Causal Inference in Healthcare: Treatment Effect Analysis

Project Overview

This Streamlit application demonstrates causal inference techniques using the Infant Health and Development Program (IHDP) dataset to understand the impact of medical interventions on outcomes.

Problem Statement

Understanding the true causal effects of medical interventions is crucial in healthcare. This project uses advanced causal inference methods to separate correlation from causation in observational data.

Key Features

• Interactive web interface for causal inference analysis
• Multiple causal inference methods implementation
• Visualization of treatment effects
• Personalized outcome prediction
• Statistical analysis of treatment impacts

Methodology

Causal Inference Techniques Implemented

1. Propensity Score Matching (PSM)

   • Balances covariates between treatment and control groups
   • Estimates treatment effects by matching similar units

2. Doubly Robust Estimation

   • Combines propensity score modeling with outcome regression
   • Provides robust estimates of treatment effects
   • Reduces bias from model misspecification

Key Metrics Calculated

• Average Treatment Effect (ATE)
• Individual Treatment Effect (ITE)
• Propensity Scores
• Model Performance Metrics (R², MSE)

Technical Components

Data Processing

• Synthetic data generation
• Confounder identification
• Data normalization

Machine Learning Models

• RandomForestRegressor for outcome prediction
• LogisticRegression for propensity score estimation
• LinearRegression for outcome modeling

Setup

git clone <repository-url>
cd causal-inference-project
pip install -r requirements.txt
streamlit run app.py

Usage

1. Navigate the sidebar to different sections
2. Use sliders to input patient characteristics
3. Select treatment status
4. Explore causal inference results and visualizations

Sections

1. Input & Results

   • Personalized treatment effect estimation
   • Outcome predictions
   • Causal inference metrics

2. Dataset Overview

   • Data distribution
   • Covariate balance
   • Treatment group characteristics

3. Model Diagnostics

   • Model performance metrics
   • Feature importance analysis

Contributors

• PixelPair