Long term cardiovascular risk models' combination - a new approach

Area of Science:

• Cardiovascular disease risk assessment
• Biostatistics
• Machine learning in healthcare

Background:

• Current cardiovascular risk scores have limitations, including a restricted number of considered risk factors.
• Existing tools often struggle to effectively manage incomplete patient data, impacting accuracy.

Purpose of the Study:

• To develop an advanced cardiovascular risk scoring strategy addressing limitations of current systems.
• To enhance risk prediction by integrating more factors and accommodating incomplete information.

Main Methods:

• A two-phase strategy was employed, starting with a Naïve-Bayes classifier for common data representation.
• Conditional probabilities were estimated using frequency methods and optimized with a Genetic Algorithm.
• A combination scheme leveraging Bayesian probabilistic reasoning was developed to integrate multiple risk models.

Main Results:

• The proposed strategy was applied to combine SCORE, ASSIGN, and Framingham cardiovascular risk models.
• Validation demonstrated promising results, indicating the strategy's potential to overcome existing drawbacks.
• The approach shows effectiveness in handling incomplete data and a broader range of risk factors.

Conclusions:

• The developed strategy offers a promising advancement in cardiovascular risk assessment.
• This method can improve the accuracy and completeness of cardiovascular risk prediction.
• The Bayesian-based combination scheme effectively integrates multiple risk models for enhanced clinical utility.