MEDIC(AIDS)

Inspiration

38 million people live with HIV globally. In resource-limited settings, who gets treated first? This question drove our project. We saw an opportunity to use UIC clinical trial data to build a modern triage tool that identifies high-risk AIDS patients using simple, affordable lab tests available anywhere.

What it does

MEDIC(AIDS) predicts AIDS treatment outcomes with 76.4% accuracy using only CD4 and CD8 immune cell counts. We combine:

Machine learning to identify high-risk patients
Survival analysis to quantify how treatment history affects outcomes
Statistical validation to ensure findings are robust

Key finding: Treatment-naive patients have 79% survival vs 67% for those with extensive prior therapy (p < 0.0001) -> validating drug resistance patterns documented in the original 1996 research.

How we built it

We built the model using a combination of data preprocessing, feature engineering, and machine learning algorithms. We removed non-contributing features, standardized numerical features, and explored several models, including random forests and ROC-AUC curves. The final model was selected based on predictive accuracy and interpretability. We created various types of plots to visualize the distribution of key variables across the target groups, to compare different method treatments, and identify correlations between variables.

Challenges we ran into

Had to research what CD4/CD8 counts actually mean for immune health
Working with unfamiliar medical terms
Class imbalance as the first model just predicted "everyone survives" with 76% accurate, but useless
Ensuring every finding was properly validated with hypothesis tests

Accomplishments that we're proud of

Advanced methodology: Survival analysis, not just classification
Research validation: Independently confirmed drug resistance findings from 2014 paper
Feature engineering: CD4/CD8 ratios appeared in top 5 most important features
Social impact framing: Connected historical data to modern global health challenges

What we learned

Survival analysis, such as Kaplan-Meier curves, log-rank tests
Clinical ML interpretation, like what does accuracy mean for patient care
Why CD4/CD8 ratios matter for immune health
Drug resistance develops with prolonged therapy
Statistical validation > just model accuracy

What's next for MEDIC(AIDS)

Short-term:

Add SMOTE for better class imbalance handling
Implement Cox proportional hazards for risk factor analysis
Fairness analysis across demographics

Long-term:

Clinical decision support web app for real-time risk scoring
Generalize to COVID-19 and future pandemics (same immune-marker framework)
Policy impact and provide recommendations for WHO treatment guidelines

Vision: Transform MEDIC(AIDS) into a real-world triage tool. When hospitals are overwhelmed, measure immune markers, run the model, and know who needs urgent care. Simple. Fast. Equitable. Life-saving.