Poisson Mixture Regression Models for Heart Disease Prediction

Area of Science:

• Biostatistics
• Cardiovascular Disease Epidemiology
• Machine Learning in Healthcare

Background:

• Effective early control of heart disease relies on efficient prediction and diagnosis.
• Model-based clustering offers a promising approach for improving diagnostic accuracy.

Purpose of the Study:

• To evaluate the efficacy of Poisson mixture regression models for heart disease prediction and diagnosis.
• To compare standard and concomitant variable mixture regression models against ordinary general linear Poisson regression.

Main Methods:

• Application of standard and concomitant variable Poisson mixture regression models.
• Utilizing Bayesian Information Criteria (BIC) for model comparison.
• Employing a Zero Inflated Poisson Mixture Regression model for risk stratification and component-wise prediction.

Main Results:

• A two-component concomitant variable Poisson mixture regression model demonstrated superior predictive performance over standard models, indicated by a lower BIC value.
• The Zero Inflated Poisson Mixture Regression model outperformed all other models, effectively clustering individuals by risk and predicting disease rates.

Conclusions:

• Poisson mixture regression models, especially the Zero Inflated variant, are effective tools for heart disease prediction.
• Component-wise identification of major risks using these models enhances prediction accuracy and facilitates early intervention.