Case Study

Bayes’ Law in Heart Failure Data
March 2022

Skills/Tools: Phyton (numpy, pandas, matplotlib, seaborn, networkx)

Just Another Practice

Data

We will use the Heart Failure Dataset. This dataset has 12 attributes, including HeartDisease. 

Goal

This work aims to find the causal inferences between medical condition attributes in the dataset and the heart disease class. We define medical condition attributes as all attributes except age and sex.

Bayes’ Law Analysis

We will analyze causal inferences of 720 rows of cleaned* Heart Failure Data using Bayes’ Law.

1. ChestPainType

P(ASY|HeartDisease) > P(NAP|HeartDisease) > P(ATA|HeartDisease) > P(TA|HeartDisease)

P(ASY|HeartDisease) > P(HeartDisease|ASY)

2. FastingBS

P(HeartDisease|FastingBS) > P(FastingBS|HeartDisease)

3. RestingECG

P(HeartDisease|LVH) > P(LVH|HeartDisease)
P(HeartDisease|ST) > P(ST|HeartDisease)

4. ExerciseAngina

P(HeartDisease|ExerciseAngina) > P(ExerciseAngina|HeartDisease)

5. ST_Slope

P(Flat|HeartDisease) > P(Up|HeartDisease) > P(Down|HeartDisease)
P(HeartDisease|Down) > P(Down|HeartDisease)

6. RestingBP

P(HeartDisease|>130) > P(>130|HeartDisease)

7. Cholesterol

P(HeartDisease|>235) > P(>235|HeartDisease)

8. MaxHR

P(>140|HeartDisease) > P(HeartDisease|>140)

9. Oldpeak

P(>0.4|HeartDisease) > P(HeartDisease|>0.4)

Visualization

dfVis = pd.DataFrame({ 'from': ['heart\ndisease', 'TA\nchestpain', 'fasting\nblood sugar\n>= 120mg/dl', 'LVH\nresting ECG\nresult', 'ST\nresting ECG\nresult', 'exercise-\ninduced\nangina','down-slope of\nthe peak exercise\nST segment','resting\nblood pressure\n> 130mmHg', 'cholesterol\nlevel\n> 235mm/dl', 'heart\ndisease', 'heart\ndisease'],
                       'to': ['ASY\nchestpain', 'heart\ndisease', 'heart\ndisease', 'heart\ndisease', 'heart\ndisease', 'heart\ndisease','heart\ndisease', 'heart\ndisease', 'heart\ndisease','maximum\nheart rate\n> 140', 'oldpeak\n> 0.4']})
G = nx.from_pandas_edgelist(dfVis, 'from', 'to', create_using=nx.DiGraph())

plt.figure(figsize=(8, 8))
nx.draw(G, with_labels = True, node_size = 6000, node_color = 'papayawhip', 
        pos = nx.circular_layout(G), arrows = True, font_size = 11)
plt.title('Causal Inferences Using Bayes\' Law on Heart Failure Data')
plt.show()
*Please see this notebook for the complete cleaning process and analysis.