Heart-Disease Signals

This project investigates which patient factors are most associated with a heart disease diagnosis (presence vs absence) in a clinical evaluation dataset (Cleveland Clinic Foundation; UCI Heart Disease — Cleveland data).

Repo includes:

• Clean, commented analysis notebook
• Presentation slides
• Poster for LinkedIn-style sharing

---

Dataset

• Source: UCI Machine Learning Repository (Heart Disease – Cleveland)
• Rows: 303 patients
• Target: heart_disease (presence / absence)
• Key variables used:
  • cp (chest pain type)
  • exang (exercise-induced angina; 0/1)
  • thalach (max heart rate during exercise test)
  • age, trestbps (resting blood pressure)
  • chol (cholesterol), fbs (fasting blood sugar >120; 0/1)

Data location in this repo: data/heart_disease.csv

Note: This is a clinical evaluation sample, not a random population sample.

---

Methods (hypothesis testing)

• Numeric vs diagnosis (2 groups): Welch two-sample t-test
  (e.g., thalach, age, trestbps, chol)
• Categorical vs diagnosis: Chi-square test of independence
  (e.g., cp, exang, sex)
• Thalach across multiple cp groups: One-way ANOVA + Tukey HSD (FWER = 0.05)
• Benchmark comparison (FBS rate vs 8%): Binomial test (one-sided)

Significance threshold: α = 0.05.

---

Key findings

Strongest signals (symptoms + exercise test)

• Chest pain type (cp) is strongly associated with diagnosis (χ² p = 1.25e−17).
  Asymptomatic patients have the highest heart disease proportion.
• Exercise-induced angina (exang) is a strong indicator:
  heart disease rate 76.8% (exang=1) vs 30.9% (exang=0), χ² p = 1.41e−13.
• Max exercise heart rate (thalach) is much lower in heart disease:
  ~19 bpm lower on average, p = 3.46e−14.
• Across chest pain types, mean thalach differs (ANOVA p = 1.91e−10).
  Tukey HSD shows the asymptomatic group has significantly lower thalach than each other cp type.

Supporting predictors (demographics/vitals)

• Age: heart disease group is older (mean +4.0 years; p ≈ 7.06e−05).
• Resting BP (trestbps): slightly higher in heart disease (mean +5.3 mmHg; p = 0.00855).
• Sex: diagnosis is more frequent among males in this sample
  (55.3% vs 25.8%; χ² p = 2.67e−06).

Benchmarks

• Cholesterol threshold (240 mg/dl): heart disease group mean 251.47 (>240; p(one-sided)=0.0035).
  However, cholesterol is not a strong between-group discriminator here (two-sample p ≈ 0.137).
• High fasting blood sugar (fbs >120): 14.85% observed vs 8% benchmark
  (45 observed vs ~24 expected; p = 4.69e−05), consistent with a clinical (non-population) sample.