was to understand patient demographics, identify billing patterns, and find insights that could help healthcare providers make better decisions.

Dataset

Source: Kaggle - Healthcare Dataset

The dataset contains 55,500 patient records from 2019-2024 with the following features:

Feature	Description
Name	Patient name
Age	Patient age (18-85)
Gender	Male/Female
Blood Type	A+, A-, B+, B-, AB+, AB-, O+, O-
Medical Condition	Cancer, Diabetes, Obesity, Asthma, Hypertension, Arthritis
Date of Admission	Hospital admission date
Doctor	Attending physician
Hospital	Healthcare facility name
Insurance Provider	Cigna, Medicare, Aetna, UnitedHealthcare, Blue Cross
Billing Amount	Total cost of treatment ($)
Room Number	Assigned room
Admission Type	Urgent, Emergency, Elective
Discharge Date	Hospital discharge date
Medication	Prescribed medication
Test Results	Normal, Abnormal, Inconclusive

Methodology

What This Project Is

This is an Exploratory Data Analysis (EDA) project - not a predictive modeling project. I focused on understanding the data through visualization and statistics rather than building machine learning models.

Why EDA Instead of ML?

Approach	What It Does	Why I Chose/Didn't Choose
EDA	Understand patterns, distributions, relationships	Chose this - best for understanding healthcare data before making decisions
Predictive Modeling	Train models to predict outcomes	Not suitable - dataset is synthetic with uniform distributions, predictions wouldn't be meaningful

What I Did

1. Data Cleaning

   • Standardized column names (lowercase, underscores)
   • Fixed inconsistent text formatting in patient names
   • Converted date strings to datetime objects

2. Feature Engineering

   • Created length_of_stay by calculating days between admission and discharge

3. Exploratory Analysis

   • Distribution analysis (age, gender, conditions)
   • Group comparisons (billing by condition, by insurance)
   • Correlation analysis between numeric variables

4. Visualization

   • Histograms and bar charts for distributions
   • Box plots for comparing groups
   • Heatmap for correlations
   • Pie charts for proportions

What I Did NOT Do

• No machine learning models trained
• No predictions made
• No classification or regression

Results

Demographics

Age is uniformly distributed from 18-85 years. Gender split is roughly 50/50. This suggests the dataset is synthetic rather than real patient data.

Medical Conditions

All six conditions have nearly equal patient counts (~9,200 each). In real healthcare data, we'd expect different prevalence rates.

Billing Analysis

Average billing ranges from $25,400-$25,700 across all conditions. The similarity suggests billing isn't strongly tied to condition type in this dataset.

Insurance Providers

Five insurance providers with roughly equal market share (~20% each). Average billing is similar across providers - no significant cost differences.

Admission Type & Length of Stay

Three admission types: Emergency, Elective, and Urgent. Length of stay varies from 1-30 days with median around 15 days. Emergency cases show slightly higher variability.

Correlation Matrix

Weak correlations between numeric variables. Age has minimal impact on billing amount or length of stay, which is interesting - older patients don't necessarily cost more.

Key Findings

Metric	Value
Total Patients	55,500
Date Range	May 2019 - May 2024
Avg Billing	$25,539
Avg Length of Stay	15.5 days
Most Common Condition	All equally distributed

Observations

1. Uniform distributions suggest this is synthetic data - useful for practice but findings shouldn't be applied to real healthcare settings

2. No strong predictors - age, condition, and insurance don't significantly impact billing, which differs from real-world healthcare where these factors matter a lot

3. Length of stay is the most variable metric - could be worth investigating what drives longer stays

4. Emergency admissions show more variance in length of stay compared to elective procedures

Limitations

• Dataset appears synthetic (too uniform)
• No patient outcomes tracked
• Missing cost breakdown (procedure vs medication vs room)
• No readmission data

Future Work

• Add predictive modeling if real data available
• Analyze seasonal admission patterns
• Build interactive dashboard with Plotly

Project Structure

├── dataset/               # data files
├── docs/                  # notes
├── results/
│   ├── plots/            # saved charts
│   └── findings/         # analysis notes
├── analysis.ipynb        # main notebook
├── requirements.txt
└── README.md

Setup

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
jupyter notebook

Tools Used

• pandas - data manipulation
• numpy - numerical operations
• matplotlib - static visualizations
• seaborn - statistical plots
• jupyter - interactive analysis

---

Anik Tahabilder Department of Computer Science Wayne State University