Rapid Screening for Cardiovascular Risk

A high-sensitivity clinical triage tool designed to predict heart disease using a subset of seven objective "hard indicators." Built for the Byte 2 Beat Hackathon.

📊 Performance Summary

• Recall (Sensitivity): 73%
• Negative Predictive Value (NPV): 96%
• Statistical Significance: $p < 0.001$ across all indicators
• Test Set Size: 50,736 observations

🔬 The "Rapid Risk Indicator" Set

Our research identified that cardiovascular risk can be effectively triaged using only seven objective, verifiable data points, bypassing the need for subjective survey data:

1. Age (Odds Ratio: 2.59)
2. Stroke History (Odds Ratio: 1.35)
3. Smoking Status (Odds Ratio: 1.33)
4. BMI
5. Sex
6. Heavy Alcohol Consumption
7. Healthcare Access

🛠️ Tech Stack

• Language: Python
• Modeling: Scikit-learn (Logistic Regression with Balanced Class Weights)
• Inference: Statsmodels (Maximum Likelihood Estimation)
• Visualization: Matplotlib, Seaborn

🚀 Methodology

1. Data Cleaning: Leveraged the CDC's BRFSS dataset (253,680 records).
2. Feature Selection: Filtered 22 variables down to 7 "Hard Indicators" to reduce subjective bias.
3. Validation: Utilized a 60/20/20 split with Calibration Curve analysis to ensure clinical reliability.

📁 Repository Structure

• notebooks/: Contains eda.ipynb for initial discovery and model.ipynb for the full experimental pipeline and statistical analysis.
• docs/: Includes primary research references and the final Rapid Screening for Cardiovascular Risk_ A High-Sensitivity Triage Approach.pdf.
• images/: Exported visualizations including confusion_matrix.png, calibration_curve.png, and p_values.png.
• data/: Structured storage for raw BRFSS data and processed curated subsets.
• requirements.txt: Environment dependencies for reproducibility.