How to reduce heart disease risk? An exploratory Model Analysis on open data set

Age going above 35 is when some of us starts thinking about fitness (and obesity). While there are many points of inspiration, mine are purely on reducing heart disease risk, particularly when it comes as hereditary. There are many questions I asked myself. Major ones are

• How do I know if jogging for 5KM everyday is effective? If so, what part of it makes it effective?
• How much sugar and tempting high calorie food I can have, if I run/not run
• What exactly a fitness band would help me? Just measuring heart beat? Sleep?

Then I came across this course in coursera which talks exactly about these questions and I was very delighted to hear that, most important factor is maintaining higher heart rate for short period of time which would keep you healthy. Even if you spend 10 minutes a day. Wow! That is great.
.

While there are studies already available from those professors, we can try something available in the open from Kaggle.

Out of these parameters, what we can control are only the following

1. Fasting blood sugar
2. Cholesterol
3. Maximum heart rate achieved

In exploratory model analysis, we can take the complete dataset, train the model and check how each of these features determine the heart risk. Following plot is created after training the model and interpretation of features using SHAP values. Factors on top are more important. Values on right favors heart disease. Values towards blue are lower for a given feature.

Since fasting blood surar is a boolean value, it is always better to be on lower side. No further check needed.

A partial dependance plot (PDPPlot) shows the effect of cholesterol If we look at the shap plot and pdp, we see that the unit effect to heart disease is 0.05-0.15.

The most influencing factor in shap plot is the maximum heart rate achieved.

We could lower the risk from 0.2 to 0.4 just by making sure that our heart rate can reach above 120. This shows that investing in a good fitness watch and doing a HIIT (high intensity interval training), where we aim to achieve the maximum heart rate in lowest possible time.

Is there an effect or shift of heart rate based on Gender?

PS: This is only for an academic interest on interpretting what data shows. I am neither qualified medical practitioner nor an expert on exercises. A separate post with details on the model building (including the code) is available here

Acknowledgements

The dataset is taken from three other research datasets used in different research papers. The Nature article listing heart disease database and names of popular datasets used in various heart disease research is shared below. https://www.nature.com/articles/s41597-019-0206-3

The data set is consolidated and made available in kaggle

Thanks to this wonderful post in Kaggle whch I have used in data cleanup