ClinicalLens — Survival Analysis + Causal Inference on ICU Data

What this demonstrates

• Survival analysis (Kaplan-Meier, Cox PH) on clinical EHR data
• Causal inference via Inverse Probability of Treatment Weighting (IPTW)
• Correcting for confounding bias (Simpson's Paradox in ICU ventilation data)
• Interactive clinical dashboard built with Streamlit

Key result

Naive comparison: HR=~1.35 (ventilation appears harmful — confounding)
IPTW-adjusted: HR=~0.75 (true protective effect recovered)

Stack

Python · Lifelines · scikit-learn · Streamlit · Pandas

To run locally

pip install -r requirements.txt
python src/data_gen.py
python src/preprocessing.py
streamlit run app/app.py

Swapping in real MIMIC-IV data

Replace data_gen.py output with MIMIC-IV's icustays, patients, chartevents tables. Column mapping in preprocessing.py covers the schema translation.